The copy number variation landscape of congenital anomalies of the kidney and urinary tract.

Congenital anomalies of the kidney and urinary tract (CAKUT) are a major cause of pediatric kidney failure. We performed a genome-wide analysis of copy number variants (CNVs) in 2,824 cases and 21,498 controls. Affected individuals carried a significant burden of rare exonic (that is, affecting coding regions) CNVs and were enriched for known genomic disorders (GD). Kidney anomaly (KA) cases were most enriched for exonic CNVs, encompassing GD-CNVs and novel deletions; obstructive uropathy (OU) had a lower CNV burden and an intermediate prevalence of GD-CNVs; and vesicoureteral reflux (VUR) had the fewest GD-CNVs but was enriched for novel exonic CNVs, particularly duplications. Six loci (1q21, 4p16.1-p16.3, 16p11.2, 16p13.11, 17q12 and 22q11.2) accounted for 65% of patients with GD-CNVs. Deletions at 17q12, 4p16.1-p16.3 and 22q11.2 were specific for KA; the 16p11.2 locus showed extensive pleiotropy. Using a multidisciplinary approach, we identified TBX6 as a driver for the CAKUT subphenotypes in the 16p11.2 microdeletion syndrome.

INTRODUCTION

Congenital anomalies of the kidney and urinary tract (CAKUT) have devastating impact on childhood renal survival[1-4]. A better understanding of the pathogenesis of CAKUT is imperative to improve the prognosis of affected children[5]. CAKUT encompasses a broad spectrum of phenotypes, which can result from early disruptions in transcription factors and signaling molecules such as PAX2, EYA1, RET, BMP4[5,6] and others, or are directly related to spatiotemporal interactions of the outgrowing ureteric bud and metanephric mesenchyme[7-9]. Early disruption of these interactions leads to renal agenesis, hypoplasia or -dysplasia, whereas later perturbations in the outgrowth of the ureteric bud result in obstructive uropathy (OU), vesicoureteric reflux (VUR) or ectopic or horseshoe kidney (EK-HK)[6,10-13]. Maldevelopment of the lower urinary tract can result in epispadias or hypospadias (LUTM) or posterior urethral valves (PUV)[14]. Genetic manipulation in mice indicates that disruption of the same cellular pathways can lead to multiple different genitourinary phenotypes[15-19]. Similarly, mutations in genes associated with Mendelian forms of CAKUT can lead to different subphenotypes in individuals from the same families[18,20,21], suggesting that a single genetic lesion can have pleiotropic manifestations across the spectrum of CAKUT. Conversely, differences in prevalence and severity of structural malformations point towards distinct molecular basis and genetic architecture[22,23]. To date, there are over 50 single-gene disorders underlying isolated and non-isolated (i.e. syndromic) CAKUT[5,24,25]. Furthermore, a significant number of CAKUT patients carry copy number variants (CNV) that have been previously associated with a syndrome diagnosis or are large and extremely rare in the general population[26-28]. Nevertheless, a molecular diagnosis can be established in less than 20% of affected individuals[15-17,29-32], emphasizing that large studies across the entire phenotypic spectrum of CAKUT are indispensable to identify those genes and allelic variants that are either specific to subcategories of disease or those that have pleiotropic effects across the entire genitourinary tract, and to discover novel cellular pathways that are implicated in kidney and urinary development.

Here, we show both the presence of a distinct genetic architecture as well as pleiotropic mutations for the different subphenotypes of CAKUT. Our CNV analysis of nearly 3,000 cases across the phenotypic spectrum of CAKUT sheds light on the genomic architecture of disease, and implicates TBX6 as a main driver for the various CAKUT phenotypes in the 16p11.2 microdeletion syndrome.

RESULTS

Burden of rare CNVs is high in CAKUT

We conducted a study in 2,824 CAKUT cases and 21,498 population controls (Supplementary Tables 1 and 2 and Supplementary Figure 1) to compare the prevalence of rare CNVs that intersect genes. The case cohort represented common CAKUT subcategories: kidney anomalies (KA; including renal agenesis, hypoplasia, dysplasia, and multicystic dysplasia), vesicoureteral reflux (VUR), obstructive uropathy (OU; including congenital hydronephrosis, ureteropelvic junction obstruction, ureterovesical junction obstruction, and congenital megaureter), duplicated collecting system (DCS; including duplications of the ureter or kidney, partial and complete), posterior urethral valves (PUV), ectopic kidney or horseshoe kidney (EK-HK), and other lower urinary tract malformations (LUTM; including anomalies of the bladder and anterior urethra). Our analysis focused on large (≥100 kb) CNVs that are present in less than 1:1,000 population controls, across different ancestries as estimated by principal component analysis (Supplementary Figure 2).

This analysis revealed marked enrichment for large, rare CNVs in CAKUT compared to controls (P = 1.04 × 10−24; Figure 1A), consistent across virtually all metrics examined, including the the number of individuals with large, rare CNVs, the median size and total CNV span per genome, and the fraction of GD-CNVs, indicating an important role for gene-disrupting CNVs across the whole CAKUT spectrum (Table 1 and Supplementary Table 3). This signal was driven by cases with and without extrarenal manifestations, since the burden difference was still highly significant when analyzing cases without extrarenal defects separately (P = 3.21 × 10−8; Figure 1A–B and Supplementary Figures 3 and 4). Even when considering only simplex isolated CAKUT cases (i.e. cases without extrarenal manifestation and with no additional CAKUT phenotypes other than the primary one), there was still an excess burden of large, rare CNVs compared to controls (P = 1.15 × 10−8; Supplementary Figure 5). Comparison of burden metrics for cases and controls indicated a population attributable risk of 4.1% for large, rare CNVs ≥500 kb in CAKUT (OR 1.64, 95% CI 1.44–1.87; P = 3.23 × 10−13). This excess burden is predominantly attributable to exonic deletions, most prominently in the KA cases (Supplementary Tables 4 and 5). Interestingly, we observed an enrichment for rare duplications compared to deletions in VUR and PUV cases when compared to controls (P = 3.33 × 10−2 and P = 8.67 × 10−4, respectively; Supplementary Figure 6). Secondary analysis also showed enrichment of the number of genes per individual genome that were affected by rare CNVs for nearly all CAKUT subcategories (Supplementary Table 6 and Supplementary Figure 7).

FIGURE 1.

