Typical and atypical ADPKD: predicted pathogenic genetic variants and population frequencies

ABSTRACT

Background and hypothesis

The population frequency of AD polycystic kidney disease (ADPKD) has not been known with certainty. We hypothesized that more accurate population frequencies for typical (PKD1, PKD2) and atypical (GANAB, ALG9, DNAJB11 ALG5, IFT140, NEK8) ADPKD were possible using a computational approach.

Methods

Initially, this study calculated the number of predicted pathogenic structural, null, and missense variants in the ADPKD-associated genes in gnomAD v.2.1.1, and compared this with disease-causing changes according to the ClinVar, HGMD, or LOVD databases. However, because of the difficulty assessing missense variant pathogenicity, ClinVar assessments were used instead for gnomAD v.4.1, which includes a larger cohort, and more structural and copy number data.

Results

Predicted pathogenic variants were found in the typical ADPKD genes in 1 in 314 people, with PKD1 and PKD2 changes present in 1 in 417 and 916, respectively, using ClinVar assessments of gnomAD v.4.1 variants and loss-of-function structural and copy number changes. Predicted pathogenic null and missense variants in the atypical ADPKD genes were found in 1 in 283 people in gnomAD v.4.1 using ClinVar. No pathogenic missense changes in the kinase domain of NEK8 previously-reported in monoallelic disease were present.

Conclusion

Predicted pathogenic variants in the genes associated with typical and atypical ADPKD are more common than suspected previously. Predicted pathogenic variants are more common for atypical than for typical ADPKD, especially when loss-of-function structural and copy number variants are included.

KEY LEARNING POINTS.

What was known:

• The population frequency of typical AD polycystic kidney disease (ADPKD) has not been known with certainty.

• In addition, the population frequency of atypical ADPKD was not known.

This study adds:

• Predicted pathogenic variants in typical ADPKD genes (PKD1, PKD2) occur in 1 in 314 of the population.

• Predicted pathogenic variants in PKD2 represent one-third of the variants seen in typical ADPKD genes.

• Predicted pathogenic variants in atypical ADPKD genes appear to be more common than in those for typical disease.

Potential impact:

• Clinicians and genetic testing laboratories should be more aware of patients with undiagnosed ADPKD.

• Atypical ADPKD may explain many people with multiple kidney cysts.

INTRODUCTION

Autosomal dominant polycystic kidney disease (ADPKD) is a genetic cystic kidney disease that accounts for at least 5% of all patients requiring renal replacement therapy [1, 2]. Typical ADPKD results from pathogenic variants in PKD1 or PKD2 and is associated with bilateral kidney enlargement from multiple cysts. PKD2 variants result in milder disease with later onset kidney failure. Atypical PKD is characterised by various imaging patterns with unilateral or localised cysts, normal sized kidneys [3] and often less-impaired kidney function. Atypical PKD results from pathogenic variants in one of at least six genes according to the ClinGen Gene Curation Expert Panel (GANAB, ALG9, DNAJB11 ALG5, IFT140, NEK8) (Table 1) [4]. Kidney cysts also occur in syndromic diseases such as tuberous sclerosis and Alport syndrome [5–7].

Table 1:

Genes, proteins, protein functions and variant characteristics in ADPKD.

Gene and protein, mode of inheritance	Protein function	Protein size and variant characteristics
PKD1 (Polycystin 1, Transient Receptor Potential Channel Interacting MIM: 601313), AD	Component of the heterotrimeric ion channel formed by polycystin 1 and 2; involved in mechanosensation by the primary cilium	4303 amino acids, all types of pathogenic variants located throughout the gene
PKD2 (Polycystin 2, Transient Receptor Potential Cation Channel MIM: 173910), AD	Component of the heterotrimeric polycystin channel complex, comprising one polycystin 1 and three polycystin 2 chains	968 amino acids, all types of pathogenic variants located throughout the gene
ALG5 (Asparagine-Linked Glycosylation 5 Homolog MIM: 604565), AD	Involved in protein N-linked glycosylation	324 amino acids, no variants reported in Simple ClinVar at 14 July 2021
ALG9 (Asparagine-Linked Glycosylation 9 Homolog MIM: 606941), AD	Encodes alpha-1,2-mannosyltransferase enzyme which is important in lipid-linked oligosaccharide assembly	611 amino acids, all types of pathogenic variants located throughout the gene
DNAJB11 [DnaJ Heat Shock Protein Family (Hsp40) Member B11) MIM: 611341], AD	Acts as co-chaperone in the folding, trafficking or degradation of proteins	358 amino acids, all types of pathogenic variants located throughout the gene
GANAB (Glucosidase II Alpha Subunit MIM: 104160), AD	Required for polycystin-1 and polycystin-2 maturation and localization to the cell surface and cilia	944 amino acids, all types of pathogenic variants located throughout the gene
IFT140 (Intraflagellar Transport 140 MIM: 614620), AD and AR	Encodes a subunit of the intraflagellar transport complex A, which is important in primary cilia, transport, signalling and development	1462 amino acids, all types of pathogenic variants located throughout the gene
NEK8 (NIMA Related Kinase 8 MIM: 609 799), AD and AR	May regulate ciliary biogenesis through targeting proteins to cilia	692 amino acids, pathogenic variants localised to the kinase domain

Information from OMIM and Simple ClinVar [19].

