Statistical validation of rare complement variants provides insights on the molecular basis of atypical haemolytic uraemic syndrome and C3 glomerulopathy

Abstract

Atypical haemolytic uraemic syndrome (aHUS) and C3 glomerulopathy (C3G) are associated with dysregulation and over-activation of the complement alternative pathway. Typically, gene analysis for aHUS and C3G is undertaken in small patient numbers, yet it is unclear which genes most frequently predispose to aHUS or C3G. Accordingly, we performed a six-centre analysis of 610 rare genetic variants in 13 mostly complement genes (CFH, CFI, CD46, C3, CFB, CFHR1, CFHR3, CFHR4, CFHR5, CFP, PLG, DGKE, and THBD) from >3500 patients with aHUS and C3G. We report 371 novel rare variants for aHUS and 82 for C3G. Our new interactive Database of Complement Gene Variants was used to extract allele frequency data for these 13 genes using the Exome Aggregation Consortium (ExAC) server as the reference genome. For aHUS, significantly more protein-altering rare variation was found in five genes CFH, CFI, CD46, C3 and DGKE than in ExAC (allele frequency <0.01%), thus correlating these with aHUS. For C3G, an association was only found for rare variants in C3 and the N-terminal C3b-binding or C-terminal non-surface-associated regions of CFH. In conclusion, the RV analyses showed non-random distributions over the affected proteins, and different distributions were observed between aHUS and C3G that clarify their phenotypes.

Introduction

Atypical haemolytic uraemic syndrome (aHUS) and C3 glomerulopathy (C3G) are two severe ultra-rare renal diseases that involve dysregulation of the alternative pathway (AP) in the complement system of innate immunity. In healthy individuals, the AP eliminates unwanted pathogens without compromising host cells due to the balance between AP activator and regulatory proteins. aHUS and C3G both feature host cell attack by the AP leading to end stage renal failure. However, aHUS is a thrombotic microangiopathy with an acute presentation, whereas C3G is not a thrombotic microangiopathy, but is characterised by an abundance of C3 deposition in the renal glomeruli with a mostly chronic presentation, with some exceptions (1).

Rare genetic abnormalities in AP and thrombosis-related genes are present in ~60% of aHUS cases (2–4). These mostly drive AP dysregulation at the endothelial cell surface (1). The penetrance of predisposing rare variants (RVs) in aHUS is ∼50% and is determined by the complement factor H (CFH; FH) and membrane cofactor protein (CD46; MCP) haplotype, additional RVs, and a trigger (5–8). As reported in our 2014 aHUS database (9), most genetic aHUS cases are heterozygous (10) and are attributed to the genes CFH (25–30% of cases), followed by CD46 (8–10%), complement C3 (C3; C3) and complement factor I (CFI; FI) (4–8% each) and complement factor B (CFB; FB) (1–4%) (6, 11). The C-terminal short complement repeat (SCR)-19/20 domains of FH are a well-known RV hotspot for aHUS, this being attributed to its functional binding sites for C3b, C3d and heparin (12). [Footnote: We utilise the term “short complement repeat” rather than “short consensus repeat” as a better descriptor of the most abundant domain type in complement]. In distinction to other genes, most RVs in CD46 lead to a quantitative decrease in the protein product and approximately a quarter are homozygous (10). Rare copy number variation (CNV) leading to large genomic rearrangements (LGRs) in the CFH-complement factor H-related (CFHR; FHR) region, such as the CFH/CFHR1 and CFH/CFHR3 hybrid genes, are risk factors for aHUS (13–16). RVs in the non-complement gene thrombomodulin (THBD; THBD) account for 3–4% of genetic aHUS (6), although no THBD variants were detected in the French aHUS cohort (17). RVs in the non-complement gene diacylglycerol kinase epsilon (DGKE; DGKE) account for ~27% of aHUS presenting under the age of one year and <4% under two years, respectively (18, 19).

In C3G, complement gene RVs have been identified in ~20% of sporadic C3G cases (6, 20–22). Familial C3G is most often linked to highly penetrant heterozygous CNV in the CFHR1–5 genes, such as CFHR5 nephropathy, as well as homozygous CFH deficiency and heterozygous gain-of-function mutation in C3 (23–30). These frequently affect AP regulation in the fluid phase, with some exceptions (12). As in aHUS, unaffected carriers of these genetic abnormalities are seen, indicating that the genetic variant only predisposes for the manifestation of C3G (31). aHUS and C3G both also involve anti-CFH autoantibodies, known as acquired factors (5–13% of cases) (6, 32).

Genetic sequencing and multiple ligation-dependent probe assessment screening panels for aHUS and C3G typically include up to 10 complement (CFH, CFHR1, CFHR3, CFHR4, CFHR5, CFI, C3, CD46, CFB and complement factor properdin (CFP; FP) (33)), two coagulation (THBD and plasminogen (PLG; PLG) (34)) and one non-complement (DGKE) genes (1). RVs in six genes (CFH, CFI, C3, CD46, CFB and DGKE) are associated with aHUS, while RVs in CFH and C3 associate with C3G. These associations result from studies of many aHUS patients, and rather fewer C3G patients by linkage and familial segregation, case-control cohorts and functional studies. However, the associations of RVs in the four genes CFHR5, CFP, PLG and THBD with aHUS, and all 13 genes with C3G are less well defined (1). In addition, an increasing number of variants of unknown significance for aHUS and C3G are being identified amongst these 13 genes. These require rapid pathogenicity evaluations for clinical interpretation, e.g. almost a third of CFH variants have limited functional characterisation (35).

In order to clarify differences in the genetic and molecular basis of aHUS and C3G, we have analysed the statistical allele frequencies (AFs) of RVs in multiple patient cohorts for comparison with reference datasets, and we have developed a new Database of Complement Gene Variants (http://www.complement-db.org). The AF provides the frequency of the variant in a given population. Variant pathogenicity can be identified by comparisons with genomic reference datasets such as the Exome Aggregation Consortium (ExAC) (36). The database also provides structural biology and evolutionary tools to predict the effect of RVs on these proteins, and present functional data from the literature and ClinVar. The database was used to analyse 610 rare genetic variants from >3500 patients in six renal centres, this being the largest dataset known to date for aHUS and C3G with 371 aHUS and 82 C3G novel RVs. Following comparisons with 60,706 genomic reference sequences from ExAC (37) and 6,500 from the Exome Variant Server (EVS), we confirmed the associations of six genes (above) with aHUS, and that three genes CFHR5, PLG and THBD were not associated with aHUS. The statistical comparisons also confirmed that CFH and C3 were associated with C3G, and suggested the involvement of CFB and THBD with C3G. Our results explain how changes in the same proteins result in the different pathologies observed with aHUS and C3G. Through the use of AFs and burden testing in our database, the new database will inform patient management by enabling clinical immunologists to interpret new variants in terms of their associations with aHUS and C3G.

Materials and Methods

Data collection

The aHUS and C3G phenotype and variant data was sourced from six centres (unpublished and published data) and literature searches. aHUS was diagnosed by the presence of one or more episodes of microangiopathic hemolytic anemia and thrombocytopenia defined on the basis of hematocrit (Ht)<30%, hemoglobin <10mg/dl, serum lactate dehydrogenase (LDH)>460U/L, undetectable haptoglobin, fragmented erythrocytes in the peripheral blood smear, and platelet count <150,000/μl, associated with acute renal failure, together with a negative Coombs test, ADAMTS13 activity >10% and negative Shiga toxin (1). C3G was diagnosed by the presence of C3 deposits by immunofluorescence in the absence, or comparatively reduced presence of immunoglobulins. The identification of dense deposits within the glomerular basement membrane by electron microscopy lead to the further classification of the C3G as dense deposit disease (DDD) (1). The database included variant data in 13 genes (C3, CD46, CFB, CFH, CFHR1, CFHR3, CFHR4, CFHR5, CFI, CFP, DGKE, PLG, THBD) (Table I). The estimated variant AFs in the aHUS and C3G datasets were based on data from the six centres only. Data from the three reference genome projects, namely ExAC (Version 0.3) (37), EVS (NHLBI GO Exome Sequencing Project, Seattle, WA (http://evs.gs.washington.edu/EVS), and the 1000 Genomes project (1000GP) (38), were downloaded and used as surrogate control reference datasets for all genes apart from CFHR1, CFHR3 and CFHR4 that were involved in CNVs only. These latter three genes were excluded because, at the time of writing, no CNV data were available from ExAC, EVS or 1000GP. These three reference datasets do not contain genomes from patients with rare renal disease phenotypes that include aHUS or C3G, however ExAC and EVS contain genomes from myocardial infarction studies. While this may affect the analyses of thrombosis-related genes such as THBD, a recent study found that the ExAC was not enriched in pathogenic RVs for these diseases (39). In ExAC, only variants with “PASS” filter status were included in our analyses. The corresponding allele frequency of each variant in ExAC, EVS and the 1000 Genomes Project was queried by using Human Genome Variation Society nucleotide level nomenclature (c.) and National Center for Biotechnology Information gene accession number. Published experimental data on each variant were sourced from literature searches using PubMed. Variants were described using DNA and protein level terms for both Human Genome Variation Society and legacy nomenclature.