Burden of rare copy number variations in CAKUT cases compared to controls. A, B. Burden of large, rare, exonic CNVs in all CAKUT cases and controls (A) and in KA and VUR cases and controls (B). C, D. Prevalence of known genomic disorders (C) and novel likely pathogenic copy number variants (D) in CAKUT cases and controls. Deletions are marked in red, duplications are marked in green. KA, OU, PUV and DCS were significantly enriched for genomic disorders. The genomic architecture of KA cases was predominantly devised by deletions, while the genetic basis of PUV and DCS cases mostly constituted of duplications. CAKUT = congenital anomalies of the kidney and urinary tract, DCS = duplex collecting system, EK-HK = ectopic kidney/horseshoe kidney, ERM = extrarenal malformations, KA = kidney anomaly, Mb = megabases, OU = obstructive uropathy, PUV = posterior urethral valves, VUR = vesicoureteral reflux.

TABLE 1.

Distribution of largest, rare, exonic CNV per individual across different size thresholds

	All CAKUT	KA	OU	PUV	DCS	VUR	EK-HK	LUTM	Controls
Total n	2,817	1,083	512	141	233	659	127	62	21,490
≥100 kb: n (%)	1,044 (37.1)	426 (39.4)	199 (38.9)	46 (32.6)	66 (28.3)	245 (37.2)	46 (36.2)	16 (25.8)	6,767 (31.5)
OR (CI)	1.28 (1.18–1.39)	1.41 (1.24–1.60)	1.38 (1.15–1.66)	1.05 (0.72–1.52)	0.86 (0.64–1.15)	1.29 (1.09–1.52)	1.24 (0.84–1.80)	0.76 (0.40–1.36)	-
P Fisher’s exact	4.01 × 10−9	1.02 × 10−17	5.25 × 10−4	0.79	0.32	2.20 × 10−5	0.25	0.41	-
≥250 kb: n (%)	615 (21.8)	260 (24.0)	107 (20.9)	26 (18.4)	38 (16.3)	147 (22.3)	25 (19.7)	12 (19.4)	3,510 (16.3)
OR (CI)	1.43 (1.30–1.58)	1.62 (1.40–1.87)	1.35 (1.15–1.66)	1.16 (0.72–1.79)	1.00 (0.68–1.42)	1.47 (1.21–1.78)	1.26 (0.78–1.96)	1.23 (0.60–2.34)	-
P Fisher’s exact	1.35 × 10−12	3.13 × 10−10	7.82 × 10−3	0.49	1.00	9.61 × 10−5	0.33	0.49	-
≥500 kb: n (%)	316 (11.2)	154 (14.2)	45 (8.8)	13 (9.2)	16 (6.9)	71 (10.8)	12 (9.4)	5 (8.1)	1,535 (7.1)
OR (CI)	1.64 (1.44–1.87)	2.20 (1.79–2.58)	1.25 (1.08–1.68)	1.32 (0.68–2.35)	0.96 (0.54–1.60)	1.57 (1.20–2.02)	1.36 (0.68–2.47)	1.14 (0.36–2.83)	-
P Fisher’s exact	3.23 × 10−13	5.67 × 10−15	0.16	0.32	1.00	7.62 × 10−5	0.29	0.80	-
≥1,000 kb: n (%)	141 (5.0)	79 (7.3)	16 (3.1)	7 (5.0)	8 (3.4)	26 (3.9)	2 (1.6)	3 (4.8)	502 (2.3)
OR (CI)	2.20 (1.81–2.67)	3.30 (2.54–4.22)	1.35 (0.76–2.23)	2.18 (0.86–4.66)	1.49 (0.63–3.00)	1.72 (1.10–2.57)	0.67 (0.08–2.48)	2.13 (0.42–6.56)	-
P Fisher’s exact	4.41 × 10−14	4.82 × 10−17	0.24	4.94 × 10−2	0.27	1.29 × 10−2	1.00	0.18	-

Proportion of individuals with their largest, rare CNV at least as large as the indicated size threshold. CAKUT = congenital anomalies of the kidney and urinary tract, CI = 95% confidence interval, CNV = copy number variation, DCS = duplication of the collecting system and/or ureter, EK-HK = ectopic kidney-horseshoe kidney, KA = kidney anomaly, kb = kilobases, LUTM = other lower urinary tract malformation, OR = odds ratio, OU = obstructive uropathy, PUV = posterior urethral valves, VUR = vesicoureteral reflux.

Genomic disorders inform about the genetic architecture of CAKUT

We cross-annotated all rare CNVs with a curated list of known genomic disorders (Supplementary Table 7)[33,34] and identified 45 distinct known genomic disorders (GD) in 112 (4.0%) independent cases (Supplementary Table 8, Figure 1C). Five cases carried more than one known GD-CNV, resulting in a total number of 117 known GD-CNVs in our cohort (Supplementary Table 9); in comparison, known GD-CNVs were found in only 134/21,498 (0.6%) population controls (OR 6.58, 95% CI 5.05–8.55; P = 7.53 × 10−41; Supplementary Table 10).

Further annotation identified 54 large, rare, exonic CNVs in an additional 47 CAKUT cases (1.7%) that fulfilled ACMG classification[35,36] as likely pathogenic imbalances (Online Methods and Supplementary Table 11). Among these, CNVs that can be classified as pathogenic were an atypical deletion at the 16p11.2 locus (see below), a 300 kb deletion involving PAX2, a 100 kb duplication containing TBX18, a 571 kb deletion spanning PBX1, and a 6.8 Mb duplication including BMP4 (Supplementary Figure 8A–D). We also identified overlapping duplications at the 15p11.2 locus in 5 cases with ureteric defects (VUR, OU) and PUV, as well as two KA cases with deletions proximal to the 16p11.2 microdeletion syndrome. In line with the burden test results, the distribution of deletions and duplications at known and novel GD-CNV loci (Figure 1C,D) was significantly different among CAKUT categories (6 × 2 Fisher’s Exact Test P = 8.93 × 10−5). In particular, subjects with KA and OU were enriched for deletion syndromes, contrary to PUV and DCS that showed an excess of duplications at the same genomic loci.

When we examined associations with disease severity, cases who carried a known GD-CNV were more likely to have multiple sites of the urinary tract affected (OR 1.60, 95% CI 1.05–2.42; P = 0.02) and more frequently harbored extrarenal malformations (OR 4.79, 95% CI 3.21–7.17; P = 6.00 × 10−15) compared to cases without a known GD-CNV. However, consistent with the burden tests conducted above, analysis of simplex isolated cases still showed increased burden compared to controls (OR 3.12, 95% CI 2.06–4.61; P = 1.86 × 10−7), strongly implicating the significance of GD-CNVs in milder forms of CAKUT. To test more complex genetic models for modes of disease determination, we examined cases and controls for second site CNVs. Among the 159 cases with GD-CNVs (112 known, 47 new/likely pathogenic: altogether called “diagnostic CNVs” (DCNVs)) eleven (6.9%) carried more than one DCNV (Supplementary Table 12). In cases, the presence of zero, one, or more than one DCNV increased the likelihood of extrarenal malformations (Chi-square test for 3 × 2 table: P = 6.94 × 10−7; Supplementary Figure 9).