AD, autosomal dominant; AR, autosomal recessive.

The population frequencies of the different forms of ADPKD are important because they improve clinician awareness, support health service planning, and encourage genetic testing and the development of new treatments. However, they have not been known with certainty. Previous estimates from clinical diagnoses suggest 1 in 2459 people have ADPKD [8], the frequency from health service data is 1 in 2326 [9], and cumulative data from eight epidemiological studies have indicated one in 3704 [10]. However half the people with typical ADPKD are not diagnosed during life [8] and these studies did not include people with asymptomatic or unrecognized disease. A more accurate population frequency is available from a bioinformatic analysis of a DNA variant dataset such as the Genome Aggregation Database (gnomAD). One such study found pathogenic PKD1 and PKD2 variants in 1 in 1471 and 3846 people, respectively, resulting in a population frequency for typical ADPKD of one in 1075 [11]. Variants may be difficult to assess accurately for pathogenicity and when “likely pathogenic” variants were included, the population frequencies for PKD1 and PKD2 variants increased to 1 in 709 and 3030, respectively, or 1 in 575 overall. When missense variants were considered pathogenic based on being positive in 12 of 16 bioinformatic prediction scores [11], the population frequencies increased further to 1 in 310 and 787, or 1 in 222 overall. However, since that publication, the bioinformatics tools for assessing pathogenicity have improved, ClinVar now includes assessments for many more PKD1 and PKD2 variants, and the latest version of gnomAD (v.4.1) includes more structural and copy number changes, which all increase the accuracy of calculated population frequencies for typical ADPKD. Importantly, genes are now associated with atypical PKD, and the number of predicted pathogenic variants in these genes may be estimated too.

The aim of this study was to determine the number of predicted pathogenic variants in both typical and atypical ADPKD in the general population (gnomAD v.2.1.1) using a computational approach; or from gnomAD variants identified as pathogenic in ClinVar (https://www.ncbi.nlm.nih.gov/clinvar/), the Human Gene Mutation Database (HGMD) (https://www.hgmd.cf.ac.uk/ac/index.php), or the Leiden Open Variation Database (LOVD) (https://databases.lovd.nl/shared/genes) [12]. A similar strategy has been used previously for Alport syndrome, Gitelman syndrome, Menkes disease, Wilson disease, and various mucopolysaccharidoses [11, 13–16], and the results confirmed independently with histology or biochemical tests in at least two of these diseases. However, our strategy proved to be insensitive for pathogenic missense variants and, we subsequently used ClinVar assessments of pathogenic and likely pathogenic changes from the larger gnomAD v.4.1 dataset which also includes more structural and copy number changes. Importantly, not everyone with a predicted pathogenic change develops kidney cysts because some variants are incompletely penetrant, and it is not clear how well the population frequencies of predicted pathogenic variants correlate with disease frequencies.

MATERIALS AND METHODS

Population datasets

gnomAD v.2.1.1 comprises whole-exome sequencing (n =141,456) and whole-genome sequencing (n =10,847) samples where about half the participants (80 000, 57%) were recruited from clinical trials of people with diabetes, cardiac, or neuropsychiatric disease, or were age and gender-matched controls (n = 60 146). Few of these samples (n = 10 847) had been studied for structural and none for copy number variants.

gnomAD v.4.1 is larger than gnomAD v.2.1.1 with ∼800 000 participants, and a greater number of samples examined for structural (n = 63 046) and copy number (n = 464,297) changes. gnomAD v.2.1.1 overlaps with gnomAD v.4.1 by ∼20% (141 456/807 162), but neither database includes clinical data so that pathogenicity cannot be assessed using the ACMG/AMP criteria [17].

All participants whose demographic and genomic data are included in gnomAD had consented to the further use of their anonymized information for research and Ethics committee approval was not required for this study.

Genes and variants

Genes for typical (PKD1, PKD2) and atypical (ALG5, ALG9, DNAJB11, GANAB, IFT140, NEK8) ADPKD were studied. The genes for atypical PKD were determined by the ClinGen Gene Curation Expert Panel according to their gene-disease validity framework [18]. All genes in Simple ClinVar, except ALG5 for which there was no entry, included both nonsense and missense variants and were affected throughout their sequence (Fig. 1, Simple ClinVar [19], https://simple-clinvar.broadinstitute.org/). However, monoallelic NEK8-associated atypical ADPKD has recently been demonstrated to result from variants (p.Arg45Trp, Ile150Met, Lys157Gln) affecting residues 4–258 in the Ser-Thr kinase domain [20] (https://www.uniprot.org/uniprotkb/Q86SG6/entry#family_and_domains). This study assessed NEK8 variants as for the other genes, and, in addition, for the reported missense variants in the kinase domain.

Figure 1:

Pathogenic and likely pathogenic variants in ADPKD from Simple ClinVar. Pathogenic or likely pathogenic variants reported to ClinVar by 14 July 2021 are shown above the horizontal line; and those present in gnomAD are shown below the line. Protein-truncating variants are colored gray, and missense variants orange. No pathogenic or likely pathogenic variants had been reported in ALG5 at this time but have been subsequently [19].

