Three non-ADPKD cortical kidney tissue samples were provided from the Christchurch Cancer Society Tissue Bank, as previously described (Bowden et al., 2018). These samples were collected from the renal cortex during nephrectomy for renal cell carcinoma, with histological testing to confirm the absence of malignant tissue. Eight single cyst samples were macrodissected from a single ADPKD kidney (male, 60 years old, PKD1 mutation c.3744_3745dup (het), NM_001009944.2). Following nephrectomy, the ADPKD kidney was stored in neutral buffered formalin for 7 days before we were able to collect it. The cyst samples were macrodissected from various sites throughout the kidney by cutting sections of the cyst wall, approximately 1 cm² in size, with surgical scissors (Figure 1A). An effort was made to ensure these samples were from a single cyst; this means samples were selected from the largest intact cysts available. Tissue was washed with 70% EtOH to remove non-cyst cells (i.e., blood). The sections of cyst were then submerged in fresh 70% EtOH and stored at 4°C prior to DNA isolation. The collection and use of renal tissue was approved by the Otago University Human Ethics Committee H15/110.

DNA methylation was assessed using RRBS. RRBS library generation of the non-ADPKD renal tissue has previously been described [data accessible on the NCBI GEO database, accession GSE110057 (Bowden et al., 2018)] and is based on previously published methodologies (Chatterjee et al., 2012a, 2013, 2017a). The generation of the cyst libraries followed the same protocol as the whole renal tissue, with modifications made for the limited tissue availability and formalin fixation. Briefly, genomic DNA was extracted from the cyst samples using the QIAamp DNA FFPE Mini Kit (Qiagen), excluding the paraffin melting steps, with an extended 16-h Proteinase K treatment at 56°C. An overnight MspI digest was performed, followed by end-repair, A-tailing and ligation of methylated adapters to the resulting fragments (Illumina). Bisulfite conversion was performed using the MethylCode DNA Methylation Kit (Invitrogen).

Libraries were amplified with PCR (15–20 cycles) and run through 3% NuSieve GTG Agarose gels (Lonza) to select for fragments 150–330 bp post-adapter ligation. Libraries were quantified with the Qubit fluorometer (Life Technologies) and assessed for quality on the 2100 Bioanalyzer using the High Sensitivity DNA chip (Agilent Technologies) according to manufacturer’s instructions. Libraries with high amounts of adapter contamination (identified as peaks just under 150 bp on the Bioanalyzer trace) were re-size selected with AMPure XP Bead Purification (New England Biolabs) before repeating Qubit and Bioanalyzer steps. RRBS libraries were sequenced in an Illumina HiSeq 2500 to produce single-ended 100 bp reads. Base calling was performed with Illumina Real Time Analyzer software.

Quality of the sequenced reads was assessed using the FastQC¹ program. Adapter sequences and methylated CpG bases at the 3′ end of the reads that were added during the end-repair step of the library preparation were removed using our in-house cleanadaptors program as part of the DMAP package (Chatterjee et al., 2012b). The sequenced reads were aligned to the human genome GRCh37 with Bismark v0.6.4 alignment tools (Krueger and Andrews, 2011). We performed stringent mapping and allowed for only one mismatch in the seed (the first 28 bp of the read).

The bioinformatic software program DMAP (Stockwell et al., 2014) was used to generate MspI fragment-based methylomes and perform analysis on the sequenced reads of each sample. A non-ADPKD reference library was made by merging the three non-ADPKD tissue libraries with the program SamTools (Li et al., 2009). Fragments were only included in analyses if they satisfied our coverage criteria (contained at least two CpGs and were covered by 10 or more sequenced reads). We have demonstrated the utility of this analysis approach previously (Chatterjee et al., 2017b).

Direct comparison between the previous whole tissue work and the cyst libraries was performed by generating a list of fragments common in all 10 RRBS libraries (non-ADPKD renal tissue, ADPKD whole tissue, eight ADPKD cysts), in order to prevent any bias due to sequencing coverage. Previously identified differentially methylated fragments (DMFs) were assessed in the eight ADPKD cysts by performing an F-test analysis of variance (ANOVA) which allowed us to assess the significance in difference between methylation in the two groups at these specific fragments. Clustering analysis was performed using unsupervised hierarchical clustering in R, using the “complete” method.

