Main

Studying the genetic basis of congenital diseases in humans can reveal unique mechanisms orchestrating human organ development that are distinct from those in other species. The development of the human pancreas has unique characteristics compared to its counterpart in rodents, underscoring the importance of investigating this process in humans to elucidate species-specific regulation mechanisms. Previous human genetic studies of individuals with pancreatic agenesis, a rare congenital condition resulting from inappropriate pancreas development, identified evolutionarily conserved genes (GATA6 (ref. 1), GATA4 (ref. 2) and CNOT1 (ref. 3)) involved in the early stages of pancreatic development which have different dosage-dependent effects in mouse and human. Here, we report the identification of loss-of-function variants in the primate-specific gene ZNF808 as a genetic cause of defective pancreatic development in humans.

To identify genetic causes of defective pancreas development, we initially studied two unrelated individuals with isolated pancreatic agenesis (defined by neonatal diabetes, which is diagnosed before 6 months of age and exocrine pancreatic insufficiency1) in whom all known genetic causes had been excluded. We performed exome sequencing in the two affected individuals and their consanguineous parents and found that both individuals were homozygous for ZNF808 loss-of-function variants (Supplementary Tables 1 and 2).

We next investigated the presence of rare ZNF808 biallelic variants in 233 more patients with neonatal diabetes without a pathogenic variant in the known etiological genes4. Homozygous loss-of-function ZNF808 variants were identified in a further 11 unrelated individuals and 2 affected siblings (Fig. 1a,b and Supplementary Table 3). All frameshift and stop-gain variants affect residues in the last exon of the gene and are predicted to result in a truncated protein lacking between 4 and all 23 of the zinc finger domains. Two patients (probands 2 and 3) were homozygous for deletions predicted to result in no ZNF808 messenger RNA. All families showed cosegregation of the ZNF808 homozygous variants with the disease consistent with recessive inheritance (Fig. 1b). No deleterious homozygous loss-of-function ZNF808 variants were identified in >680,000 individuals without neonatal diabetes or pancreatic agenesis (UK BioBank (n = 454,756), Gnomad v.2.1.1 (n = 141,071), 100,000 genomes project from Genomics England (n = 75,118) and Genes and Health (n = 8,921)). Overall, our findings show that ZNF808 biallelic loss-of-function variants cause pancreatic agenesis/neonatal diabetes.

Fig. 1: Identification of homozygous variants in ZNF808 as a cause of pancreatic agenesis.
figure 1

a, Schematic representation of the ZNF808 gene and the pathogenic variants identified in 13 families. Deletions are highlighted in red above the gene cartoon while the loss-of-function variants are represented below. The KRAB and zinc finger domains are annotated at the gene level. b, Partial pedigrees of the 13 families with homozygous ZNF808 variants. NA, sample not available for testing; N/N, variant not detected; N/M, heterozygote for variant; M/M, homozygote for variant. Two black lines between parents indicate individuals who were known to be related. A black line and a dashed line between parents indicate individuals who were not known to be related at testing but were confirmed to be consanguineous by homozygosity mapping of next-generation sequencing data calculated with SavvyHomozygosity34.

Full size image

The patients’ phenotype supports a role for ZNF808 in early pancreatic development affecting both endocrine and exocrine pancreatic functions (pancreatic agenesis). All individuals showed markedly reduced insulin secretion in utero, as they all had low birth weight (median −2.98 s.d., interquartile range (IQR) −3.56 to −2.47) reflecting reduced insulin-mediated fetal growth and also postdelivery, as they rapidly developed insulin-requiring diabetes (median age at diagnosis 17 weeks, IQR 3–23). There was evidence of early exocrine failure, with 5 of 5 patients tested having undetectable fecal elastase and 4 further patients having symptoms or investigations consistent with fat malabsorption (Supplementary Table 3). The patients’ phenotype is restricted to pancreas as no extra-pancreatic features were consistently observed in the cohort.

Analysis of transcriptomic data from GTEx reveals that ZNF808 is broadly expressed across human adult tissues (Extended Data Fig. 1a). Transcriptomic analysis of 15 human embryonic tissues5 showed that ZNF808 is maximally expressed in developing pancreas (Extended Data Fig. 1b).