Population frequency of predicted pathogenic variants for typical and atypical ADPKD using gnomAD v.2.1.1

Variants in the genes were downloaded (December 2023 to June 2024) from gnomAD v.2.1.1 (GRCh37/hg19) [21], annotated with ANNOVAR (https://annovar.openbioinformatics.org/), and assessed for pathogenicity as described previously [21] (Fig. 2).

Figure 2:

Our filtering strategy for identifying predicted pathogenic variants in gnomAD v.2.1.1. Copy number variants were available for gnomAD v.4.1.

Structural variants

Structural variants that were deletions, affected an exon and were classified as loss of function in gnomAD were assessed as predicted pathogenic. The total number of variants was corrected for the smaller number of people who were studied for structural changes. All structural variants assessed as predicted pathogenic were included in the total count regardless of their allele frequency.

Null variants

Null variants including nonsense and frameshift changes (except those in the last exon or the last 50 nucleotides of the penultimate exon, which were not expected to undergo nonsense-mediated decay [22]), as well as canonical splice site variants were assessed as predicted pathogenic. The canonical mRNA (MANE transcripts) and the locations of the cut-offs were obtained from Alamut (https://www.sophiagenetics.com). Again, all null variants assessed as predicted pathogenic were included in the total count regardless of allele frequency.

Missense variants

Missense variants were predicted pathogenic if they were rare (≤5) and disease causing in: SIFT4G (Sorting Intolerant From Tolerant) score ≤0.05; PP2 (Polymorphism Phenotyping v.2, PolyPhen-2) score ≥0.95 (http://genetics.bwh.harvard.edu/pph2/); and MT (MutationTaster) if they were “disease causing or probably deleterious” (D) or “disease-causing automatic and known to be deleterious” (A) (https://www.mutationtaster.org/info/documentation.html). Missense variants were required to be rare to be assessed as predicted pathogenic. The appropriate cutoff for allele frequency depends on how common the disease is, how many genes are affected, and how many pathogenic variants affect the gene, which differs for each gene. An allele frequency was set at ≤5 because most predicted pathogenic variants in the ADPKD genes were found in only one person, or, at most, up to five.

Missense variants were also examined for conservation of the affected amino acid, that is, the variant was the same (*) or similar (:) in three vertebrate species (chicken, mice, humans), using Clustal Omega (https://www.ebi.ac.uk/Tools/) and the Ensembl reference sequences (http://asia.ensembl.org/index.html).

Accuracy of our approach for assessing predicted pathogenic missense variants

Our strategy's accuracy was assessed as follows. Predicted pathogenic missense PKD1 and PKD2 variants were very common, so were further evaluated in Alamut. Variants with an Alamut pathogenicity score of ≥5 (variant of uncertain significance) or ≥6 (likely pathogenic) were recorded separately in the expectation that the true population frequencies lay between these.

In addition, we assessed our method's sensitivity, specificity, and positive and negative predictive values using pathogenic/likely pathogenic missense variants and benign/likely benign missense variants from the LOVD database (Supplementary Table S1). This demonstrated that missense variants with an Alamut score of ≥5 or ≥6 were still insensitive for pathogenic variants in both PKD1 and PKD2.

Finally, we examined all the genes for missense variants with a REVEL (Rare Exome Variant Ensembl Learner) score >0.932²³, which differentiates pathogenic changes from those of unknown significance. Population frequencies for each gene were then calculated including missense variants with this REVEL score.

Calculating the population frequencies of predicted pathogenic variants in the genes for ADPKD

The total number of people with a predicted pathogenic variant was divided by the mean number who had undergone gene sequencing in gnomAD (v.2.1.1, v.2.1.1 Controls; and gnomAD v.4.1), assuming that each affected person had inherited only one predicted pathogenic variant. The calculations for IFT140 and NEK8 were based on monoallelic inheritance, although both genes are also associated with biallelic disease [20, 24].

Population frequencies of predicted pathogenic variants using different variant databases

The population frequencies of predicted pathogenic variants were also calculated from the number of gnomAD changes assessed as pathogenic or likely pathogenic in ClinVar, HGMD, or LOVD. Numbers were counted even if variants were present in more than five people because these database assessments were considered likely to be accurate. For ClinVar, variants with a conflicting assessment (a variant of uncertain significance and a pathogenic or likely pathogenic, but not benign or likely benign, variant), were also included as predicted pathogenic.

Population frequency of predicted pathogenic variants in typical and atypical ADPKD genes using ClinVar and gnomAD v.4.1

Our initial studies suggested that our strategy was insensitive for identifying predicted pathogenic variants in gnomAD v.2.1.1 and ClinVar assessments were more sensitive and therefore more accurate. The accuracy was possibly because ClinVar assessments were submitted from accredited genetic testing laboratories, often from patients with suspected kidney disease, and used both the ACMG/AMP criteria and recent data on rarity. However, ClinVar assessments were not available for each gnomAD variant.