KA, OU, and VUR show distinct genomic characteristics

Comparison of CNV landscape between the largest CAKUT subcategories revealed both commonalities and differences between KA, OU, and VUR (Table 2). We found significant CNV enrichment for all three phenotypes, which was most uniformly shown by a larger proportion of cases carrying exonic CNVs ≥100 kb compared to controls. KA cases had the highest CNV burden (P = 9.01 × 10−25; Figure 1B), as evidenced by median CNV size and total span as well as proportion of individuals with large imbalances that could be classified as pathogenic GD-CNVs (80 cases with GD-CNVs, OR 12.65, 95% CI 9.40–16.94; P = 8.53 × 10−50; Table 2 and Figure 1C). After removal of individuals with a known GD-CNV, the strength of the excess CNV burden in KA was still detectable but markedly attenuated (P = 1.27 × 10−3), consistent with the major role for GD-CNVs in the pathogenesis of KA (Supplementary Figure 10). In contrast, VUR cases were also affected by a high CNV burden (P = 7.56 × 10−7), but these CNVs predominantly involved duplications and were less likely to be classified as GD-CNVs (7 cases with GD-CNVs, OR 1.71; 95% CI 0.67–3.64; P = 0.20; Figure 1C). The large rare CNVs in VUR were mostly classified as likely pathogenic, potentially indicating that novel syndromes account for the molecular basis of this disorder (Table 2, Figure 1D and Supplementary Figures 3 and 10). Finally, OU cases fell in an intermediate category, with CNVs that were not statistically larger compared to controls, yet the excess CNV burden (P = 0.01; Supplementary Figure 3) was reflected by a greater proportion of individuals with large imbalances that were also more likely to be classified as pathogenic GD-CNVs (11 cases with GD-CNVs, OR 3.50, 95% CI 1.70–6.52; P = 5.96 × 10−4; Table 3 and Supplementary Figure 3). These data suggest a distinct genomic architecture among CAKUT subcategories, with enrichment of GD-CNVs in KA, and enrichment for novel, large or intermediate size imbalances in VUR and OU, respectively.

TABLE 2.

Comparison of CNV landscape across major CAKUT subcategories

	CAKUT(n=2,824)	KA(n=1,088)	OU(n=512)	VUR (n=660)	Controls(n=21,498)
Median CNV size (kb) (IQR)[‡]	245 (315)	258 (394)	223 (244)	248 (273)	223 (248)
P value	3.4 × 10−6	3.6 × 10−6	0.87	1.9 × 10−2	-
Median total CNV span (kb) per genome (IQR)[‡]	350 (598)	414 (815)	306 (451)	353 (503)	300 (437)
P value	1.4 × 10−7	1.9 × 10−8	0.61	1.7 × 10−2	-
% individuals with at least 1 large, rare CNVs[‡]	37.1	39.3	38.9	37.2	31.5
P value	4.0 × 10−9	1.0 × 10−7	5.3 × 10−4	2.2 × 10−3	-
% individuals with at least 2 large, rare CNVs[‡]	10.5	12.6	10.4	9.7	8.1
P value	1.6 × 10−5	1.0 × 10−6	7.1 × 10−2	0.12	-
% individuals with known GDs	4.1	7.4	2.1	1.1	0.6
P value	7.5 × 10−41	8.5 × 10−50	6.0 × 10−4	0.20	-
Ratio of No. of large, rare deletions: duplications[‡]	0.90	1.02	1.05	0.74	0.94
P value	0.34	0.35	0.37	3.3 × 10−2	-

CNVs refer to exonic CNVs ≥100 kb with frequency of < 1:1,000 in controls.

Metrics derived from burden analysis.

CAKUT = congenital anomalies of the kidney and urinary tract, CNV = copy number variation, GD = genomic disorder, IQR = interquartile range, KA = kidney anomaly, OU = obstructive uropathy, VUR = vesicoureteral reflux.

TABLE 3.

Clinical characteristics of CAKUT cases affected by the heterozygous 16p11.2 microdeletion.

Case	Sex	Ethnicity	Chr	Start(Mb)	End(Mb)	Size(Mb)	Number ofgenes	Age ofDiagnosis(years)	Phenotype	AdditionalCAKUT	Extrarenalmalformations	Familyhistory	eGFR(ml/min/1.73m2)
CAKUT1	M	White	16	28.4	30.1	1.7	69	9	KA	N	N	NA	NA
CAKUT2	M	Black	16	29.5	30.1	0.6	31	NA	OU	N	N	NA	NA
CAKUT3	F	White	16	29.5	30.1	0.6	31	NA	KA	N	N	NA	NA
CAKUT4	M	White	16	29.5	30.1	0.6	31	0	PUV	Y (VUR, KA bilateral)	N	Y	158
CAKUT5	F	White	16	29.5	30.1	0.6	30	NA	VUR	N	Y (scoliosis)	NA	NA
CAKUT6	M	White	16	29.5	30.1	0.6	40	NA	KA	N	Y (craniofacial dysmorphia, radial agenesis, bilateral thumb aplasia)	N	44
CAKUT7	M	Admixed	16	29.6	30.1	0.5	28	NA	DCS	Y	N	NA	NA
CAKUT8	M	White	16	29.6	30.1	0.5	34	NA	KA	N	Y (cryptorchidism)	NA	NA
CAKUT9	M	White	16	29.9	32.6	2.7	94	N	KA	N	Y (craniofacial dysmorphias, abnormality of the feet, cardiac defect, corpus callosum hypoplasia)	NA	5

CAKUT = congenital anomalies of the kidney and urinary tract, Chr = chromosome, DCS = duplicated collecting system and/or ureter, eGFR = estimated glomerular filtration rate, F = female, KA = kidney anomaly, M = male. Mb = megabases, N = no, NA = not available, OU = obstructive uropathy, PUV = posterior urethral valves, VUR = vesicoureteral reflux; Y = yes.

Six GD loci account for 65% of CAKUT cases with known GD-CNVs

We conducted a literature search, including a survey of databases of genomic variants such as DECIPHER[37,38] and ISCA[35], and found either a known link to CAKUT[5,23,26,39-46] or case series or reports in which CAKUT was part of the clinical phenotype[47-53] in 43 out of 45 GD-CNVs. This finding underlines that these GD-CNVs are causally related to CAKUT.