ZNF808 is a member of the KRAB zinc finger protein (KZFPs) family, the largest group of DNA binding factors in the human genome (∼350 protein-coding genes). KZFPs primarily act as epigenetic silencers of transposable elements through establishment of heterochromatin-associated H3K9me3 (refs. 6,7). The KZFP family is rapidly evolving with new members found at most phylogenetic branches since its emergence at the dawn of tetrapods8,9.

ZNF808 is exclusively found in primates and its evolutionary origin can be functionally traced through its zinc finger signature to a common ancestor of Old World monkeys (Fig. 2a), with no similar array of zinc fingers found in New World monkeys or any other mammals8. Our genome-wide analysis of sequence identity between primate and non-primate mammalian orthologues for all human protein-coding genes did not highlight any other primate-specific gene confirmed to be causative for a human developmental disorder (Extended Data Fig. 1c and Supplementary Table 4). These findings confirm how extremely rare it is for a primate-specific gene to cause a human congenital developmental disorder.

Fig. 2: ZNF808 is a primate-specific gene targeting transposons of similar evolutionary age.
figure 2

a, Reconstructed phylogeny of ZNF808 using a zinc finger signature approach. The three amino acids of each zinc finger directly contacting DNA were used to build a specific functional signature to track evolution of ZNF808, as previously described8. Zinc finger domains are color-coded according to the number of variants in each triplet compared with the human version. Notable events of loss or gain of zinc fingers are also represented. No appreciable homology with any zinc finger array was detected in New World monkeys or in any other mammals. Silhouettes of representative species are all from PhyloPic.org. b, ZNF808 binds primarily MER11 transposable elements. Analysis of ZNF808 ChIP–seq data8 reveals that it primarily intersects with transposable elements, although a few binding sites are found on gene promoters and other genomic regions. Further analysis of transposons shows that ZNF808 binds primarily elements of the MER11 family—MER11A, MER11B and MER11C. c, MER11 transposable elements are primate-specific. The origin of each individual MER11 element in the human genome was traced using a comparative multiple alignment of 241 species35. The age of each element was determined as corresponding to the farthest phylogenetic branch where we could find a similar copy at a syntenic locus. Data points are plotted as the sum of elements found to have originated at each phylogenetic branch per subfamily—the x axis is scaled according to time in million years from estimates between the human genome and each phylogenetic branch common ancestor. A curve is interpolated between the data points to show the estimated rate of replication between phylogenetic branches. The scale is relative to each subfamily and is indicative of proportional changes between phylogenetic categories—the highest point for each subfamily is annotated with the number of new elements for scale.

Source data

Full size image

Genome-wide binding profiling of ZNF808 shows that it primarily targets the long terminal repeat of endogenous retroviruses classified as MER11 elements which comprise subfamilies A, B and C (Fig. 2b). These elements also originated in Old World monkeys and spread during evolution in successive waves until recently (Fig. 2c). It was suggested that remnants of MER11 elements might be domesticated10,11,12,13, providing regulatory potential that can modulate gene expression in certain cellular contexts even though they have lost their ability to transpose. Using the chromatin immunoprecipitation ChIP-Atlas14 database, we found that MER11 elements show enriched occupancy for many DNA binding factors, including some involved in early embryonic development (Extended Data Fig. 2a,b). Three of these (GATA4, GATA6 and HNF4A) are known causes of pancreatic agenesis and/or diabetes in humans when mutated1,2,15,16 and are involved in endoderm, liver and pancreas specification. We also found that subsets of MER11 elements are targeted by other primate-specific KZFPs (Extended Data Fig. 2b). This suggests that MER11 elements have the potential to be regulatory elements and are silenced by a group of primate-specific KZFPs, with ZNF808 being the central effector.

To characterize the molecular events triggered by ZNF808 loss, we functionally inactivated ZNF808 in H1 human embryonic stem cells (hESC) using CRISPR-Cpf1 (hereafter, ZNF808 KO) (Fig. 3a and Extended Data Fig. 3). Also, we generated induced pluripotent stem cells (iPSCs) derived from proband 2 harboring a homozygous deletion of ZNF808 exons 4 and 5 to validate our findings.