Table I.

Summary of mRNA and protein identifiers for the 13 genes

Gene	Protein	RefSeq (NCBI)a		Ensembla
		mRNA	Protein	Transcript ID	Protein ID
CFH	FH	NM_000186.3	NP_000177.2	ENST00000367429	ENSP00000356399
CFI	FI	NM_000204.3	NP_000195.2	ENST00000394634	ENSP00000378130
C3	C3	NM_000064.3	NP_000055.2	ENST00000245907	ENSP00000245907
CD46	MCP	NM_002389.4	NP_002380.3	ENST00000358170	ENSP00000350893
CFB	FB	NM_001710.5	NP_001701.2	ENST00000425368	ENSP00000416561
CFHR1	FHR1	NM_002113.2	NP_002104.2	ENST00000320493	ENSP00000314299
CFHR3	FHR3	NM_021023.5	NP_066303.2	ENST00000367425	ENSP00000356395
CFHR4	FHR4	NM_001201550.2	NP_001188479.1	ENST00000367416	ENSP00000356386
CFHR5	FHR5	NM_030787.3	NP_110414.1	ENST00000256785	ENSP00000256785
THBD	THBD	NM_000361.2	NP_000352.1	ENST00000377103	ENSP00000366307
CFP	FP	NM_001145252.1	NP_001138724.1	ENST00000396992	ENSP00000380189
DGKE	DGKE	NM_003647.2	NP_003638.1	ENST00000284061	ENSP00000284061
PLG	Plasminogen	NM_000301	NP_000292	ENST00000308192	ENSP00000308938

^aThe databases of annotated genomic, transcript and protein reference sequences are found at https://www.ncbi.nlm.nih.gov/refseq/ (RefSeq) and http://www.ensembl.org/index.html (Ensembl).

Data cleansing, duplications and maintenance

Patient duplicate tests were carried out within and between datasets from the six renal centres, firstly by identifying potential duplicates using the patient’s variant profile, disease, gender and year of birth. Potential duplicates were investigated by requesting the full date of birth details from each renal centre and deleting as necessary. For maintenance of the database in the long term, each collaborating group will provide an annual data update as a spreadsheet that will be automatically uploaded to the database with automated duplication and error checks that are programmed into the MySQL backend. As an open resource, other clinical centres are invited to upload formatted data updates through the curator. To prevent against SQL injection attacks and related software vulnerabilities, the database inputs were sanitised and tested using the penetration testing tool SQLMAP.

Web-database development

The Database of Complement Gene Variants (http://www.complement-db.org) was constructed using a MySQL platform and its user interface was developed using a combination of PHP, JavaScript, jQuery, CSS and HTML. The design is based on the European Association for Haemophilia and Allied Disorders coagulation Factor IX web-database (http://www.factorix.org/) (40). The database is mounted on a Linux server within the Information Services Division at University College London.

Data retrieval

Variant data were retrieved from the Database of Complement Gene Variants using simple or advanced search tools. The user first defined the data source and gene, and then the other search options, including protein domain, location, and variant type and effect. The advanced search tool featured a customisable default ExAC AF cut-off value of 1%, which was used to ensure that any variant with an ExAC AF >1% was not retrieved. Variants were mapped to their genomic location using the reference human genomes GRCh37 and GRCh38. The protein and transcript locations were identified using RefSeq and Ensembl (Table I). The database provided protein structural views of each missense variant using the JSmol Java applet (https://sourceforge.net/projects/jsmol/). These views facilitated a structural understanding of the variant using the webpage ‘structure and function’ tab. For each disease dataset, the statistics page showed (i) the distribution of variants in genes by protein domain, exon/intron location, and variant type and effect; (ii) the RV burden per gene, and (iii) the number of RVs per case. The AF webpage summarised the number of variants in each gene using their reference genome AF, with links to their variant database entries, for each disease dataset. The world map webpage showed the number of laboratory-sourced cases that have citizenship in each country, where data were available. The new variants webpage predicted the effects of any theoretical missense variants using AF, structural and functional analyses. The structures webpage listed all known protein structural models with links to their Protein Data Bank (PDB) code. The amino acid alignments webpage showed the multiple sequence alignment of up to ten vertebrate species for each protein depending on sequence availability, this being calculated using the UniProt database and the Probabilistic Alignment Kit program (PRANK) (41).

Rare variant burden

The RV burden per gene was computed for the aHUS, C3G, ExAC and EVS datasets. The RV burden was defined as the proportion of screened cases with an identified RV per gene. For aHUS and C3G, the burden was calculated by dividing the number of cases with an identified RV by the total number of cases screened per gene. For ExAC and EVS, this was calculated for each gene by dividing the total mean adjusted allele count, this being determined from the allele count minus the number of homozygous cases, by the total mean adjusted number of subjects with RVs (42). An AF cut-off value of 0.01% was used for all RV burden calculations, which is advised for a Mendelian disease (43). The RV burden was calculated for all protein-altering RV only; truncating (nonsense, frameshift and splice) and non-truncating (missense and in-frame). RVs classified as ‘benign’ and ‘likely benign’ (see next section) were not filtered out because these data were unavailable for the ExAC and EVS datasets.

RV assessment

Each non-LGR RV was classified as ‘pathogenic’, ‘likely pathogenic’, ‘uncertain significance’, ‘benign’ or ‘likely benign’ using categorisation guidelines (1) that followed the American College of Medical Genetics and Genomics and the Association for Molecular Pathology (44). The following categories were also added to cover all RVs in our datasets:

1. Minor allele frequency (MAF) <0.1%; likely benign by clinical testing (Illumina; ClinVar); in silico analyses predict: tolerated, neutral and benign. Categorised as ‘likely benign’ for all genes except for C3 and CFB which were ‘uncertain significance’ instead.

2. MAF <0.1%; likely benign by clinical testing (Illumina; ClinVar); in silico analyses predict: uncertain deleterious effects. Categorised as ‘uncertain significance’.

3. MAF <0.1%, predicted as loss-of-function (includes nonsense, frameshift and splice acceptor/donor variants) in C3 or CFB. Categorised as ‘uncertain significance’. C3 or CFB loss-of-function is unlikely to over-activate complement and lead to aHUS or C3G. Since their effects on complement have not been experimentally proven, it was safer to categorise these as ‘uncertain significance’.

4. MAF <0.1%; synonymous change; in silico analyses predict: tolerated, neutral and benign. Categorised as ‘likely benign’.

5. 0.1% < MAF <1%; no functional data; in C3 or CFB. Categorised as ‘likely benign’.

Loss-of-function RVs in all genes, except in those only involved in LGR variants or in the complement activators C3 and CFB, were classified as ‘likely pathogenic’ for aHUS and C3G, unless functional data reported otherwise (1). The Ensembl Variant Effect Predictor tool (45) was used for PolyPhen-2 (46) and SIFT (47) analyses and the results were used for combinatorial variant analyses. These classifications were made available for each variant identified in the aHUS and C3G datasets.