While the majority of GD-CNVs affected unique or few cases (Supplementary Table 8), six loci explained 73 of the 112 (65%) cases who carried a known GD-CNV (Figure 2). These common GD loci included chromosome 1q21.1 (7 cases = 6.2% of cases with a GD-CNV; five deletions, two duplications), chromosome 4p16.1-p16.3 (Wolf-Hirschhorn syndrome; 5 cases = 4.4% of cases with a GD-CNV; all deletions), chromosome 16p11.2 (9 cases = 8.0% of cases with a GD-CNV; eight deletions, one duplication), chromosome 16p13.11 (9 cases = 8.0% of cases with a GD-CNV; four deletions, five duplications), chromosome 17q12 (26 cases = 23.0% of cases with a GD-CNV; 23 deletions, three duplications), and chromosome 22q11.2 (17 cases = 15.0% of cases with a GD-CNV; 14 deletions, three duplications) loci. Genotype-phenotype correlations identified that microdeletions, but not duplications, were enriched in upper urinary tract defects, especially KA (1q21.1, 4p16.1-p16.3, 17q12, 22q11.2), whereas the 16p11.2 microdeletion syndrome was identified in all CAKUT subcategories, suggesting a high pleiotropic effect. These observations provide further support that some genetic lesions result in specific CAKUT subphenotypes while other genetic lesions, such as the 16p11.2 microdeletion syndrome, have high pleiotropic effects across the genitourinary tract. These loci provide a list of regions that are likely to encompass critical regulators of kidney and urinary tract development in humans, offering a unique opportunity for gene discovery.

FIGURE 2.

Common genomic disorders loci in CAKUT cases and their prevalence in controls. Deletions are marked in red, duplications are marked in green. Among these common genomic loci, the chromosome 16p11.2 locus showed high pleiotropy, whereas the Wolf-Hirschhorn, 17q12 and 22q11.2 loci were mostly identified in KA cases. CAKUT = congenital anomalies of the kidney and urinary tract, Ctrls = controls, DCS = duplex collecting system, EK-HK = ectopic kidney/horseshoe kidney, KA = kidney anomaly, LUTM = other lower urinary tract malformation, OU = obstructive uropathy, PUV = posterior urethral valves, RCAD = renal cysts and diabetes, VUR = vesicoureteral reflux.

Exome sequencing indicates haploinsufficiency underlying CNV deletions

To test the mechanism by which pathogenic CNVs confer risk to CAKUT, we conducted whole exome sequencing (WES) in 23 patients harboring pathogenic microdeletions at 14 independent loci (Supplementary Table 13). Based on recessive loss-of-function (LOF) inheritance, WES would uncover a hemizygous LOF mutation on the non-deleted allele (unmasking effect). WES was performed as previously described[18,23,54]. We retrieved all identified candidate hemizygous LOF variants located within the case-specific deletion loci. Overall we identified only one LOF variant in EFCAB12 (p.Q437*; observed once in heterozygosity in the ExAC database[55,56]) in a patient affected by unilateral KA and multiple extrarenal manifestations, including neurodevelopmental delay, epilepsy, corpus callosum agenesis, left radial bone agenesis and a patent ductus arteriosus, who harbored a 14.9 Mb deletion at chromosome 3q13.22–13.1 (Supplementary Figure 11)[57]. The EFCAB12 variant was inherited from a heterozygous unaffected mother, while the CNV occurred de novo (Supplementary Figure 11)[57]. Query of 15,469 control individual exomes at the Institute of Genomic Medicine at Columbia University did not reveal any homozygous or compound heterozygous truncating variants, suggesting that biallelic truncating mutations are likely not tolerated in humans and propose recessive mutations in EFCAB12 as contributors to the developmental syndrome of this individual. In addition, we performed clinical annotation of genes that are known to be implicated in Mendelian forms of CAKUT, by querying WES data for an in-house gene list as described[58]. We did not find any pathogenic variants in known CAKUT genes for any of the 23 deletion carrier patients. Hence, our WES studies suggest haploinsufficiency as main pathogenetic mechanism for the CAKUT-associated deletion CNVs.

CAKUT is a common phenotype in the 16p11.2 microdeletion syndrome

We identified nine CAKUT cases and five controls with overlapping 16p11.2 microdeletions (0.32% vs 0.02%; OR 13.7, 95% CI 4.1–52.2; P = 4.39 × 10−6; Figure 2, Table 3 and Supplementary Tables 9 and 10), implicating CAKUT as an important feature of this syndrome. To better estimate the prevalence of genitourinary malformations in individuals who harbor this GD-CNV, we first analyzed the clinical reports from 186 cases with 16p11.2 microdeletion syndrome in the DECIPHER database (Supplementary Table 14). The most prevalent associated conditions were abnormalities of the nervous system (92.5%), abnormalities of head or neck (26.9%), growth defects (23.7%), and abnormalities of the limbs (14.5%). Abnormalities of the genitourinary system (including KA, hydronephrosis, and VUR) were reported in only 10 cases (5.4%), but these estimates likely reflect the standard clinical indication for ordering a DNA microarray for diagnosis of a syndromic disease, i.e. neurodevelopmental delay and dysmorphic features. In fact, abnormalities of the skeletal system, which are a hallmark for the 16p11.2 microdeletion syndrome[59,60], were reported in only 19 cases (10.2%), since, similarly to individuals with CAKUT, these patients are rarely referred for genetic testing. To obtain a more unbiased estimate of prevalence of genitourinary defects in patients with 16p11.2 microdeletion, we queried the data warehouse of the Center for Applied Genomics (CAG) at the Children Hospital of Philadelphia (CHOP). We analyzed medical records for 42 children with the 16p11.2 microdeletion syndrome (Supplementary Table 15). Neurodevelopmental defects and brain anomalies, and spine defects were the most prevalent conditions, recorded in 22 (52.4%) and 6 (14.2%) cases, respectively. Imaging studies and clinical data assessing the kidney and urinary tract were available for 15 out of 42 and genitourinary anomalies were present in 6 out of these 15 cases (40.0%), highlighting that CAKUT, similar to scoliosis, is a common feature in this syndrome characterized by incomplete penetrance. Consistent with our observation of a high pleiotropic effect for the 16p11.2 microdeletion, the phenotype was also variable in this cohort: two patients with OU, two with VUR, one with PUV, and one with LUTM. These data implicate CAKUT as a significant feature of the chromosome 16p11.2 microdeletion syndrome and provide an opportunity to identify novel key regulators of kidney and urinary tract development.