Cyst-specific DMFs were identified in each cyst by performing Fisher’s exact test between each cyst library and the non-ADPKD cortical tissue reference library we generated. DMFs were selected using the Bonferroni correction for multiple testing (p = α/n; α = 0.001, cut off values in Supplementary Table S1), and an absolute difference in methylation greater than or equal to 0.25 (i.e., 25%) between ADPKD and non-ADPKD tissue. Allosomes were excluded from the analyses between multiple individuals due to sex differences between the two groups. This is different to the ANOVA test above as this is identifying changes specific to the single cyst rather than within between two groups of whole tissue which contain multiple cysts.

To identify inter-cyst variation, a Chi-square test was performed with n – 2 degrees of freedom (where n is the number of individual cysts investigated). This resulted in the analysis of 45,954 fragments (referred to as “total analyzed fragments”). To control for multiple testing, the Bonferroni correction was applied with a stringent significance value of 0.001; the adjusted cut off p-value was 2.18 × 10^–8. Fragments with p-values below this threshold were deemed inter-cyst variants (ICVs) (Figure 1B). An overlap between ICVs and DMFs common in ≥50% of the cysts identified variably DMFs. The cut off was 50% of cysts to account for some of the variably methylated fragments having the same methylation value as the non-ADPKD library.

The identgeneloc program of the DMAP package was used to label fragments based on the distance to the transcription start site (TSS) of a gene. Intragenic fragments were subclassified as promoters (between 5 kb upstream and 1 kb downstream of the TSS), and those within the gene body (>1 kb downstream of the TSS) were labeled as intron, exon, and intron/exon boundary fragments by identgeneloc. All fragments not within this defined intragenic region were classed as intergenic fragments. Identgeneloc also allows for fragments to be classified based on their CpG feature as annotated by SeqMonk²; that is, whether they are in the core, shore or shelf of a CpG island (CGI).

Gene ontology was performed using Metascape (Zhou et al., 2019). Gene lists (Supplementary Tables S3–S7) were analyzed alongside a list of background genes (n = 17,917), which was generated from a list of all sequenced fragments in the nine RRBS libraries. A list of genes (n = 2,415) associated with intragenic ICVs (n = 4,204) was analyzed to characterize pathways associated with ICVs, as intragenic fragments were more likely to have consequential effects on gene transcription.

Genes were ranked with a variation score, to indicate the susceptibility of the gene to DNA methylation variation. This was calculated by dividing the number of intragenic ICVs associated with the gene, divided by the total number of intragenic fragments associated with the gene in the analysis. Genes were required to contain at least three ICVs in order to be included in the analysis as this conferred a reasonable amount of certainty about the amount of variation in the gene. Fragments with a variation score of 1.0 had total variation in all sequenced fragments in the analysis.

To identify inter-cyst variation, a Chi-square test was performed with n – 2 degrees of freedom (where n is the number of cysts investigated). To control for multiple testing, the Bonferroni correction with a stringent significance value of 0.001 was applied to each analysis to determine the p-value threshold. The statistical significance for other analyses (such as the proportion of ICVs in the chromosome and genomic features) were calculated with a Chi-square test with Yate’s correction, where p < 0.001 was the minimum value for significance. To plot the distribution of p-values, QQ-plots were generated using R (Supplementary Figure S6).

Cysts were visually confirmed to be throughout the kidney upon dissection. As the kidney had been cut up for formalin fixation many had drained, but some still contained cyst fluid. The larger of the cysts were collected in order to extract as much DNA for analysis as possible. These had thicker membranes and were more distinct from the stroma. As the kidney was partially dissected when we had access to it, we could not determine the spatial organization of the cysts in the kidney.

We carried out genome-wide bisulfite sequencing on eight single kidney cyst samples from an ADPKD patient using RRBS. We used this data alongside previously described whole renal tissue RRBS data from non-ADPKD and ADPKD kidneys (Bowden et al., 2018). We first wanted to compare global methylation data from the eight ADPKD cysts to ADPKD whole tissue samples and non-ADPKD samples which we had previously generated. The non-ADPKD samples clearly clustered separately from the cysts and ADPKD tissue samples (Figure 1C). Next, we identified the global change in median methylation in each cyst in comparison to non-ADPKD and ADPKD whole tissue samples using only the common analyzed fragments, where it was observed that all eight ADPKD cysts were hypomethylated by 3–4% in comparison to our non-ADPKD reference genome (Figure 1D and Supplementary Table S2). This is consistent with our whole tissue data which showed a 2% reduction in median methylation in ADPKD compared to non-ADPKD. When compared to non-ADPKD using the same common analyzed fragments, the whole tissue ADPKD samples were only hypomethylated by 1.6%.