Spatial distribution of missense RVs in the proteins

For each protein domain, the AFs of missense RVs in the aHUS and C3G datasets were summed, and then divided by the proportion of protein residues in the corresponding domain, in order to identify mutational hotspots. Missense RVs that were categorised as ‘benign’ or ‘likely benign’ were excluded from these analyses. For FH, FI, C3, MCP and FB, where there were enough missense RVs to identify mutational hotspots, these were represented graphically as bar charts. For FH, FI, C3, MCP and FB, each unique missense RV was mapped onto the structural model for visualisation, but noting that these did not take the aHUS or C3G AF into account. The structural models for FH (PDB code 3N0J (48)), C3 (PDB code 2A73 (49)), FI (PDB code 2XRC (50)), MCP (PDB code 3O8E (51)) and FB (PDB code 2OK5 (52)) were sourced from the PDB. An alternative model for FH in which the C-terminal domains were extended and not the N-terminal domains was also used but not shown here (PDB code 3GAV) (9). The R statistical package was used for statistical analyses and artwork (http://www.R-project.org/). PyMol was used for protein structural visualisation and artwork (53).

Statistical analyses

The categorical variables shown in Tables II and VI, Supplementary Tables II, III and IV, and Figures 4 and 5 were examined using the two-tailed Chi-squared test with Yates’ correction with a 0.05 significance level that was Bonferroni-corrected where applicable. For the common variants analyses in Table II, the 0.05 significance level was Bonferroni-corrected by dividing by the 14 variants, to give 0.0036. For the patient gender analyses in Table VI, the 0.05 significance level was Bonferroni-corrected by dividing by 12 genes to give 0.0042 for aHUS, and by 11 genes to give 0.0045 for C3G. The ‘ALL’ genes category for both aHUS and C3G were each subjected to a 0.05 significance level (no Bonferroni adjustment needed). For the rare variant burden analyses presented in Figure 4 and Supplementary Tables II and III, the 0.05 significance level was Bonferroni-corrected by dividing by the 9 genes to give 0.0056. For the statistical analyses of CFH in Supplementary Table IV and Figure 5, a 0.05 significance level was used. The categorical variables shown in Figure 2B and Table III were examined using the two-tailed Fisher’s exact test with a significance level of 0.05.

Table II.

The 14 common genetic variants identified in at least one of the three reference datasets (1000GP, EVS and ExAC) at an allele frequency of ≥1%. Bold text denotes statistical significance^a,b.

Gene	Genetic Variant	Protein Variant	Dataset Allele Frequency (%)
			1000GP	EVS	ExAC	aHUS (a)	C3G (b)
C3	c.1407G>C	p.Glu469Asp	1.62a	1.41a	0.40	0.16	0
CFB	c.754G>A	p.Gly252Ser	1.02a	2.82a	2.22a	0.43	0
CFB	c.1598A>G	p.Lys533Arg	1.92a,b	0.44	1.05a	0.20	0.26
CFB	c.1697A>C	p.Glu566Ala	1.02	0.73	1.12a	0.53	0.13
CFB	c.1953T>G	p.Asp651Glu	1.04a	0.94a	0.22	0.04	0
CFH	c.1652T>C	p.Ile551Thr	1.91a	1.68a	0.50a	0.16	0
CFH	c.2669G>T	p.Ser890Ile	6.23a	6.57a	1.99a	0.34	0
CFH	c.2808G>T	p.Glu936Asp	20.33a	13.80a	19.55a	0.02	0
CFHR5	c.136C>T	p.Pro46Ser	0.90	1.13	0.70	0.63	0
CFHR5	c.1067G>A	p.Arg356His	1.04	2.09a	1.78a	1.42	0.12
CFI	c.884–7T>C	-	2.56a	0	2.31a	0.02	0
CFI	c.1534+5G>T	-	0.30	1.13	0.90	0.33	0.12
CD46	c.1058C>T	p.Ala353Val	0.40	1.25a	1.53a	0.60	0.24
PLG	c.1567C>T	p.Arg523Trp	0.24	1.01	0.68	0.35	0

^aDetermined to be significantly more common in ExAC than in aHUS using a two-tailed Chi-square test with Yates’ correction and a Bonferroni-corrected significance level of 0.0036.

^bDetermined to be significantly more common in ExAC than in C3G using a two-tailed Chi-square test with Yates’ correction and a Bonferroni-corrected significance level of 0.0036.

Table VI.

Gender of the 1231 aHUS and 116 C3G cases with an identified rare variant.

Disease	Gene	Female	Male	Unknown	Proportion of females (known) (%)	P a
aHUS	C3	76	61	99	55	0.20
aHUS	CD46	77	91	67	46	0.28
aHUS	CFB	13	11	11	54	0.68
aHUS	CFH	196	138	196	59	0.002
aHUS	CFHR1	8	13	1	38	0.28
aHUS	CFHR3	1	1	0	50	1.00
aHUS	CFHR5	4	3	0	57	0.71
aHUS	CFI	80	60	54	57	0.09
aHUS	CFP	1	1	0	50	1.00
aHUS	DGKE	16	9	0	64	0.16
aHUS	PLG	7	5	0	58	0.56
aHUS	THBD	16	8	10	67	0.10
aHUS	ALL b	440	363	428	55	0.007
C3G	C3	23	20	0	53	0.65
C3G	CD46	1	1	0	50	1.00
C3G	CFB	4	4	0	50	1.00
C3G	CFH	19	24	0	44	0.45
C3G	CFHR1	0	1	0	0	0.32
C3G	CFHR3	0	1	0	0	0.32
C3G	CFHR5	0	4	0	0	0.05
C3G	CFI	5	6	0	45	0.76
C3G	DGKE	1	2	0	33	0.56
C3G	PLG	2	5	0	29	0.26
C3G	THBD	2	7	0	22	0.10
C3G	ALL b	53	63	0	46	0.35

^aP value from a two-tailed Chi-square test with Yates’ correction using a Bonferroni-corrected significance level of 0.0042 for aHUS and 0.0045 for C3G. The ‘ALL’ genes category was subject to a significance level of 0.05 (no Bonferroni correction). Bold text denotes a P level less than the Bonferroni-corrected significance level.

^b‘ALL’ is not equal to the gene sum, because some patients possessed RVs in more than one unique gene.

Figure 4.

The RV burden (%) per gene for the nine relevant genes in the four aHUS (Allele number (AN): 634–6256), ExAC (AN: 74194–121246), EVS (AN: 8202–13005) and C3G (AN: 208–886) datasets. These were based on an ExAC MAF cut-off of 0.01%. *** denotes p < 0.0001. * denotes p = 0.0052.

Figure 5.

The distribution and disease allele frequencies (AFs) of non-benign missense RVs in the domains of FH, C3, FI, MCP, and FB in the aHUS and C3G datasets. The largest bars correspond to missense RV hotspots (e.g. FH SCR-20 for aHUS; SCR-18 for C3G). Each domain missense RV AF is normalised for its size by dividing it by the proportion of residues in the protein domain. In each of (A-E), red represents the total AF missense RVs identified in the aHUS dataset, and likewise dark blue for C3G. For those missense RVs identified in both the aHUS and C3G datasets, pink represents the AF for aHUS and light blue for C3G. On the x-axes, the domain names are shown.

(A) The total AF of missense RVs in each of the 20 SCR domains in FH. Beneath the x-axis, the functional binding sites associated with each SCR domain are shown by coloured arrows (identified in the inset).

(B) The total AF of missense RVs in each C3 domain. Beneath the x-axis, the functional binding sites associated with each C3 domain are shown by arrows to correspond to the SCR domains in FH (pink) or other sites in FI or on the cell surface. The C3d binding site on SCR-19/20 corresponds to the TED domain, however this is not shown.

(C) The total AF of missense RVs in each FI domain.

(D) The total AF of missense RVs in each MCP domain.

(E) The total AF of missense RVs in each FB domain.

Figure 2.

Summary of cases and variants in aHUS and C3G.

(A) The source of the RV data in the database for aHUS and C3G. “Both” (yellow) indicates RVs that were identified both in the laboratory-sourced datasets (purple) and published in the literature (green).

(B) The number of unique RVs (0 to 3) per patient case in the aHUS and C3G datasets, totalling 3127 patients. For aHUS, there is a further case with four RVs that is too small to be seen.

(C) A matrix showing the genetic profiles of the 182 aHUS cases with compound heterozygous (in single or in two different genes) or homozygous RVs from panel (B). Their frequencies n are graded in colour (inset).

(D) A matrix showing the genetic profiles of the 31 C3G cases with compound heterozygous (in single or in two different genes) or homozygous RVs from panel (B), graded in colour (inset).

Table III.