TBX6 is a driver for CAKUT in the 16p11.2 microdeletion syndrome

We recently demonstrated that the combinatorial use of genomic analyses with functional modeling in vertebrates can lead to the identification of key drivers of CAKUT phenotypes from microdeletion syndromes[23]. Analysis of breakpoints in 9 CAKUT cases with typical or atypical 16p11.2 microdeletions identified a ~175 Kb minimal region of overlap (MRO) predicted to harbor the genetic driver(s) for CAKUT (Supplementary Figure 12). Among the 19 genes included in the MRO, TBX6 appeared as a strong candidate: it is included in the list of essential genes from the International Mouse Phenotyping Consortium[61]; it is depleted from LOF mutations in subjects from the ExAC database (aggregate prevalence of LOF <0.001); it has a role in paraxial and intermediate mesoderm development[62,63]; and its inactivation in humans and rodents leads to spine defects including congenital scoliosis and spondylocostal dysostosis[59,62,64-66]. The connection of TBX6 mutations with vertebral anomalies is of particular interest because of the historically well-known clinical association between congenital scoliosis and CAKUT[67-69].

To provide a functional link between inactivation of TBX6 and CAKUT, we generated an allelic series by cross breeding mice with two different alleles: a Tbx6 null allele (Tbx6)[70] and a Tbx6 hypomorphic allele, the spontaneous mutant Tbx6 (hereafter referred to as Tbx6 and Tbx6, respectively) previously studied for vertebral development and spine defects[71]. Since the homozygous null mutation is embryonically lethal at E9.5, precluding analysis of the developing urinary tract, we first studied compound heterozygous embryos (Tbx6), which retain sufficient residual expression of Tbx6 for survival past E9.5. We studied 19 Tbx6 embryos at E17.5-E18.5, and observed full penetrance of CAKUT with variable expressivity of phenotypes characterized by unilateral or bilateral renal agenesis, unilateral or bilateral renal hypoplasia and dysplasia, and obstructive uropathy characterized by hydronephrosis and hydroureter (Table 4, Figure 3A–L). Microscopic analysis of renal tissue from Tbx6 embryos at stages from E13.5 to E18.5 showed variable severity of kidney anomalies (Figure 3D–L). Severe phenotypes included unilateral or bilateral renal agenesis, rudimentary kidneys and undeveloped renal parenchyma embedded in the paraspinal musculature (Figure 3F,I,L). Milder phenotypes included unilateral and bilateral renal hypoplasia with hydroureter, tubule dilation and hydronephrosis (Figure 3E,H,K). These data strongly implicate TBX6 as a major driver for CAKUT phenotypes. Comparative analysis was performed on wildtype and Tbx6 embryos at E11.5 after staining with E-cadherin (Cdh1), an epithelial marker, and Pax2, which labels mesenchymal progenitors that produce nephrons, the ureteric bud which gives rise to the renal collecting duct system, and the common nephric duct (CND)[72-74]. This analysis revealed that the events leading to CAKUT (renal parenchyma abnormalities as well as obstruction) are present at very early stages during development (Figure 3M,N). In wildtype E11.5 embryos, the ureteric bud has invaded the metanephric blastema and has undergone a round of branching (Figure 3N). In the Tbx6 mutant (Figure 3N), the ureteric bud has not fully invaded the metanephric blastema and had not branched, a defect that is predicted to lead to KA. Additional Cdh1-positive cells in Figure 3N might represent persistent mesonephros or ectopic ureteric buds. In this latter possibility, ectopic buds would explain the duplication of kidney and ureter phenotype as observed in Figure 4.

TABLE 4.

Tbx6 dosage-dependent kidney and urinary tract phenotypes

Phenotype	Tbx6rv/-	Tbx6rv/rv			Tbx6+/−	Tbx6rv/+	Tbx6+/+
	E17.5–E18.5*	E15.5	E18.5	P0-P1	E18.5*	E18.5*	E18.5
Bilateral renal agenesis	4/19	0/8	0/5	0/3	0/4	0/6	0/12
Unilateral renal agenesis	6/19	0/8	0/5	0/3	0/4	0/6	0/12
Bilateral renal hypoplasia/dysplasia	8/19	3/8	0/5	0/3	0/4	0/6	0/12
Unilateral renal hypoplasia/dysplasia	3/19	3/8	4/5	1/3	0/4	0/6	0/12
Hydronephrosis/ hydroureter	5/19	1/8	1/5	3/3	0/4	0/6	0/12
Duplex kidney/ureter	1/19	0/8	4/5	0/3	0/4	0/6	0/12
Total embryos with defects (%)	19/19 (100)	6/8 (75)	4/5 (80)	3/3 (100)	0/4 (0)	0/6 (0)	0/12 (0)

Based on gross morphology only. We analyzed four Tbx6 and six Tbx6+/rv E18.5 embryos by gross morphology thus hampering ability to assess milder phenotypes and incomplete penetrance in these embryos.

FIGURE 3.

Analysis of urinary tract phenotypes in Tbx6 mutants. A-C. Whole mounts of urogenital tracts isolated from E18.5 wild type (A) and Tbx6 mutants showing severe bilateral renal hypoplasia (B) and unilateral renal agenesis with contralateral renal hypoplasia (C); K = kidney, U = ureter, B = bladder. D-F. H&E stained sagittal sections from E18.5 wild type (D) and Tbx6 mutants (E,F). In the wild type it is appreciable the normal developing nephrogenic zone (nz) and kidney medulla (m). The arrows in E point to the dilated renal pelvis (upper arrow) and ureter (lower arrow), which are indicative of hydronephrosis and hydroureter, respectively. In E the kidney parenchyma also appears severely hypoplastic. The arrowhead in F points to the rudimentary kidney, which is embedded in paraspinal musculature. Few dilated tubule and microcysts are present. G-I. H&E stained kidney from and E15.5 wild type embryo (G) and E15.5 Tbx6 mutant embryos (H, I). The mutants show moderate to severe hypoplasia with reduction of nephrogenic zone (nz) (H); the arrowhead indicated severely underdeveloped kidney tissue with tubule dilation and microcysts (I). J-L. H&E histological analysis of kidneys from E13.5 wild type embryos (J) and Tbx6 mutants (K,L). The arrowheads point to the rudimentary kidneys that are embedded in the body wall. M,N. Immunostaining of E11.5 wild type and Tbx6 mutant embryos stained with Pax2 (red) and Cdh1 (green) showing the ureteric bud (ub) nephron progenitors (np), and nephric duct (nd). Note in the wild type embryo, the ureteric bud has invaded the metanephric mesenchyme and branched, while in the mutant, the ureteric bud has not fully invaded the metanephric mesenchyme.