The ClinVar evaluation of gnomAD v.4.1 was then undertaken. Variants in the typical (PKD1, PKD2) and atypical (ALG5, ALG9, DNAJB11, GANAB, IFT140, NEK8) ADPKD genes were downloaded from gnomAD v.4.1 between August and November 2024, and checked for ClinVar assessments, using the same approach as previously [21]. gnomAD v.4.1 also includes copy number variants, which were classified as predicted pathogenic where gnomAD categorised them as loss of function and they affected an exon. The number was corrected for the smaller cohort size.

The population frequencies were then calculated for gnomAD variants considered pathogenic/likely pathogenic in ClinVar and loss-of-function structural and copy number changes in each gene associated with typical or atypical ADPKD.

Statistical analysis

Results were compared with chi-squared calculations with Yates’ correction using GraphPad (https://www.graphpad.com/).

RESULTS

Population frequencies of predicted pathogenic variants in typical and atypical ADPKD genes using gnomAD v.2.1.1

Population frequency of predicted pathogenic variants in typical ADPKD genes using our strategy

Overall, the population frequency for predicted pathogenic variants in typical ADPKD (PKD1 plus PKD2) from gnomAD v.2.1.1 using our strategy and Alamut scores was 1 in 964 people for scores ≥6 (VUS, likely pathogenic or pathogenic) or one in 155 for scores ≥5 (likely pathogenic or pathogenic) (Supplementary Table S1).

Our strategy evaluating PKD1 found 48 predicted pathogenic variants in 75 people with an Alamut score ≥6 or 395 variants in 625 people with a score ≥5 (Table 2). These corresponded to predicted pathogenic variants in 1 in 1400 or 1 in 168 people, respectively.

Table 2:

Population frequencies of predicted pathogenic variants in ADPKD genes in gnomAD v.2.1.1 using our strategy and Alamut scores for missense variants.

	Pathogenic Structural variants (n = 10 847, corrected for whole cohort)	Pathogenic null variants	Predicted pathogenic missense variants		Total Pathogenic variants
			No. of variants and affected people		No. of variants and affected people
Gene	No. of variants and affected people	No. of variants and affected people	Alamut score ≥6	Alamut score ≥5	Alamut Score ≥6	Alamut score ≥5	Number of people with a predicted pathogenic variant in cohort	Population frequency in cohort
PKD1 (n = 105 068)	0	43 variants in 70 people	5 in 5 people	352 in 555 people	48 in 75 people	395 in 625 people	75 (≥6); or 625 (≥5)	1 in 1 400 (≥6); or 1 in 168 (≥5)
PKD2 (n = 109 908)	0	28 variants in 37 people	2 in 2 people	27 in 46 people	30 in 39 people	55 in 83 people	39 (≥6); 83 (≥5)	1 in 2 818 (≥6); or 1 in 1 324 (≥5)
Typical PKD genes	PKD1 plus PKD2 variants				78 variants in 114 people (≥6); 476 variants in 658 people (≥5)		1 in 964 people (≥6); 1 in 155 people (≥5)
ALG5 (n = 113 325)	14 (1 × 14) variants in 14 people	d10 in 13 people	26 in 36 people		50 in 63 people		63	1 in 1 798
ALG9 (n = 117 325)	0	17 in 21 people	20 in 30 people		37 in 51 people		51	1 in 2 300
DNAJB11 (n = 113 325)	0	12 in 12 people	18 in 22 people		30 in 34 people		34	1 in 3 033
GANAB (n = 120 987)	0	21 in 26 people	5 in 9 people		26 in 35 people		35	1 in 3 457
IFT140 (n = 118 804)	28 (2 × 14) variants in 28 people	116 in 164 people	154 in 281 people		298 in 473 people		473	1 in 251
NEK8 (n = 122 172)	0	50 in 71 people	59 in 92 people		109 in 163 people		163	1 in 750
Atypical PKD genes					550 variants in 819 people		1 in 148 people
Typical and atypical PKD: 628 variants in 933 or 1 in 130 people (PKD1 and PKD2 score of at least 6) Or typical and atypical PKD: 1000 variants in 1527 or 1 in 79 people (PKD1 and PKD2 score of at least 5)

For PKD2, there were 30 predicted pathogenic variants in 39 people with a score ≥6 and 55 variants in 83 people with a score ≥5, corresponding to predicted pathogenic variants in 1 in 2818 and 1 in 1324 people, respectively.

Structural variants were not detected in PKD1 or PKD2 in gnomAD v.2.1.1 (Table 2). Null variants were common in both PKD1 and PKD2 occurring in 70 and 37 people corresponding to predicted pathogenic variants in 1 in 1500 and 1 in 2970 people, respectively, making null variants nearly twice as common in PKD1 as PKD2.

For PKD1 missense variants, five people had an Alamut score ≥6 and 555 had an Alamut score ≥5. For PKD2, two people had an Alamut score of ≥6, and 46 had an Alamut score ≥5.

Population frequency of predicted pathogenic variants in typical ADPKD in the control cohort

The population frequencies of predicted pathogenic variants in PKD1 and PKD2 were 1 in 2181 and 1 in 2771 people in the control cohort, respectively (Table 3). The overall population frequency of predicted pathogenic variants in PKD1 and PKD2 variants in the controls was 1 in 1240, which was not different from the whole cohort including missense variants with an Alamut score ≥6 (P = .21).