The total number of aHUS and C3G cases screened per gene

Gene	Disease	Total number of cases screened by the reference centres 1–6
		1	2	3	4	5	6	All
C3	aHUS	410	480	286	252	409	618	2455
CD46	aHUS	524	480	286	461	578	613	2942
CFB	aHUS	395	480	286	328	350	618	2457
CFH	aHUS	662	480	286	483	578	639	3128
CFHR1	aHUS	7	250		442			699
CFHR3	aHUS		250		348			598
CFHR5	aHUS			286		31		317
CFI	aHUS	533	480	286	425	578	621	2923
DGKE	aHUS		76	286	191	150		703
PLG	aHUS			286				286
THBD	aHUS		480	286	348	214		1328
C3	C3G	115	160	104				379
CD46	C3G	142	160	104				406
CFB	C3G	115	160	104				379
CFH	C3G	179	160	104				443
CFHR5	C3G			104				104
CFI	C3G	144	160	104				408
DGKE	C3G		23	104				127
PLG	C3G			104				104
THBD	C3G		160	104				264

Centre 1: Institute of Genetic Medicine, Newcastle University, Newcastle upon Tyne, United Kingdom

Centre 2: Clinical Research Center for Rare Diseases “Aldo e Cele Daccò”, IRCCS-Istituto di Ricerche Farmacologiche “Mario Negri”, Ranica Bergamo, Italy

Centre 3: Molecular Otolaryngology and Renal Research Laboratories, Carver College of Medicine, University of Iowa, Iowa City, Iowa, United States of America;

Centre 4: Department of Cellular and Molecular Medicine, Center for Biological Research and Center for Biomedical Network Research on Rare Diseases, Madrid, Spain;

Centre 5: Assistance Publique-Hopitaux de Paris, Hôpital Européen Georges Pompidou, Service d’Immunologie Biologique, Paris, France;

Centre 6: Department of Pediatric Nephrology, Radboud University Medical Center, Nijmegen, The Netherlands

For each row of one independent t-test (protein-altering only) in Supplementary Tables 2 and 3, we undertook a power calculation using the program PS Power and Sample Size Calculations (Version 3.0) (54). Our calculations used expected differences of: ≥ 5% for CFH, C3 and CD46, ≥ 2.6% for DGKE in the aHUS and EVS groups, and ≥ 2.5% for DGKE in the other three groups, and ≥ 2% for all other genes, and took into account the unequal sizes of the experimental (aHUS or C3G) and control (ExAC or EVS) groups. If the true difference between the experimental and control groups was as expected, we were able to reject the null hypothesis that their frequencies were equal with a certain probability (power). The Type I error probability associated with all of these tests was 0.0056. The power was at least 80% in all tests apart from the following cases in Supplementary Table II, with ExAC as the reference dataset, for aHUS: DGKE (74%), and for C3G: DGKE (55%) and PLG (32%), and also in Supplementary Table III, with EVS as the reference dataset, for aHUS: DGKE (67%), and for C3G: DGKE (68%), PLG (10%), THBD (38%) and CFB (41%).

AF analyses

A two-tailed chi-squared test with Yates’ correction was used to assess the difference in protein-altering RV AF (ExAC AF < 0.1%) between the aHUS/C3G and reference datasets. Protein-altering variants included non-truncating and truncating variants only (42). For each RV, the 0.05 significance level was Bonferroni-adjusted by dividing by the number of RVs identified in its gene for aHUS and ExAC (C3: 485, CD46: 191, CFB: 400, CFH: 469, CFHR5: 262, CFI: 215, DGKE: 193, PLG: 263, THBD: 145), and for C3G and ExAC (C3: 452, CD46: 130, CFB: 391, CFH: 334, CFHR5: 261, CFI: 173, DGKE: 176, PLG: 263, THBD: 143) and rounding to 4 decimal places to give significance levels of 0.0001, 0.0002, 0.0003 or 0.0004 accordingly. The AF was not available in the aHUS dataset for the two CFP RVs so CFP was not analysed. CFHR1, CFHR3 and CFHR4 were not analysed because large genomic rearrangements were not included in ExAC at the time of writing. Related individuals were taken out the analyses by including familial alleles only once (twice for homozygous cases). This procedure refers to the AF assessments that were displayed on the web database, and were also used to verify that none of the RVs were significantly more common in ExAC than in aHUS or C3G.

Results

Genetic variants in aHUS and C3G

In order to perform the comparisons between aHUS and C3G using AF analyses, six Reference Centres from the United Kingdom, France, Italy, Spain, Holland and the United States of America provided phenotype and variant data sets for 3128 and 443 patients from National and International Registries. The total of disease variants was 543 for aHUS and 111 for C3G in 13 genes (Figure 1). The aHUS and C3G variants were analysed in terms of their reference AFs for these variants found in three reference datasets (1000GP, EVS and ExAC). First, 10, 10 and 8 common variants with AF > 1% were filtered out (red in Figure 1). The RVs that were retained were those with an AF of <1% in the reference datasets (light blue, green, dark blue, and pink, Figure 1), or were rare LGRs not covered by the reference datasets. By this filtering, a total of 610 retained RVs (96–97% of the total) were identified in 13 genes (Table I) for aHUS (542 variants) and C3G (110 variants), with 42 being shared by aHUS and C3G, including the LGRs. The remaining 3–4% of common variants with AF >1% in at least one of the reference datasets comprised 14 and four in aHUS and C3G respectively (Table II). These 14 and four variants were only just over the 1% AF threshold except for three (CFH p.Ser890Ile, CFH p.Glu936Asp and CFI c.884–7T>C). In the aHUS dataset, 14% (75), 19% (103), and 32% (174) of variants were found to have an AF between 0–1% (light blue/green/dark blue, Figure 1A) in the 1000GP, EVS and ExAC reference datasets, respectively. In the C3G dataset, these proportions were significantly higher, being 31% (35), 46% (52) and 58% (65) respectively (light blue/green/dark blue, Figure 1B). In confirmation of this outcome, all three analyses gave p<0.0001 in the two-tailed Fisher’s exact test. This outcome was still true when the ‘benign’ and ‘likely benign’ RVs were excluded from our aHUS and C3G datasets (p<0.0001; Fisher’s exact test) (see below). Our aHUS and C3G RV datasets were compared with the literature (9) in order to determine how many RVs were novel. Of the 542 aHUS and 110 C3G RVs, 56% (371) and 58% (82) respectively were not found in the literature and therefore novel (purple, Figure 2A). The RVs that were reported in the literature but not in the six Reference Centres were not part of our current analyses because we could not calculate their aHUS and C3G AF values (green, Figure 2A). However, they are included in the updated database from this study.

Figure 1.

Stacked bar analyses showing the reference AF of the variants identified in the (A) aHUS and (B) C3G datasets. The total numbers of unique variants (n_var) are shown within the bars. A unique variant is defined as, when a variant is seen in more than one patient, the variant is only counted once. Unique variants (excluding the 13 aHUS and 3 C3G CNVs) are categorised by their AF in each of the three reference datasets: the 1000 Genomes Project (1000GP) with an allele number (AN; total number of alleles screened for each gene) of 3775–5008, the Exome Variant Server (EVS) with an AN of 8202–13005 and the Exome Aggregation Consortium (ExAC) with an AN of 14708–121412. The aHUS and C3G datasets each had ANs of 634–6256 and 208–886, respectively. The pink bars indicate a reference AF of 0%, dark blue bars indicate a reference AF of between 0% and 0.01% (non-inclusive), green bars indicate a reference AF of between 0.01% (inclusive) and 0.1%, light blue bars indicate a reference AF of between 0.1% (inclusive) and 1%, and red bars indicate a reference AF ≥ 1%.

RV frequencies in cases

The frequencies of the RVs in the aHUS and C3G datasets were surveyed. First, we investigated differences in the screened cases with RVs for aHUS and C3G. The aHUS and C3G datasets comprised 3128 aHUS and 443 C3G patients respectively (Figure 2B; Table III). Of these totals, 1231 (39%) aHUS and 116 (26%) C3G patients harboured 542 and 110 unique RVs respectively, with an overlap of 42 unique variants between them. A significantly greater proportion of screened aHUS patients had at least one RV compared to C3G patients (p<0.0001; two-tailed Fisher’s exact test). There were 0.44 unique RVs per case in aHUS, compared to 0.95 in C3G. This suggested that almost every C3G case had a different, unique RV, whereas aHUS cases were more likely to share the same RV. The proportion of aHUS patients with RV (1231/3128=39.3%) (Figure 2B) was low compared to literature reports of ~60%. This difference most likely arose from the omission in the AF analyses of the 119 literature-sourced aHUS RVs in the web-database, for which no aHUS patient AF data from the six centres were available (green, Figure 2A).