FIGURE 4.

Analysis of urinary tract phenotypes in Tbx6 mutants. A-C. H&E stained sagittal sections from a wild type P0 pup (A) and P0 Tbx6 pups (B,C). In the wild type can be appreciated the renal cortex (rc) resulting from the development of the nephrogenic zone; the medulla (m) and the renal pelvis (p). The arrows in B point to the duplicated kidneys. The arrows in C point to hypoplastic kidney and the dilated renal pelvis and proximal ureter. D-F. H&E stained sections from E15.5 wild type embryo (D) and Tbx6rv/rv mutant embryos (E,F). The arrow in E points to the dilated renal pelvis and proximal ureter. The arrows in F points to the hypoplastic kidney and dilated ureter.

Since more subtle CAKUT phenotypes like OU or DCS may require a minimum glomerular filtration rate or might be masked in more severe models, we examined a milder model that might be closer to the genetic architecture of humans harboring the 16p11.2 microdeletion syndrome. We therefore analyzed embryos homozygous for the Tbx6 hypomorphic allele (Tbx6). We generated 25 homozygous Tbx6 animals at E15.5 (n=10), E18.5 (n=10), and P0 (n=5) and observed incomplete penetrance and variable expressivity of multiple phenotypes that are typical of human CAKUT, including mild unilateral or bilateral hypoplasia with asymmetric kidneys, bifid ureter or DCS, and OU with profound hydronephrosis (Table 4 and Figure 4). This pleiotropy of CAKUT phenotypes, ranging from renal parenchyma defects (KA) to hydronephrosis (OU), duplication of ureters (DCS) is highly reminiscent of the pleiotropic manifestations of the 16p11.2 microdeletion in humans, suggesting that TBX6 gene dosage is a major determinant for the pathogenesis of CAKUT and the observed variable expressivity of phenotypes in this syndrome.

DISCUSSION

CAKUT has profound impact on child health and alone accounts for about 50% of kidney failure requiring dialysis and transplantation in children[1,3,75-78]. Moreover, CAKUT is often accompanied by extrarenal comorbidities, such as neurodevelopmental and cardiovascular diseases, contributing further to the disease burden in affected children[5,26,27,79]. Understanding the genetic architecture of CAKUT has important implications for the development of therapeutic tools that aim to slow down progression of kidney disease and to mitigate the associated neurocognitive and cardiac disease. In that respect, besides the value in achieving molecular diagnosis and hence solving the etiology of disease. In this respect, it, is relevant to address two features of CAKUT: incomplete penetrance and variable expressivity of disease. While prediction of a targeted treatment that prevents or reverses the disease is still limited by current knowledge, understanding the intrinsic inter-individual mechanisms of compensation or amplification of disease are of critical for devising treatments that might slow down or halt progression of kidney disease and associated extrarenal malformations.

The study of CNVs has provided enormous insight into the genetic architecture of many developmental traits[80-8889]. We conducted a large study on rare CNVs in nearly 3,000 CAKUT cases and over 21,000 controls. By studying the burden of rare CNVs, we identified that KA, OU, and VUR are particularly enriched for large and rare structural variants that affect coding regions of the genome. When examining known GD-CNVs, we found particular enrichment in CAKUT cases with KA, OU, DCS, and PUV when compared to controls. Overall, we identified 45 distinct GDs at 37 independent genomic loci in 4.1% of the CAKUT cases, indicating substantial genetic heterogeneity. We identified novel GD-CNVs in an additional ~2% of CAKUT cases, providing multiple novel susceptibility loci to kidney disease. With respect to mechanism, WES in cases harboring deletions was consistent with haploinsufficiency as the most frequent pathogenic mechanism in individuals affected by GD-associated deletions. Moreover, increased burden of second site CNVs was associated with KA and increased prevalence of extrarenal malformations. These data indicate a role for background genomic burden in variable penetrance and expressivity of disease. CNVs at six of the known 37 loci accounted for ~65% of the cases with a known GD, thus identifying major susceptibility CNVs for CAKUT. At these six loci, deletions were associated with KA, whereas duplications were enriched in cases with ureteric and lower tract defects, such as DCS and PUV. This preliminary observation gives rise to the hypothesis that CAKUT subcategories such as ureteric and lower urinary tract defects (DCS, PUV) may represent, at a molecular level, mirror traits of conditions affecting the upper urinary tract (KA, OU). This “mirroring phenomenon” has been recognized in autism spectrum disorder caused by structural variation at the 16p11.2 in which patients with deletions show macrocephaly and patients with duplications are microcephalic[90,91]. Larger human cohorts and experimental data will be required to investigate this hypothesis. A closer look at common GD loci assigns specificity to CAKUT subcategories, such as deletions, but not duplications, at 4p (Wolf-Hirschhorn syndrome)[92], 17q12 (renal cysts and diabetes syndrome)[40], and 22q11.2 (DiGeorge syndrome)[93], that were nearly exclusively identified in cases with upper urinary tract malformations (mainly KA). On the other end, microdeletions at 16p11.2 were observed across the whole phenotypic spectrum of CAKUT, underlining the highly pleiotropic effect of this genomic region on human kidney and urinary tract development. The finding that pleiotropic urogenital defects are an important feature of this syndrome was replicated in two independent series from DECIPHER and CHOP.

We previously reported on the potential of large-scale genetic studies coupled with functional modeling in vertebrates in the identification of genetic drivers for kidney phenotypes of microdeletion syndromes[23]. Here we focused on the 16p11.2 microdeletion syndrome, because its pleiotropic effect and incomplete penetrance provide an ideal scenario to identify genetic factors that might be amenable for devising therapeutic intervention. Deletion mapping and prioritization analyses pointed to TBX6 as the main genetic driver for CAKUT in cases with this syndrome. Tbx6 mouse allelic series showed that compound heterozygous embryos for a null (Tbx6) and a hypomorphic (Tbx6) allele displayed fully penetrant CAKUT with variable expressivity of phenotypes, from bilateral renal agenesis to hypodysplasia and obstructive uropathy. Analysis of E11.5 embryonic tissue stained for renal progenitors and epithelial markers suggested that the initiating events for the CAKUT phenotypes are likely to occur very early in development in these mice. These data are in line with recent fate mapping studies that show that Tbx6 is expressed in renal progenitor cells prior to commitment to ureteric bud or metanephric mesenchyme lineages[94]. When we analyzed embryos and mice with milder Tbx6 inactivation (Tbx6) as a closer model to the human 16p11.2 microdeletion, we observed incomplete penetrance of multiple kidney and urinary tract malformations. Interestingly, similar to our 16p11.2 microdeletion cases, mutant mice showed phenotypes across the phenotypic spectrum of CAKUT, including KA, OU and DCS. Overall, our mouse data at least partly intimate that the variable expressivity and incomplete penetrance of CAKUT in cases with the 16p11.2 microdeletion is attributable to a fine regulation of TBX6 gene expression during organogenesis.