Table 3:

Population frequencies of predicted pathogenic variants in ADPKD genes in gnomAD v.2.1.1 using our strategy, controls and other databases.

Gene	Our assessment of whole cohort using Alamut scores		Our assessment of controls using Alamut scores		ClinVar P/LP and conflicting (VUS/P/LP)	HGMD pathogenic	LOVD pathogenic or likely pathogenic
	No. people with pathogenic variant	No. of people tested	No. people with pathogenic variant	No. of people tested	No. of variants and people	No. of variants and people	No. of variants and people
PKD1 (n = 105 068)	75 (≥6)	105 068	21	45 806	31 in 109 people	140 in 13 221 people	36 in 56 people
	1 in 1 400		1 in 2 181		1 in 964	1 in 8	1 in 584
PKD2 (n = 109 909)	39 (≥6)	109 909	17	47 113	19 in 27 people	31 in 33 people	20 in 34 people
	1 in 2 818		1 in 2 771		1 in 4 071	1 in 3 331	1 in 3 233
Typical PKD1 plus PKD2	1 in 964 (≥ 6) (Ref); 1 in 155 people (≥ 5)		1 in 1 240 people P = .21 (cf whole cohort ≥6)		1 in 714 people P = .13 (cf ≥6)	1 in 8 people P < .0001 (cf ≥6)	1 in 1 221 people P = .11 (cf ≥6)
ALG5 (n = 113 325)	63	113 325	21	49 592	1 in 2 people	1 in 2 people	1 in 2 people
	1 in 1 798		1 in 2 361		1 in 56 663	1 in 56 663	1 in 56 663
ALG9 (n = 117 325)	51	117 325	24	49 947	4 in 6 people	Not present	Not present
	1 in 2 300		1 in 2 081		1 in 19 554	Not present	Not present
DNAJB11 (n = 103 135)	34	103 135	11	45 661	4 in 4 people	5 in 5 people	2 in 2 people
	1 in 3 033		1 in 961		1 in 25 784	1 in 20 627	1 in 51 678
GANAB (n = 120 987)	35	120 987	19	52 018	4 in 4 people	2 in 2 people	1 in 1 person
	1 in 3 457		1 in 2 738		1 in 30 247	1 in 60 494	1 in 120 987
IFT140 (n = 118 804)	473	118 804	181	51 499	56 in 220 people	30 in 64 people	17 in 31 people
	1 in 251		1 in 285		1 in 540	1 in 1 856	1 in 3 832
NEK8 (n = 122 172)	163	122 173	78	53 220	10 in 37 people	17 in 525 people	3 in 32 people
	1 in 750		1 in 682		1 in 3 302	1 in 233	1 in 3 818
Atypical PKD	1 in 148 people		1 in 159 people		1 in 447	1 in 204	1 in 1771
Typical and atypical PKD	1 in 130 people (score ≥6) (Ref); or 1 in 79 people (score ≥5)		1 in 143 people (score ≥6)(P = .13)		1 in 299(P < .0001)	1 in 9 (P < .0001)	1 in 768(P < .0001)

VUS/P/LP, variant of uncertain significance/pathogenic/likely pathogenic. Ref, reference group. Our strategy only included missense variants found in ≤5 people. By contrast, all disease-causing pathogenic variants in ClinVar, HGMD, or LOVD were included because their assessments were for patients from accredited testing laboratories.

Population frequency of predicted pathogenic variants in typical ADPKD using other variant databases

Population frequencies for predicted pathogenic variants in PKD1 and PKD2 were then deduced from pathogenic variants found in ClinVar (1 in 964, 1 in 4071), HGMD (1 in 8, 1 in 3331) and LOVD (1 in 584, 1 in 3233), respectively (Table 3). Overall, the population frequency of PKD1 plus PKD2 variants was 1 in 714 using ClinVar, 1 in 8 using HGMD, and 1 in 1221 using LOVD. The population frequency was more common with ClinVar than with our method (1 in 714 compared with 1 in 964, Alamut score ≥6) although this was not statistically different (P = 0.13). The ClinVar assessment resulted in a population frequency between the results we obtained using Alamut scores ≥5 and ≥6.

Population frequency of predicted pathogenic variants in atypical PKD using our strategy

Our strategy found that predicted pathogenic variants in the six atypical PKD genes had a population frequency of 1 in 148 in gnomAD v.2.1.1. The population frequency for predicted pathogenic variants in typical and atypical PKD for these genes was 1 in 130 (Alamut score ≥6 for typical ADPKD) or 1 in 79 people (score ≥5) (Table 3).

Structural variants were not commonly detected in the atypical genes in gnomAD v.2.1.1, and for all genes associated with atypical PKD, missense variants (n = 470) were about as common as null variants (n = 307). Using our strategy, IFT140 was the commonest affected gene with a population frequency of 1 in 251; NEK8 was the second commonest and found in 1 in 750 people when all variants assessed as predicted pathogenic with our method were included.

Population frequencies for the atypical ADPKD genes deduced from pathogenic variants were 1 in 447 with ClinVar, 1 in 204 with HGMD, and 1 in 1771 with LOVD (Table 3).

The population frequencies for predicted pathogenic variants in both typical and atypical ADPKD genes together were 1 in 299 (ClinVar, P < .0001), 1 in 9 (HGMD, P < .0001), and 1 in 768 (LOVD, P < .0001) compared with our strategy (Table 3).