We then sought to validate the previously identified DMFs in each cyst to confirm that these changes are universal in each ADPKD cyst. There was coverage in ≥50% of the cysts within nine of the whole tissue DMFs, with statistically significant differences between non-ADPKD tissue and ADPKD cysts in six DMFs (Figure 1E).

Differential methylation has largely been the focus of DNA methylation analyses, since such changes in the promoters of genes may confer stable changes in gene expression (Jones, 2012; Chatterjee et al., 2018). For this reason, we decided to identify DMFs within each ADPKD cyst to determine whether key changes occurred within each individual cyst. An analysis was performed between each ADPKD cyst and the non-ADPKD methylome using Fisher’s exact test. The number of DMFs identified in cysts (Supplementary Table S1) ranged from 474 (in Cyst 4) to 6,087 (in Cyst 1). Of these fragments, 96 common DMFs, which all demonstrated the same direction of methylation change, were identified in all eight cysts. The common DMFs were slightly enriched within the gene body of protein-coding genes (54%; Figure 2A) and were enriched for pathways associated with animal organ maturation as well as cell-cell adhesion and Notch signaling (Figure 2B and Supplementary Table S9).

Of our previously identified whole tissue DMFs (Bowden et al., 2018), the TMEM18- and GET4-associated fragments were also differentially methylated in some of the single cysts according to the stricter criteria (Supplementary Figure S2).

As illustrated by comparison to our DMF data, there was a significant amount of variability across the eight cysts. To quantify this further, we selected fragments present in at least six of the eight cysts (to account for lower methylome coverage in some cysts), which resulted in 45,954 total analyzed fragments. Fragments displaying significant inter-cyst variation were identified by applying a Chi-square goodness of fit test, followed by the Bonferroni adjustment for multiple test correction at a significance level of 0.001 (adjusted p-value cut off = 2.18 × 10^–8). This analysis identified 6,727 fragments (14.6% of the total analyzed fragments) which were termed “inter-cyst variants” (ICVs).

Next we characterized the ICVs based on their chromosomal and genomic location. We found that the proportion of ICVs in two chromosomes (4 and 17) was significantly different to the proportion of total analyzed fragments by ≥12% (p = 0.001, Figure 3A). When ICVs were characterized by relation to CGIs, ICVs were shown to be more strongly enriched in fragments within the core of the CGI (Figure 3B). The proportion of ICVs increased in intragenic fragments and conversely decreased in intergenic fragments (Figure 3C).

Genes were ranked with a variation score, which was generated based on how many fragments associated with each gene were variably methylated, in order to determine the genes with the most variable methylation states. A total of 18 genes (Figure 3D and Supplementary Figure S3) had a variation score of 1.0 (at least three fragments associated with the gene, all classed as ICVs), and were functionally enriched for the gene ontology processes sulfur compound metabolic processes and GTPase regulation (Supplementary Figure S4 and Supplementary Table S8).

When the null hypotheses are true, the distribution of p-values is uniform on the interval (0, 1). The histogram of actual p-values (Supplementary Figure S5) shows a pronounced skew toward values near zero. The QQ-plot (Supplementary Figure S6) displays the same information: observed quantiles lie below the quantiles of the uniform distribution (indicated by the 45-degree line).

In this analysis there were 4,204 intragenic ICV fragments, associated with 2,415 unique genes. We investigated the functional pathways of this group of genes, as methylation at these fragments is historically considered more likely to impact the gene function than methylation within intergenic fragments (Klose and Bird, 2006).

Gene ontology analysis through Metascape revealed the intragenic ICV-associated genes were enriched for a number of developmental and morphogenesis pathways (Supplementary Figure S7 and Supplementary Table S10; intergenic ICV-associated gene ontology data in Supplementary Table S11).

As there were differences in the number of ICVs identified in each cyst, we aimed to investigate if there was an overlap between DMFs and ICVs, i.e., were fragments which showed variable methylation measurably different to the non-ADPKD tissue.

A total of 2,204 fragments were differentially methylated in at least 50% of the cysts. 837 of these DMFs shared an overlap with the 6,727 ICVs (Figure 4A), occurring in 586 unique genes. These fragments which overlapped with DMFs and ICVs could be said to be variable DMFs. Variable DMFs were predominantly located within the gene body of protein-coding genes (Figure 4B), which were functionally enriched for development pathways (Figure 4C).

We previously identified 13 DMFs in ADPKD tissue and have now identified TMEM18- and GET4- associated differential methylation also occurs individually in single cysts. These DMFs did not demonstrate variable methylation.