RV profiles of cases

Next, we compared the genetic RV profiles of aHUS and C3G patients. Among the 1231 aHUS and 116 C3G patients with RVs, 1024 (83%) and 82 (68%) respectively harboured a single RV in one of the 13 genes (Figure 2B). Two RVs were identified in 182 (15%) aHUS and 31 (29%) C3G cases (light blue bars; Figure 2B), three in 24 (2%) and 3 (2%) respectively (red bars; Figure 2B), and four in 1 (<1%) and 0 (0%) respectively. However, the number of RVs per aHUS case in the dataset from Reference Centre 6 was unknown except for three homozygotes. All other RV occurrences in this dataset were analysed as one heterozygous aHUS case. When the 322 aHUS cases from Reference Centre 6 were excluded, 702/909 (77%) harboured a single RV, 182/909 (20%) had two, 24/909 (3%) had three, and 1 (<1%) had four RVs. Excluding Reference Centre 6, no significant difference was seen between aHUS and C3G in the cases with two RVs (p=0.113; two-tailed Fisher’s exact test). For aHUS, cases with two RVs in CD46 (23 cases including 19 homozygotes; 11% of all aHUS cases with CD46 RVs), were the most frequent, followed by CFH (37 cases, including 18 homozygotes; 9%) (Figure 2C; Table IV). For C3G, cases with two RVs in CFH (7 cases including 6 homozygotes; 30%) were the most frequent, followed by C3 (6 cases including 2 homozygotes; 23%) (Figure 2D; Table IV). These analyses showed differences between aHUS and C3G, summarised in Figures 2C, 2D. No CFHR3 RVs were seen in any aHUS or C3G cases that had more than one RV.

Table IV.

Homozygous RVs present in aHUS and C3G

Gene	cDNA change	Protein change	Classification	Condition	Number of patients
C3	c.3322_3333del	p.1108_1111del	Likely benign	aHUS	1
CD46	c.97G>C	p.Asp33His	Uncertain significance	aHUS	2
CD46	c.100G>A	p.Ala34Thr	Uncertain significance	aHUS	1
CD46	c.286+2T>G	-	Pathogenic	aHUS	4
CD46	c.286+1G>C	-	Pathogenic	aHUS	1
CD46	c.496T>C	p.Cys157Arg	Uncertain significance	aHUS	1
CD46	c.535G>C	p.Glu179Gln	Pathogenic	aHUS	1
CD46	c.565T>G	p.Tyr189Asp	Likely pathogenic	aHUS	1
CD46	c.718T>C	p.Ser240Pro	Pathogenic	aHUS	2
CD46	c.736T>A	p.Phe246Ile	Uncertain significance	aHUS	1
CD46	c.800_820del	-	Uncertain significance	aHUS	1
CD46	c.810T>G	p.Cys270Trp	Uncertain significance	aHUS	1
CD46	c.811_816delGACAGT	p.Asp271_Ser272del	Pathogenic	aHUS	1
CD46	c.881C>T	p.Pro294Leu	Uncertain significance	aHUS	1
CD46	c.1027+2T>C	-	Pathogenic	aHUS	1
CFH	c.79_82delAGAA	-	Likely pathogenic	aHUS	1
CFH	c.157C>T	p.Arg53Cys	Pathogenic	aHUS	2
CFH	c.158G>A	p.Arg53His	Likely pathogenic	aHUS	1
CFH	c.2697T>A	p.Tyr899*	Likely pathogenic	aHUS	2
CFH	c.2880delT	p.Phe960fs	Likely pathogenic	aHUS	1
CFH	c.2918G>A	p.Cys973Tyr	Uncertain significance	aHUS	1
CFH	c.3048C>A	p.Tyr1016*	Likely pathogenic	aHUS	2
CFH	c.3628C>T	p.Arg1210Cys	Pathogenic	aHUS	1
CFH	[c.3674A>T; 3675_3699del]	p.Tyr1225Tyrfs*38	Likely pathogenic	aHUS	2
CFH	c.3693_3696 delATAG	p.*1232Ilefs*38	Likely pathogenic	aHUS	5
CFI	c.341T>C	p.Val114Ala	Uncertain significance	aHUS	1
CFI	c.1357T>C	p.Cys453Arg	Uncertain significance	aHUS	1
CFI	c.1456T>C	p.Trp486Arg	Uncertain significance	aHUS	1
CFI	c.1642G>C	p.Glu548Gln	Uncertain significance	aHUS	2
DGKE	c.889–1G>A	p.IVS5–1	Likely pathogenic	aHUS	1
DGKE	c.966G>A	p.Trp322*	Likely pathogenic	aHUS	4
DGKE	c.1000C>T	p.Gln334*	Likely pathogenic	aHUS	1
DGKE	c.1608_1609del	p.His536Glnfs*16	Likely pathogenic	aHUS	1
PLG	c.1481C>T	p.Ala494Val	Likely benign	aHUS	1
C3	c.168_169delTG	p.Thr56Thrfs*16	Uncertain significance	C3G	1
C3	c.1682G>A	p.Gly561Asp	Uncertain significance	C3G	1
CFB	c.2035C>T	p.Arg679Trp	Uncertain significance	C3G	1
CFH	c.232A>G	p.Arg78Gly	Pathogenic	C3G	1
CFH	c.262C>A	p.Pro88Thr	Uncertain significance	C3G	1
CFH	c.694C>T	p.Arg232*	Likely pathogenic	C3G	1
CFH	c.3286T>A	p.Trp1096Arg	Uncertain significance	C3G	3

^*Denotes a nonsense variant

Gender analyses

Finally, we examined gender dependences in aHUS and C3G cases. Of the 1231 aHUS patients with at least one identified RV, 36% (440) were female, and 30% (363) were male (Table V). In aHUS, assuming the remaining 34% of patients for which the gender was unknown showed a similar pattern, the bias towards females was significant (p=0.007). However, when CFH was removed from the analyses, the numbers of females (244) and males (225) was not significantly different (p=0.38) (Table VI). For the 116 C3G patients, no significant difference in the number of females (53) and males (63) was seen (p=0.35) (Table VI). In terms of gender differences, our aHUS dataset (542 variants) showed a bias towards females in CFH, which is likely to be a reflection of potential triggering factors. For example, in pregnancy-aHUS, most patients harbour CFH variants (55, 56). This trend was not observed for C3G.

Table V.

Demographics of the 1231 aHUS and 116 C3G cases showing an identified rare variant

Disease	Gender	Number of cases aged 18+ years a	Number of cases aged <18 years b	YOB c unknown	Total
aHUS	Female	328 (27%)	84 (7%)	28 (2%)	440 (36%)
aHUS	Male	254 (20%)	84 (7%)	25 (2%)	363 (29%)
aHUS	Unknown	2 (<1%)	1 (<1%)	425 (34%)	428 (35%)
aHUS	ALL	584 (48%)	169 (14%)	478 (38%)	1231
C3G	Female	34 (29%)	12 (10%)	7 (6%)	53 (46%)
C3G	Male	40 (35%)	16 (14%)	7 (6%)	63 (54%)
C3G	Unknown	0	0	0	0
C3G	ALL	74 (64%)	28 (24%)	14 (12%)	116

^aBorn in or before 1997

^bBorn after 1997

^cYear of birth

RV pathogenicity classification

RVs were categorised for their pathogenicity using published experimental evidence, reference AFs and in silico predictive analyses, in accordance with genetic variant guidelines (1) (Figure 3A). The average healthy genome also contains benign RVs that occur at similar AFs to disease-related RVs, therefore AF analyses alone cannot be used as evidence for pathogenicity (43). In terms of pathogenicity, only gain-of-function RVs in the activators C3 and CFB are expected to predispose for aHUS or C3G disease; here gain-of-function results from the loss of ability to interact with an inhibitory complement regulator. The predictive tools PolyPhen-2 and SIFT were unable to predict gain-of-function phenotypes, thus many C3 and CFB RVs without experimental data could only be classified as ‘uncertain significance’ (green bars; Figure 3A). For example, the C3 RV p.Arg161Trp was predicted to be damaging and deleterious, while functional studies showed a C3 gain-of-function (57). The number of ‘pathogenic’ or ‘likely pathogenic’ RVs in C3 and CFB were therefore lower than for the other genes. In aHUS, the majority of RVs in CFH, CFI, CD46 and DGKE were either ‘pathogenic’ or ‘likely pathogenic’ (red and light blue bars; Figure 3A), this being attributed to their loss-of-function phenotypes. In C3G, the majority of RVs in all genes were of ‘uncertain significance’ (green bars; Figure 3A) except for CFH. In terms of the non-pathogenic RVs in the aHUS and C3G datasets, a survey of all 542 RVs in aHUS and 110 in C3G showed that 43 (8%) and 27 (25%) RVs respectively were categorised as ‘benign’ and ‘likely benign’, and these were filtered out (Supplementary Table I). This suggested that the RVs in the C3G dataset may be less pathogenic compared to aHUS overall; this observation reinforced the importance of using classification guidelines prior to our protein domain hotspot analyses below. We also note that ‘likely pathogenic’ RVs still require further functional studies to confirm or disprove the predicted effects and disease relevance of the variants.