Assessment of our strategy for estimating the population frequency of predicted pathogenic variants in the ADPKD genes

Our assessment method for missense variants in gnomAD v.2.1.1 that included an Alamut pathogenicity score of ≥5 or ≥6 for both PKD1 and PKD2 had a median sensitivity of 30% (range 20% to 65%) specificity of 98% (75% to 100%), and positive predictive value of 98% (range 44% to 100%) and negative predictive value of 60% (range 48% to 73%) (Supplementary Table S2).

When we performed a REVEL score for each missense variant, there were 34 variants in 47 people, which resulted in fewer missense changes than with our strategy (Supplementary Table S3). Using REVEL meant that predicted pathogenic PKD1 variants were present in 1 in 1459 people, and PKD2 variants in 1 in 2556. Overall, there were predicted pathogenic variants in the typical ADPKD genes in 1 in 949 people. However, predicted pathogenic variants in the genes for atypical ADPKD were present in 1 in 315 people, mainly because of the large number of structural IFT140 variants.

The variation in pathogenicity for missense variants with Alamut scores ≥5 and ≥6 in our assessment meant that the population frequency for PKD1 variants ranged from 1 in 168 to 1 in 1400, and for PKD2 variants from 1 in 1324 to 1 in 2818, for the different Alamut scores. We concluded that the actual prevalence was likely to lie between these limits for each gene, and that it was not possible to calculate the population frequency accurately using our approach.

Population frequency of predicted pathogenic variants in typical and atypical ADPKD genes using ClinVar and gnomAD v.4.1

The population frequencies of predicted pathogenic variants in typical and atypical ADPKD were then estimated using ClinVar assessments and gnomAD v.4.1. These calculations demonstrated that predicted pathogenic variants in PKD1 were present in 1 in 417 people, PKD2 variants in 1 in 916 and together typical PKD1 and PKD2 variants were found in 1 in 314 of the population (Table 4). Loss-of-function structural variants were the commonest change in PKD1, occurring in 960 people, compared with null variants, in 124, and predicted pathogenic missense changes in 789. Loss-of-function structural variants were less common in PKD2, being present in 128, compared with null variants in 474, and pathogenic missense variants in 17. These data confirm that PKD1 variants occur about twice as often as PKD2 variants, and thus that PKD2 variants represent about one-third of all pathogenic variants in typical ADPKD in the gnomAD cohort.

Table 4:

Population frequencies of predicted pathogenic variants in ADPKD genes including loss-of-function structural and copy number variants and pathogenic/likely pathogenic variants from ClinVar in gnomAD v.4.1.

			No. of variants and people from ClinVar
Gene	Loss-of-function structural variants (x16)	Loss-of-function copy number variants (x1.7)	Pathogenic/LP null variants	Pathogenic/LP missense variants	Total Pathogenic/LP null and missense variants from ClinVar	Total Pathogenic/LP variants	Population frequency in whole cohort
PKD1 (n = 787 413)	5 (80) in 60 (960) people	4 (7) in 8 (14) people	71 in 124 people	43 in 789 people	114 in 913 people or one in 862	201 in 1887 people	1 in 417
PKD2 (n = 570 761)	3 (48) in 8 (128) people	1 (2) in 2 (4) people	56 in 474 people	5 in 17 people	61 in 491 people or one in 1162	111 in 623 people	1 in 916
Typical PKD (n = 787 413)	8 (128) in 68 (1088) people	5 (9) in 10 (18) people	127 in 598 people	48 in 806 people	175 in 1404 people or one in 560	312 in 2 510 people	1 in 314
ALG5 (n = 795 881)	1 (16) in 1 (16) people	2 (4) in 2 (4) people	2 in 3 people	2 in 6 people	4 in 9 people or one in 88 431	24 in 29 people	1 in 27 444
ALG9 (n = 790 262)	None	3 (6) in 6 (11) people	5 in 24 people	2 in 802 people	7 in 826 people or one in 957	13 in 837 people	1 in 944
DNAJB11 (n = 800 790)	1 (16) in 2 (32) people	1 (2) in 1 (2) people	8 in 36 people	1 in 1 person	9 in 37 people or one in 21 643	27 in 71 people	1 on 11 343
GANAB (n = 805 322)	None	3 (5) in 4 (7) people	12 in 36 people	2 in 6 people	14 in 42 people or one in 19 174	19 in 49 people	1 in 16 435
IFT140 (n = 803 520)	10 (160) in 501 (8 016) people	11 (19) in 16 (30) people	65 in 1 200 people	22 in 706 people	87 in 1906 people or one in 422	266 in 9 952 people	1 in 81
NEK8 (n = 806 143)	None	3 (5) in 4 (7) people	15 in 333 people	1 in 3 people	16 in 336 people or one in 2399	21 in 343 people	1 in 2 350
Atypical PKD (incl IFT140 and NEK8) (n = 806 143)	192 in 8064 people	23 (41) in 33 (61) people	107 in 1632 people	30 in 1524 people	137 in 3156 people or one in 255	370 in 11 281 people	1 in 71
Typical plus atypical PKD (incl IFT140 and NEK8) (n = 806 143)	320 variants in 9 152 people	28 (50) variants in 43 (79) people	234 variants in 2 230 people	78 variants in 2 330 people	312 variants in 4560 or one in 177	682 variants in 13 791 people	1 in 58