In summary, with this study on children and young adults with kidney and urinary tract malformations we provide significant insight into the genomic landscape of human CAKUT. We identify several susceptibility genetic loci and genes, and we highlight TBX6 as a main genetic driver for CAKUT in cases with the chromosome 16p11.2 microdeletion syndrome, with significant implications for understanding, and potentially modifying, penetrance of disease.

URLs

Annovar: http://annovar.openbioinformatics.org/en/latest/

ATAV: https://redmine.igm.cumc.columbia.edu/projects/atav/wiki

DECIPHER: https://decipher.sanger.ac.uk/

Exome aggregation consortium: https://exac.broadinstitute.org

ISCA: https://www.iscaconsortium.org/

Python Software Foundation: https://www.python.org/

lifelines: https://github.com/CamDavidsonPilon/lifelines/

pandas: https://pandas.pydata.org/

R Foundation for Statistical Computing: https://www.R-project.org/

dplyr package: https://CRAN.R-project.org/package=dplyr

survival package: https://CRAN.R-project.org/package=survival

Seattle seq: http://snp.gs.washington.edu/SeattleSeqAnnotation138/

UCSC genome browser: https://genome.ucsc.edu

METHODS

Study cohorts and Methods, including statements of data availability and any associated accession codes and references, are available in the online version of the paper.

ONLINE METHODS

Study subjects

CAKUT cases consisted 2,824 affected, unrelated individuals with different subphenotypes across the entire CAKUT spectrum. Cases were recruited in the United States, Europe and Brazil (for an overview of recruitment sites, see Supplementary Figure 1). Baseline characteristics of the study cohort are shown in Supplementary Table 1. One in every four cases presented with more than one CAKUT phenotype (i.e. complex CAKUT), whereas a significant proportion of patients had extrarenal manifestations such as neurocognitive defects, congenital heart disease or dysmorphic features. A positive family history for renal disease was identified in 15% of the cases. Cases included CAKUT patients from the CKiD[27] and KIMONO[28] studies; genotyping results of 823 (29%) subjects had been partially reported in previous publications[23,26-28]. The control population consisted of 21,498 individuals recruited as part of genome-wide genotyping studies of complex traits that are not associated to nephropathy or developmental defects (Supplementary Table 2).

CNV discovery and annotation

Genomic DNA was obtained from peripheral blood samples or, in the case of CKiD participants, lymphoblastoid cell lines derived from peripheral blood samples. Genome-wide genotyping was performed in all cases and controls using HumanHap550 or higher density Illumina (San Diego, CA, USA) or Affymetrix SNP6.0 (Santa Clara, CA, USA) microarrays (Supplementary Table 2). Raw data processing and subsequent analyses were performed in the same fashion for both cases and controls to avoid bias.

Raw data was first processed with Affymetrix Power Tools and the PennCNV-Affy protocol or with Illumina GenomeStudio v2011, to obtain probe level logR-ratio and b allele frequency values. Raw intensity data were processed in GenomeStudio v2011 (Illumina). PennCNV software[95] was used to determine CNV calls. PennCNV and PLINK software[96] were used for quality control. Principal component analysis (PCA) was conducted using smartPCA[97] based on SNP genotypes derived from the same genotyping arrays. CNV calling and all analyses were performed on hg18 coordinates, CNVs were then mapped to hg19 using UCSC liftover software (see URLs). Only high quality CNVs with confidence scores ≥30 were included in the analyses based on independent experimental validation from our prior CNV study on kidney anomalies (KA)[26].

CNVs in cases were compared to those in controls and to known CNV coordinates, and annotated with RefGene (see URLs) and curated sets of genes using custom Perl code[98]. Known CNVs were based on the DECIPHER[37,38] and the International Standards for Cytogenomic Arrays (ISCA) databases[35] and the literature (Supplementary Table 7). Curated sets of genes included those known to be associated with kidney disease and/or development from the OMIM and MGI databases and the literature. Two CNVs were considered to be identical when they had the same copy number value and had a reciprocal overlap equal to or greater than 70%. All reported CNVs were visually inspected using Illumina Genome Viewer 1.9.0 or Affymetrix ChAS to exclude potential artifacts.

CNV classification

CNVs were classified as known genomic disorders (GD-CNV) or likely pathogenic CNVs (“novel GD-CNV”). A CNV was defined as a known GD-CNV when it overlapped at least 70% of a known syndromic CNV. The criteria to define a likely pathogenic CNV were adapted from prior recommendations for interpretation of microarray data[26-28,35]. A CNV was classified as likely pathogenic if it (1) intersected at least one exon, (2) was at least 100 kb in size, (3) had a frequency in controls of at most 0.02%, (4) did not overlap (<70%) with a benign or likely benign CNV in the ISCA database and fulfilled at least one of the following additional criteria: (a) ≥70% overlap with a reported pathogenic or likely pathogenic CNV in the ISCA database, (b) intersected a causative autosomal dominant gene for CAKUT in humans or mice and/or (c) is the reciprocal of a known GD-CNV (coordinates with ≥70% overlap). Known GD-CNV and likely pathogenic taken together were termed as ‘diagnostic CNVs’. CNVs that did not meet the criteria for diagnostic CNVs were defined as variants of unknown significance, if they were at least 100 kb in size and had a frequency of at most 0.1% in controls, including homo-/hemizygous deletions of loci that underlie a recessive disorder for human disease with a renal phenotype.

CNV burden analysis

We restricted burden analyses to autosomal CNV of size ≥100 kb in cases and controls, with a frequency ≤1% in the whole control dataset and in any one control cohort comprising it. CNVs were further filtered at a frequency ≤0.1% in control population subgroups based on PCA, to avoid including CNVs that might be relatively common within ancestry groups represented in controls. Prior to determining the CNV burden, seven cases (5 KA, 1 vesicoureteral reflux and 1 lower urinary tract malformations) and 8 controls were removed because they were unmatched outliers in PCA, yielding a dataset of 2,817 cases and 21,490 controls for burden analyses.