Figure 3.

RV effects and classifications in aHUS and C3G. The colour coding of each RV effect and its classification is shown in the insets.

(A) In terms of pathogenicity, the total number of unique RVs for each gene and their classification, based on the pathology guidelines (Methods), are shown for the aHUS and C3G datasets, and for both datasets, in three panels.

(B) In terms of functional annotation (such as that used in ExAC), the total number of unique RVs for each gene and their effect on each protein are shown for the aHUS and C3G datasets, and for both datasets in three panels.

RV abundance in genes

In order to identify differences in the frequency of unique RVs per gene between aHUS and C3G, and therefore the molecular pathogenesis of both diseases, the abundances of RVs in each gene were analysed. We stress that only those RVs that were not classified as ‘benign’ or ‘likely benign’ were analysed (red, light blue, green and grey bars; Figure 3A). The CFHR1-CFHR4 genes were subject to multiple ligation-dependent probe assessment only, and were not sequenced, thus only LGRs were identified. In aHUS, CFH showed the most RVs (204 variants; 41%) (Figure 3A), followed by CFI (82; 17%), CD46 (79; 16%), and C3 (68; 14%). This outcome confirmed our 2014 analyses for CFH, CFI, CD46 and C3 (9). Lesser abundances for aHUS were seen for CFB (20; 4%), DGKE (20; 4%); THBD (11; 2%), PLG (4; 1%), CFHR5 (3; 1%), CFHR1 (4; 1%), CFHR3 (2; <1%) and CFP (2; <1%). Only one unique variant in CFHR4, a CFHR1/CFHR4 deletion, was seen in three heterozygous and one homozygous aHUS cases. This was classified under the CFHR1 gene in the database. In C3G, C3 (31 variants; 37%) and CFH (25 variants; 28%) showed the most RVs (red, light blue, green and grey bars; Figure 3A). Lesser abundances for C3G were seen for CFI (5; 6%), CD46 (1; 1%), CFB (8; 10%), DGKE (2; 2%); THBD (2; 2%), PLG (2; 2%), CFHR5 (1; 1%), CFHR1 (1; 1%), CFHR3 (1; 1%) and CFP (0; 0%). It was concluded that RVs in CFI and CD46 were substantially reduced in C3G compared to aHUS (p=0.0121 and p<0.0001, respectively; two-tailed Fisher’s exact test). When analysed in terms of genetic effect for both aHUS and C3G, most RVs were non-truncating (yellow; Figure 3B), except for the aHUS variants in DGKE that were mostly truncating, and the LGRs in CFHR1 and CFHR3. The most frequent RV in our aHUS dataset was C3 p.Arg161Trp, seen at an AF of 1.16% in 52 aHUS cases. However in C3G, none of the RVs were notably more frequent than others. Data on RVs found in the complement inhibitor vitronectin and clusterin genes in the aHUS and C3G datasets were not available for analysis at the time of writing (58, 59).

Gene-based RV Burden

In order to confirm that the amount of rare variation seen in the genes of aHUS and C3G patients was greater than in the genes of individuals without these diseases, we determined the burden of protein-altering rare variation (ExAC MAF<0.01%) per gene for each dataset (Methods). These were compared to the ExAC and EVS reference datasets (Figure 4; Supplementary Tables II and III). Because ExAC and EVS did not contain data on LGRs, CFHR1–4 were not analysed. No aHUS or C3G AF data were available for CFP. This left nine out of 13 genes (CFH, CFI, CD46, C3, DGKE, CFB, CFHR5, PLG, THBD) for analysis. For the aHUS dataset, a significantly greater burden of rare variation was revealed in patients than in the ExAC and EVS datasets for five genes, namely CFH, CFI, CD46, C3 and DGKE (Chi-square test with Yates’ correction using a Bonferroni-corrected significance level of 0.0056; the five genes each gave p<0.0001), and also for CFB when compared with the EVS dataset only (also with p<0.0001) (Figure 4; Supplementary Tables II and III). No association with aHUS was observed with RVs in THBD, PLG and CFHR5. The tests for these three genes showed a power >80% thus their false negative rates were expected to be very low. In the C3G dataset, C3 (p<0.0001; p<0.0001) and CFH (p<0.0001; p<0.0001) showed a significantly greater burden of protein-altering rare variation in patients than in the ExAC and EVS datasets (Chi-square test with Yates’ correction using a Bonferroni-corrected significance level of 0.0056) (Figure 4; Supplementary Tables II and III). C3G was also associated with RVs in CFB (p<0.0001) and THBD (p=0.0052) when only EVS was used as the reference dataset (green bar, Figure 4; Supplementary Table III), despite both tests showing a lack of power (CFB: 41% and THBD: 38%). The lack of association of DGKE and PLG with C3G may also relate to lack of power (DGKE: 55% ExAC and 68% EVS, and PLG: 32% ExAC and 10% EVS) shown in the tests for these genes.

Distribution of aHUS and C3G RVs in FH

The aHUS and C3G location and AF of the missense RVs for each gene resulted in the identification of mutational hotspots in each protein structure. For FH, as seen in the 2014 study, most aHUS missense RVs (78) occurred in the C-terminal ten domains, compared to the N-terminal ten domains (47; p=0.0042) (Figure 5A; Supplementary Table IV) (9). In aHUS, the total frequency of CFH alleles with a missense RV in the C-terminal ten domains (3.2%) was significantly greater than for the N-terminal ten domains (1.2%; Chi-square test with Yates’ correction using a significance level of 0.05; p<0.0001). In SCR-20, the total missense RV aHUS AF of 2.03% (Supplementary Table IV) was the highest for all 20 domains. These non-random distributions supported the functional association of SCR-20 with cell surface dysregulation in aHUS, although laboratory experimentation and functional characterization will be required to validate this result. This was still the case when normalised for the size of the FH domain (62/1231 residues; 5.5%), giving 37.2% (Figure 5A; Supplementary Table IV). Most of the six FH domains (SCR-2, SCR-5, SCR-8, SCR-12, and SCR-13) with the least number of aHUS RVs did not correspond to known FH binding sites. For C3G, in contrast to aHUS, the C3G missense RVs in FH were clustered at the N-terminal C3b binding site, with comparatively few at the non-surface associated C-terminal domains such as SCR-15 (Figure 5A). No C3G clusters were seen in the heparin binding regions of FH. The only three unique C3G missense RVs found in SCR-7 (p.Cys431Tyr; C3G AF: 0.2%) and SCR-20 (p.Arg1210Cys, 0.1%; p.Cys1218Arg, 0.1%) in the heparin binding regions were also seen in aHUS at similar AFs of 0.01%, 0.4%, and 0.01% respectively, suggesting an overlap in phenotypes. The C3G missense RVs in FH were also not found in eleven SCR domains (SCR-4/6, SCR-8/9, SCR-11/13, SCR-16/17 and SCR-19). There was a C3G missense RV cluster at the N-terminal SCR-1/3 C3b binding region (31) (Figure 5A; Supplementary Table IV); we infer that SCR-1/3 may be a mutational hotspot for C3G. In summary, the distribution of FH hotspots between aHUS and C3G showed clear differences between the two diseases.