LP, likely pathogenic. Pathogenic/LP missense variants also included missense variants classified as conflicting that included both VUS and pathogenic or likely pathogenic changes. The numbers of structural and copy number variants were corrected (x16; x1.7) for the smaller cohort size to make them equivalent to the whole cohort

Atypical PKD occurred in 1 in 72 people, mainly due to the large number of IFT140 apparent loss-of-function structural changes and copy number variants, and to a lesser extent three pathogenic IFT140 founder variants that were present in 474, 184, or 167 people. These were assessed as pathogenic (c.2399+1G>T) or pathogenic/likely pathogenic (c.1010–1G>A, c.634G>A) in ClinVar, and were mainly present in people with a European ancestry.

Otherwise, the population frequencies of the atypical genes varied from pathogenic IFT140 changes of 1 in 81 to 1 in 31 835 for ALG5 variants. There were very few pathogenic or likely pathogenic variants reported in ClinVar for ALG5, DNAJB11, or GANAB. Nevertheless, the overall population frequency for predicted pathogenic variants in typical and atypical ADPKD was 1 in 59 people if all the predicted pathogenic variants in IFT140 and NEK8 were interpreted correctly and included.

However, the NEK8 variants were examined further. There were nine pathogenic/likely pathogenic NEK8 variants in 18 people identified by ClinVar in gnomAD v.4.1. Eight were deletions or null variants. One was a missense change (p.Gly580Ser) that did not affect the kinase domain. None of the three reported pathogenic missense variants (p.Arg45Trp, Ile150Met, Lys157Gln) was present [20]. Thus, there were no pathogenic or likely pathogenic missense residues in the kinase domain (residues 4–258), and therefore no disease-causing variants in NEK8 in gnomAD v.4.1 according to ClinVar. The pathogenic NEK8 variants found in gnomAD are likely associated with biallelic disease.

If only predicted pathogenic null and missense variants in ALG5, ALG9, DNAJB11, GANAB, and IFT140 that were classified disease causing by ClinVar were counted, then there were 2820 variants in the gnomAD v.4.1 cohort, which corresponded to a population frequency of 1 in 283. This is still more common than for typical ADPKD, but the population frequency is likely to be even greater because of the number of apparent loss-of-function structural variants in IFT140 that may be pathogenic.

DISCUSSION

The population frequency for predicted pathogenic variants in typical ADPKD was 1 in 314 using the ClinVar assessment of PKD1 and PKD2 variants and the loss-of-function structural and copy number changes in gnomAD v.4.1. Predicted pathogenic variants in PKD1 and PKD2 were found in 1 in 417 and 916 people, respectively. These estimates were likely to be more accurate than those derived from gnomAD v.2.1.1 because of the inclusion of structural and copy number changes, but are nevertheless similar to the previous finding of 1 in 575 for pathogenic and likely pathogenic variants in typical ADPKD [11]. The actual population frequency for predicted pathogenic variants may be even greater because people with a known diagnosis of ADPKD due to PKD1 or PKD2 variants were less likely to have been recruited into gnomAD v.4.1.

The population frequency for predicted pathogenic variants in atypical ADPKD was more difficult to interpret. The frequency that assumed all monoallelic pathogenic IFT140 and NEK8 variants in ClinVar were associated with cystic kidney disease was 1 in 71. This was mainly due to the large number of loss-of-function structural and founder variants in IFT140, and by including all the NEK8 variants classified disease causing by ClinVar. However, none of the reported pathogenic NEK8 variants (p.Arg45Trp, Ile150Met, p.Lys157Gln) [20] was present in gnomAD, and there were no further missense variants in the kinase domain identified as disease causing by ClinVar. The NEK8 variants classified as pathogenic or likely pathogenic in ClinVar probably only encoded biallelic disease. On the other hand, ClinVar also identified relatively few pathogenic variants in ALG5, DNAJB11, and GANAB. The total number of null and missense changes after excluding loss-of-function structural and copy number and all NEK8 variants was 1 in 283, which is still more common than the population frequency of variants in the typical ADPKD genes.

Our final population frequencies of predicted pathogenic variants in typical and atypical ADPKD may still be underestimates. Although ClinVar assessments are generally considered accurate, with only a small risk of error [25], evaluations are not available for each gnomAD variant. This is because many people with atypical ADPKD do not undergo genetic testing, some affected genes have only recently been added to cystic kidney panels [26], and many other genes are also affected in atypical disease (PKHD1, other genes for polycystic liver disease, and some unknown [27, 28]).

However incomplete penetrance means that not everyone with a predicted pathogenic variant develops kidney cysts. About half the variants demonstrated here were missense changes that often result in less penetrant disease [29]. Nevertheless, variants from ClinVar demonstrated to be disease causing are more likely associated with cysts because many are from genetic testing of patients with kidney disease.