Whole exome sequencing

Whole exome sequencing was performed on 23 CAKUT patients harboring 14 distinct known or novel GD-CNVs at the New York Genome Center (NYGC). Briefly, for each capture experiment, 1 μg of genomic DNA was fragmented, linkers were ligated to the ends and a library was prepared. Next, genomic DNA was annealed to Agilent V4 capture probes, and bound genomic DNA was eluted and subjected to Next-Generation sequencing performed on an Illumina HiSeq 2500 machine. Sequence reads were converted to FASTQ format and mapped to the reference genome. We used uniform procedures for variant calling to prevent technical bias. Samples were processed using a consistent alignment and variant calling pipeline, consisting of primary alignment using bwa-0.5.10, duplicate removal using Picard tools, index realignment and variant calling using GATK 3.6, and variant annotation using snpEff-3.3, AnnoVar (see URLs), and SeattleSeq (see URLs), with Ensembl-GRCh37.73 annotations.

After variant calling, resultant calls and their underlying quality statistics were then stored in a database of variants (AnnoDB) that is used by Analysis Tool for Annotated Variants (ATAV) analyses (see URLs). Next, we queried ATAV for the individual deleted regions of all 23 CAKUT cases and annotated all hemizygous variants (minimal coverage >8X, minimal genotype quality score >30) with a minor allele frequency of <1% (autosomal recessive model) in the Exome Aggregation Consortium database (ExAC) and in 3,653 ethnically-matched controls available in-house controls, and with a Combined Annotation Dependent Depletion (CADD) score ≥20 (i.e. strongly predicted to be deleterious). We retained one homozygous variant in one individual CAKUT case: a homozygous truncating mutation in EFCAB12 found in a case with a 14.9 Mb deletion at chromosome 3q13.22–13. The variant was confirmed by Sanger sequencing according to standard protocols. Finally, we queried the ExAC database as well as the data warehouse of the Columbia Institute of Genomic Medicine (IGM), which contains whole exome data for 15,469 control individuals that are recruited for other reasons than chronic kidney disease. We did not find any homozygous or compound heterozygous truncating variants in EFCAB12. We performed clinical annotation of genes that are known to be implicated in Mendelian forms of CAKUT, by querying WES data for an in-house gene list as previously described[58]. We did not find any pathogenic variants in known CAKUT genes for all deletion carrier patients

Generation and analysis of Tbx6 mutant mice and embryos

The null allele null expression reporter allele, Tbx6, which deletes exon 2, part of exon 3 and has an H2B-EYFP fusion gene inserted in frame into exon 170, was maintained on a mixed genetic background of 129 and ICR (Taconic). B6L-Tbx6/J mice (JAX) were mated with C57BL/6Tac mice (Taconic) and maintained in a small closed colony. Embryos were collected from timed matings of mice heterozygous for either allele, with noon on the day of the plug considered embryonic day (E) 0.5. Embryos were genotyped by PCR using the following primers pairs for Tbx6 (1) 5’-GTACCATCCACGAGAGTTGTAC-3’, (2) 5’-GGGAAGAATGAGGATCCAGG-3’ to obtain a 220bp wild type allele fragment; and (3) 5’-ATTGCACGCAGGTTCTCCGG-3’; (4) 5’-GTCACGACGAGATCCTCGCC-3’ to obtain a 550bp mutant allele fragment. For Tbx6 (1) 5’-CTCGCAGCTTCACTAGTCC-3’, (2) 5’-GTGTCTGGCGTATCAGCTCA-3’ to obtain a 322bp wild type allele fragment and 504bp mutant allele fragment.

We analyzed different allele combinations for the Tbx6 null (Tbx6) and hypomorphic (Tbx6) alleles. Hematoxylin and eosin (H&E) staining: embryos were fixed overnight in 4% PFA and dehydrated through a graded ethanol series (25–50-75–100%) followed by Xylene washes. The embryos were incubated in 2 changes of paraffin wax at 55°C under vacuum. The embryos were then embedded in paraffin wax and sectioned at 10μm. Sections were dewaxed in Xylene, rehydrated through ethanol, stained with Harris haematoxylin and Eosin-Y, and finally dehydrated before mounting in Cytoseal. Immunostaining: embryos were fixed in 4% PFA. Tissues were dehydrated into 100% methanol post-fixation. For sectioning, tissues were embedded in paraffin or frozen in OCT and sectioned at various thicknesses (5–20μm). Paraffin sections were deparaffinised with Histoclear and rehydrated via an ethanol-series. Cryosections were prepared for immunolabelling by removing OCT by washing in PBS. Antibodies: Cdh1: Goat R&D AF741 at 1:400 dilution. Pax2: Rabbit Zymed 716000 1:70.

Statistical analysis

All burden metrics calculations, statistical tests and plots were performed with R v3.1 software (R Core Team 2016. R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria, see URLs) and its dplyr (Hadley Wickham and Romain Francois (2016). dplyr: A Grammar of Data Manipulation, see URLs), survival (Therneau T (2015). A Package for Survival Analysis in S. version 2.38, and ggplot2[99] packages, and the Python v2.7 language (see URLs) and its package pandas[100] and lifelines (see URLs). Unadjusted P-values were reported for all tests. Summary statistics (mean, standard deviation, median, interquartile range) were calculated for CNV metrics such as size and total span, and non-parametric Wilcoxon test was used for comparison between cases and controls. Proportions were compared using 2-sided Fisher’s Exact test. Kaplan-Meier survival curves were calculated to represent the largest CNV per genome and non-parametric log-rank test was used to compare the survival distributions between cases and controls. Logistic regression was used to test if the number of genes intersected by large and rare CNVs detected (quantitative predictor) was associated with the case-control status (outcome). The reciprocal cumulative distribution of the number of intersected genes in cases versus controls were plotted.

Data Availability Statement

Raw data that support the findings of these study will in part be available from the corresponding authors upon reasonable request and in part available from dbGaP (https://www.ncbi.nlm.nih.gov/gap; accession pending). Some restrictions may apply according to participants’ consent and privacy protection. All images generated from mouse experiments reported in this study will also be available from the corresponding authors upon reasonable request.

Code Availability Statement

Custom Perl code used for CNV comparison and annotation as well as R and Python code used in burden analyses are available upon request from the corresponding authors.

Ethical Statement

All aspects of the study involving human research participants adhered to the principles of the Declaration of Helsinki and the study protocol was approved by the Institutional Review Boards (IRB) of Columbia University Medical Center and each participating recruitment site. Signed written informed consent by the participant and/or their parents or guardians was obtained according to the protocol of local IRBs.

All animal experiments followed protocols approved by the Institutional Animal Care and Use Committee at Columbia University.