Distribution of aHUS and C3G RVs in C3

For C3, the 67 aHUS C3 missense RVs occurred in 12 of the 16 C3 domains (Figure 5B; Supplementary Table IV). Macroglobulin (MG)-2 showed the highest missense RV AF for C3 (1.3%), normalised by the proportion of domain residues (21.7%), followed by MG-6b (0.5%; 18.5%) (Figure 5B). Both the MG-2 and MG-6b domains were thus inferred to be aHUS hotspots. The 29 C3G missense RVs occurred in 13 of the 16 C3 domains (Supplementary Table IV). In contrast to aHUS, these were spread more evenly throughout C3 and no hotspots were inferred (Figure 5B). No aHUS or C3G variants involved the key C3 thioester residues (Glu991, Cys998, His1104 and Glu1106), or the anaphylatoxin (ANA), aNT, and beta-sheet (CUB)-a domains. Again the distribution of C3 missense RVs between aHUS and C3G showed clear differences between the two diseases.

Distribution of aHUS and C3G RVs in FI, CD46, FB and other proteins

For FI, the 65 missense RVs in aHUS were distributed across all five domains of FI (Figure 5C). For MCP, the 45 aHUS rare missense variants were distributed across the four MCP domains (Figure 5D). For FB, there were no aHUS or C3G missense RVS in SCR-1 or SCR-3 (Figure 5E). In FB, most aHUS missense RVs occurred in the Von Willebrand Factor Type A (VWFA) domain, whereas the C3G missense RVs were spread across the SCR-2, VWFA and serine protease (SP) domains. No mutational hotspots were evident for either aHUS or C3G in FI, MCP or FB.

Missense RVs Mapped onto Protein Structures

The missense RVs for aHUS and C3G were mapped onto protein structural models for FH (48), C3 (49), FI (50), MCP (51) and FB (52). For FH, in confirmation of the above analyses, SCR-5 and SCR-12/13 were sparsely populated, implying that the variants were spatially clustered only at functional regions of FH (Figure 6A). For C3, MG-2 showed a higher density of aHUS variants (red, Figure 6B) while a higher density of C3G variants occurred at the core of C3 (yellow). For FI, the aHUS variants were distributed evenly throughout FI (Figure 6C). For CD46, the aHUS variants were likewise distributed throughout the protein (Figure 6D). Similar patterns were seen for FB (Figure 6E).

Figure 6.

Missense RVs mapped onto protein structural models. Only the aHUS and C3G RVs that were not classified as ‘benign’ or ‘likely benign’ are shown. For (A) FH, (B) C3, (C) FI, (D) MCP, and (E) FB, the red spheres represent missense RVs identified in the aHUS dataset, the yellow spheres represent missense RVs in C3G and the black spheres represent missense RVs in both aHUS and C3G. Note that some of the spheres overlap, especially if different RVs affect the same amino acid. The protein domains are shown in alternating colours (A) or in unique colours (B-E) and labelled in that colour for clarity.

Minor AF Analyses

The AFs of each of the protein-altering RVs per gene in the aHUS and C3G datasets were compared with the corresponding AF in ExAC. The two CFP RVs in aHUS were not analysed because there were no available patient AF data. CFHR1, CFHR3 and CFHR4 were not analysed because large genomic rearrangements were not included in ExAC at the time of writing. There were no variants that were rare in ExAC (MAF<0.01%) yet were significantly more common in ExAC, than in aHUS or C3G. The AFs and their significance are shown on the database for users.

Discussion

Summary of RVs in aHUS and C3G

Compared to our 2014 study (9) in which 324 aHUS- and C3G-associated genetic variants in CFI, CFH, C3 and CD46 were used to identify variant hotspots in these four proteins (www.fh-hus.org), this more detailed study now reports 610 RVs in 13 genes from 3128 aHUS and 443 C3G patients, together with their associated AFs. The expansion to 13 genes reflects new candidate genes for potential association with these diseases, and clarifies the extent to which they are indeed associated. To our knowledge, this is the largest study of aHUS and C3G patients to date. The much increased totals of RVs and cases now make possible for the first time AF analyses and RV burden analyses of the aHUS and C3G datasets, including comparisons to three reference datasets. In turn, the AF analyses provided more detailed statistical analyses of the RV distributions in both diseases. The analyses of distinct protein domain hotspots for aHUS and C3G clarify molecular differences that rationalise the occurrence of their two different phenotypes, the involvement of the RVs that are present, and the molecular mechanisms involved in both diseases. In particular, this study has reduced the earlier knowledge gap in the genetics and genotype-phenotype correlations of C3G to bring these closer to that of aHUS (1).

Differences between aHUS and C3G using AF Analyses

The AF analyses revealed three new insights into the individual RVs associated with aHUS and C3G (and not the full genes). Our analyses raised the question of what AF cut-off to employ. Firstly, our AF analyses verified the rarity of 97% of the aHUS and C3G variants when compared to the ExAC reference (Figure 1), especially given the ability of ExAC to resolve ultra-rare variant AFs as low as 1×10⁻⁵ (0.001%) (42). Such disease-predisposing alleles in aHUS and C3G are by definition deleterious. In theory, evolutionary pressures will maintain these alleles at very rare frequencies in the general population through negative selection (60–62). Other studies report a histogram of variants in which pathogenicity increases with rarity and low AF (43). These results therefore justify our focus on RVs with AF<1% in the reference datasets. A more stringent RV AF cut-off of 0.01% is applicable to rare diseases of Mendelian inheritance (43). This 0.01% cut-off was used for the RV burden calculations in order to restrict them to RVs with a higher confidence of pathogenicity (Figure 4; Supplementary Tables II, III). However, RVs observed in both aHUS and C3G have reduced penetrance, where disease-free individuals also harbour these pathogenic RVs, and the onset of aHUS or C3G disease depends upon a trigger and other factors. To enable our hotspot analyses, a less stringent reference AF cut-off of <0.1% was thus used, but accompanied alongside experimental data and prediction tools, as specified in genetic variant classification guidelines¹. Some variants show AFs of >1% (Table II). While these may be risk factors for aHUS (e.g. CFH p.Val62Ile) (12), their analysis was beyond the scope of this study. All the variants are available to view in the Database of Complement Gene Variants by adjusting the value of the ExAC AF filter.

The second insight involved AF differences between aHUS and C3G that relate to their different phenotypes. Significantly greater proportions of C3G RVs were identified in the reference datasets, with AFs between 0–1% rather than 0%, unlike aHUS (Figure 1). This meant that our C3G RVs occurred more frequently in individuals without C3G (the reference datasets) than our aHUS RVs. This result may explain the mostly chronic presentation of C3G that accumulates over time, in distinction to the mostly acute presentation of aHUS.

Thirdly, the AF analyses of the disease datasets (i.e. not the reference datasets) revealed differences between the most common RVs in the aHUS and C3G datasets that were likely to reflect their different phenotypes. In our aHUS dataset, while CFH had the most RVs, the most frequent RV was p.Arg161Trp in C3, corresponding to an AF of 1.16% of 3128 aHUS cases, i.e. 52 aHUS cases. C3 p.Arg161Trp has a surface exposed position in the MG2 domain (Figure 6B), and forms a hyperactive C3 convertase with an increased affinity for factor B, thus leading to over-activation of the AP. C3 p.Arg161Trp was not seen in the reference datasets, and classification guidelines confirmed its pathogenicity (1). In our C3G dataset, C3 had the most RVs. Further analyses (below) reveal distinct domain hotspots for aHUS and C3G.