This study initially compared the population frequencies of predicted pathogenic variants derived from our computational approach and ClinVar for gnomAD v.2.1.1, and subsequently with ClinVar alone for gnomAD v.4.1. Our assessment for typical ADPKD in the gnomAD v.2.1.1 cohort for variants that fulfilled all criteria and had an Alamut score ≥6 was not different from the control cohort, and for ClinVar and LOVD. However, the absolute population frequencies for typical ADPKD were commonest for the ClinVar assessments: 1 in 714 for gnomAD v.2.1.1 and 1 in 314 for gnomADv.4.1. The difference is likely explained by the inclusion of more structural and copy number changes in gnomAD v.4.1 suggesting that the frequency of 1 in 314 is closer to the actual population frequency of predicted pathogenic variants in ADPKD. The ratio of PKD1: PKD2 variants in v.4.1 was ∼2:1, which is different from the previously cited 7:1 [30]. This may have occurred because the v.4.1 cohort includes healthy community-based normal people, with the unrecognised and milder cystic kidneys that occur with PKD2 variants. A higher proportion (30%) of PKD2 variants has been reported previously in a cohort with mild disease [31].

The population frequency of predicted pathogenic variants in the atypical ADPKD genes was estimated to be 1 in 71 from the ClinVar assessment of gnomAD v.4.1. This too was different from the 1 in 148 in our assessment of gnomAD v.2.1.1 (P < .0001). The frequency of 1 in 71 was largely due to the inclusion of loss-of-function structural changes as well as the three pathogenic founder IFT140 variants in gnomAD v.4.1. The founder variants were included in the frequency derived from the ClinVar assessment of gnomAD v.4.1 variants but not in our assessment of v.2.1.1 because they were present in more than five people.

IFT140 variants were the commonest change (1 in 81) found in atypical PKD and even more common than PKD1 (1 in 417) or PKD2 (1 in 916) according to ClinVar. Pathogenic variants in ALG5 (1 in 27 444), DNAJB11 (1 in 11 343), and GANAB (1 in 16 435) were much rarer. The low frequencies of IFT140 variants reported in clinical series may occur because monoallelic forms of this gene have only recently been recognized as a cause of cystic kidney disease, and fewer people with mild or incompletely penetrant disease have been tested [32]. In addition, many of the IFT140 variants noted here in gnomAD v.4.1 were structural or copy number changes that may not have been detected with previous bioinformatics approaches. Furthermore, the number of structural and copy number variants was corrected to be equivalent to the larger cohort, which may have distorted the final number of predicted pathogenic changes.

These results are different from those reported previously using “high confidence” pathogenic PKD1 and PKD2 variants including truncating variants and definitely or highly likely pathogenic changes from the Mayo Clinic database (1 in 1075); or after including likely pathogenic PKD1 and PKD2 variants (1 in 709) [11]. These differences reflect the limitations of assessing variants for pathogenicity using computational tools. The Mayo database included mainly truncating variants rather than the missense changes associated with milder disease [29], and gnomAD v.2.1.1 likely excluded patients with known genetic kidney disease or kidney failure whereas gnomAD v.4.1 comprised a more normal community-based population.

The strengths of this study were the calculation of the population frequencies of predicted pathogenic variants from the ClinVar analysis of the larger gnomADv.4.1 cohort because of ClinVar's accuracy and because gnomAD v.4.1.1 comprises a more normal population.

The study's major limitations were ClinVar's underestimation of the number of pathogenic variants because assessments were not available for each gnomAD variant. Indeed, pathogenic variants were rare for some atypical genes, few atypical genes were examined, and more genes for atypical ADPKD have still to be identified. Other limitations were that many participants had undergone whole-exome sequencing, which did not demonstrate structural or copy number variants reliably, but recent advances mean that these variants are now detected accurately. If the results are confirmed they suggest that IFT140 structural and copy number variants may be underestimated in the diagnostic setting. However, even truncating variants in IFT140 have a variable penetrance and expression, with differing numbers of large kidney cysts, few liver cysts, and often only mildly impaired kidney function [24].

In conclusion, this study used current computational tools and variant datasets to demonstrate that predicted pathogenic variants in genes associated with typical and atypical ADPKD are more common than reported previously, and that atypical disease is likely more common than typical disease. These population frequencies are still underestimates because not all gnomAD variants had a ClinVar assessment, and more disease-associated genes and pathogenic variants have still to be described. Nevertheless, the predicted pathogenic variants used to calculate population frequencies for at least typical ADPKD using ClinVar assessments are more likely to be penetrant because they were often from patients referred for genetic testing for kidney disease. Pathogenic variants in the genes affected in atypical ADPKD may be associated with milder and more variably penetrant disease.

This study demonstrates the value of assessing predicted pathogenic variants in a cohort of the general population, but improved computational tools will result in an even greater accuracy. Databases with linked clinical information will be of further value [33].

DATA AVAILABILITY STATEMENT

All of the data for this study are included in this paper or in the supplementary material.

FUNDING

We thank the Polycystic Kidney Disease Australia Foundation for their support.

AUTHORS' CONTRIBUTIONS

S.V. performed the computational analysis and drafted the paper; M.H. downloaded the genes and supervised S.V.; and J.S. supervised the project as a whole, checked the results, and finalized the manuscript.

CONFLICT OF INTEREST STATEMENT

None declared.