RV Burden Testing

The RV burden is the proportion of screened cases with an identified protein-altering RV for which the ExAC AF was <0.01%. As opposed to the AF analyses that look at each variant one-by-one, the RV burden now provides insight into all the RVs associated with each gene. In general, RV burden tests assume that all tested RVs influence the phenotype in the same direction (63). We therefore separated RVs into ‘truncating’ (loss-of-function) and ‘non-truncating’ (either loss- or gain-of-function, or neutral) in order to aid interpretations. RV burden tests showed clear differences between aHUS and C3G when compared to reference datasets, and this clarified the molecular mechanisms of the two diseases. Previous knowledge of experimental functional characterization of some of the RVs collectively support that aHUS is more related to surface AP dysregulation, while C3G is more related to fluid phase AP dysregulation. Here we now extend these earlier functional results:

• (i)For aHUS, our RV burden analyses confirmed the association of rare variation in the six genes CFH, CFI, CD46, DGKE (18, 64), C3 and CFB. For the five genes of the AP, CFH, CFI, C3 and CFB are involved in both cell surface and fluid phase regulation, but CD46 is only involved in cell surface regulation. Therefore, the RV burden analyses suggest that aHUS involves defects that result in both cell surface and fluid phase dysregulation. Different domains of C3 and FH are involved in cell surface dysregulation compared to fluid phase dysregulation, and this is explored in terms of the aHUS and C3G RV distributions in the next section. The protein encoded by the sixth gene, DGKE, is found in endothelium, platelets and podocytes, and normally inactivates the signalling of arachidonic acid-containing diacylglycerols which activate protein kinase C (PKC) and promote thrombosis. Loss of DGKE function may thus result in a prothrombotic state and lead to microangiopathic hemolytic anemia seen in aHUS outside the complement system (18). For three more genes CFHR5, PLG and THBD, no association of RVs in these with aHUS was observed. Each of their tests showed a high power of >80%. This is unexpected from their known function, where FHR5 is likely to compete with FH for regulation (65, 66), while PLG and THBD are inhibitors of thrombosis, and THBD also regulates complement (67–70).
• (ii)For C3G in contrast, the RV burden analyses showed that the four genes C3, CFH, CFB and THBD were associated with C3G, while the two genes CD46 and CFI were not associated with C3G. This outcome suggested that C3G is not caused by defects in cell surface regulation by CD46 or defects in cell surface or fluid phase regulation by FI. Our results also suggested that non-LGR RVs in CFHR5 and PLG are not causative for C3G. We did not analyse LGRs in CFHR5 such as those identified in CFHR5 glomerulopathy. Despite pathogenic RVs in known cardiac genes not being overrepresented in ExAC (39), THBD is involved in cardiac disease cases. Furthermore, the prevalence of RVs may differ across different centres, especially for genes such as CFHR5 and PLG in aHUS, and CFHR5, PLG and DGKE in C3G, that are less mutated and/or were sequenced by one or two centres only. This outcome is potentially affected by sampling bias, thus being difficult to interpret, and the results for these genes should be considered as preliminary only. In addition, the lack of association of DGKE and PLG with C3G may also be related to a lack of power (10% – 68%) shown by their tests.

Hotspots for Missense RVs

The 2014 identification of hotspots in four complement proteins (9) can now be expanded to examine clusters of missense RVs in each of aHUS and C3G. Based on the RVs with 0.1% reference AF cut-offs, clear differences were seen between the phenotypes of aHUS and C3G at the molecular level. In particular, RV ‘hotspots’ were identified in FH and C3 that could be rationalised on the basis of their importance in protein-protein interactions. For example, FH is involved in both fluid phase and cell surface regulation by (i) being a cofactor for FI, (ii) possessing C3 convertase decay-accelerating activity, and (iii) blocking the formation of the C3 convertase. Despite the FH N-terminal SCR-1/4 and C-terminal SCR-19/20 domains both binding to C3b, required for both fluid phase and cell surface regulation, the C-terminal region of FH is critical for cell surface regulation (Figure 6A) and is not required for fluid phase regulation (71). Thus surface-exposed missense RVs in CFH that map to individual SCR domains can be correlated with FH function. Other variants predicted to affect FH stability may lead to FH aggregation, making this unable to perform fluid or cell surface phase regulation. In the fluid phase, FH is the only AP complement regulator that has cofactor activity for FI, however at the cell surface, both FH and MCP can act as the FI cofactor. Thus, if the FH C-terminal SCR-19/20 domains are compromised, wild-type MCP may save cell surface regulation. An additional scenario is that if the FH variant is heterozygous, this would affect only half of the FH in plasma, thus altering the resulting phenotype. Overall, the consequence of each RV on FH function can be complex to interpret.

Different FH SCR domains were identified as hotspots in aHUS and C3G:

• (i)For aHUS, the AF analyses confirmed that SCR-20 with 31 RVs and a RV density of 37.2% was a notable missense hotspot (Figures 5A, 6A; Supplementary Table IV). SCR-20 is functionally important for FH binding to C3b, C3d, heparin-like oligosaccharides and sialic acid (9, 72–74). The occurrence of SCR-20 as a RV hotspot is well explained by the disruption of FH binding to surfaces, leading to host cell damage from excess complement activation caused by unregulated C3b. aHUS missense RVs were also identified in the remaining 19 SCR domains in FH (Supplementary Table IV). Four of the five FH domains (SCR-2, SCR-5, SCR-8, SCR-12, and SCR-13) do not correspond to known FH binding sites and have only single missense RVs. The distribution in Figure 5A suggested that the aHUS missense RVs affect mainly FH cell surface binding.
• (ii)In contrast, for C3G, the missense RVs were clustered at the N-terminal C3b binding site (SCR-2/3) (Figure 5A). No missense RVs in C3G were now clustered at cell surface heparin binding sites in SCR-6/7 or SCR-20. The SCR-2/3 domains were identified as C3G hotspots, likely attributed to its binding to MG-2 and MG-6 in C3b (Figure 5A) (72, 75). This different clustering best correlates our C3G variants with dysregulation of the complement AP in the fluid phase (C3, FH) and not at the cell surface (CD46).

Different C3 domains were likewise identified as hotspots in aHUS and C3G:

• (i)For aHUS, the MG-2 and MG-6b domains with the highest RV density (Supplementary Table IV) were deduced to be RV hotspots (Figure 6B). Both MG domains interact with FH SCR-2 and SCR-3 to enable C3 regulation by FH in both the fluid phase and on the cell surface (75). The disruption of the MG-2 and MG-6b domains would reduce C3 regulation by FH. It is not clear if these two C3 domains also bind MCP thus affecting cell surface regulation further (76). While the TED domain contained 22 RVs, its RV density, which takes into account the number of residues in the domain (300), was not as high as might be expected from its functionally important thioester group and its binding to cell surfaces.

• (ii) In distinction, for C3G, too few missense RVs in C3 have been reported for a clear outcome. The RVs were distributed in 11 of its 13 domains with no clustering seen to date.

Utility of the Database

The new Database of Complement Gene Variants enhances our understanding of rare genetic variants in aHUS and C3G for clinical applications. Improvements include the use of AFs, predictive comparisons of wild-type and mutant amino acids, in silico analyses using PolyPhen-2 and SIFT, examination of evolution-conserved residues across species, and correlations with functional binding sites. These tools enable clinicians to assess RVs in disease, for example, to investigate which variants within these genes conferred predisposition to aHUS and C3G, and to identify mutational hotspots within these protein structure. This is especially useful for variants of uncertain significance for which no experimental data exists. Ethnicity data was only recorded for less than 50% of the aHUS and C3G patients in our datasets. The RVs are displayed on the database in comparison with ExAC ethnicity data with a full record of ethnicity. While the disease datasets are incomplete in this regard, the new web-database has the capacity to capture new ethnicity data for aHUS and C3G cases for future AF comparisons. Because the six renal clinics of this study are based in Western Europe and the US, which are also the source of much of the ExAC dataset, the effect of ethnicity are expected to be minimal on the aHUS and C3G analyses.

Abbreviations used in this article:

AF

allele frequency

aHUS

atypical haemolytic syndrome

ANA

anaphylatoxin

AP

alternative pathway

C3

C3 complement C3 gene, protein

C3G

C3 glomerulopathy

CD46

MCP membrane cofactor protein gene, protein

CFB

FB complement factor B gene, protein

CFH

FH complement factor H gene, protein

CFHR

FHR complement factor H-related gene, protein

CFI

FI complement factor I gene, protein

CFP

CFP complement factor properdin gene, protein

CNV

copy number variation

DGKE

DGKE diacylglycerol kinase epsilon gene, protein

EVS

Exome Variant Server

ExAC

Exome Aggregation Consortium

LGR

large genomic rearrangement

MAF

minor allele frequency

MG

macroglobulin

PLG

PLG plasminogen gene, protein

RV

rare variant

SCR

short complement repeat

SP

serine protease

THBD

THBD thrombomodulin gene, protein

1000GP

1000 Genomes project

VWFA

von Willebrand factor Type A