Fig. 3: Loss of ZNF808 unmasks the regulatory potential of MER11 elements during pancreatic differentiation.
figure 3

a, ZNF808 gene editing using CRISPR. Cpf1 and two guide RNAs targeting the zinc finger array were used in H1 stem cells to produce the ZNF808 KO. b, In vitro differentiation protocol to pancreatic progenitors. Overview of the multistep differentiation protocol used in this study for both epigenetic and transcriptomic analysis of differences induced by ZNF808 KO. Stage numbers from S0 to S4 are used through the text to refer to specific steps. c, ZNF808 is important for the maintenance of epigenetic repression on MER11 elements. Left, H3K9me3 ChIP–seq peaks intersecting with MER11 elements reveal that most sites are covered in heterochromatin-associated H3K9me3 and are bound by ZNF808. A proportion of these peaks lose H3K9me3 in ZNF808 KO clones at various stages of differentiation (top). Results show that many MER11 elements that had H3K9me3 signal in the WT gain H3K27ac in the ZNF808 KO, especially at the early stages of differentiation (bottom). d, A subset of MER11 elements is activated in the ZNF808 KO heatmap showing clustering of 220 MER11 elements displaying a loss of H3K9me3 followed by a gain of H3K27ac in at least one stage of differentiation. Normalized ChIP–seq signal is shown—color scale ranges from +3 (red) to −3 (blue) on a z-score scale.

Source data

Full size image

We assayed epigenetic changes at key stages of in vitro differentiation from pluripotent stem cells to pancreatic progenitor cells (Fig. 3b)17. We first quantified the genome-wide presence of H3K9me3 (Fig. 3c, top), a hallmark of heterochromatin induced by KZFPs18 and focused our analysis on MER11 elements as they represented 91.3% of ZNF808-bound sites with H3K9me3 (Extended Data Fig. 4a–c). H3K9me3 was detected on 1,367 MER11 elements in wild-type (WT) stem cells, 1,041 of which are known to be targets of ZNF808. The ZNF808 KO cells show a loss of 693 H3K9me3-positive MER11 loci at S0 (embryonic stem cells), with 684 of these being known targets of ZNF808. The loss of H3K9me3 on MER11 elements known to be bound by ZNF808 is statistically significant when considering all stages (Fisher exact test P < 3.6 × 10−61). This implies that ZNF808 plays a key role in silencing MER11 elements in early development.

We then surveyed whether the loss of silencing revealed active regulatory potential at MER11 elements. In the ZNF808 KO, we observed emergence of 226 H3K27ac-positive MER11 elements at S0 (Fig. 3c, bottom), most (198) at known targets of ZNF808. The gain of H3K27ac on known MER11 elements targeted by ZNF808 is statistically significant when all stages are considered (Fisher exact test P < 1.8 × 10−16).

We identified a subset of 220 MER11 elements where both loss of H3K9me3 and gain of H3K27ac were observed at the same stage in the ZNF808 KO. These have distinct patterns of activity during differentiation which we could group in six clusters (Fig. 3d and Extended Data Fig. 5), reflective of when the elements gain H3K27ac. We found that MER11 elements gaining activity in the ZNF808 KO later during differentiation (cluster nos. 5 and 6) are more frequently bound by GATA4 or members of the HNF4 family (Extended Data Fig. 5c). These transcription factors are dynamically expressed during our differentiation protocol (Extended Data Fig. 6) and are known to be important in specification of endodermal derived tissues19,20 including pancreas21,22 and liver20,23,24. Analysis of chromatin accessibility from a single-cell ATAC-seq tissue atlas25 and the NIH Roadmap26 dataset shows that MER11 elements are not active in pancreas, however multiple MER11 loci are normally active in specific cell types, including fetal trophoblasts, enterocytes and hepatocytes (Extended Data Fig. 7a,d,e), with enrichment for those unmasked in the ZNF808 KO cells (Extended Data Fig. 7b,c). We found strong agreement between our ZNF808 KO and the patient-derived iPSCs in terms of loss of H3K9me3 and gain of H3K27ac (Extended Data Fig. 8a).

These results show that ZNF808 silences specific MER11 elements during differentiation toward pancreatic lineages and that loss of ZNF808 unmasks their regulatory potential which is normally found active in other cellular contexts.

We next quantified transcriptomic changes induced in the ZNF808 KO during differentiation. We found that the number of genes whose expression was dysregulated increased from 443 genes at S0 (stem cell state) to a peak of 2,124 genes at S3 (posterior foregut) (Fig. 4a). Comparable changes in gene expression were detected in the patient-derived iPSCs by both RNA-seq and quantitative PCR (qPCR) with reverse transcription (Extended Data Fig. 8b,c). We found that unmasked MER11 elements drive proximal gene activation in the ZNF808 KO predominantly at the first two stages of differentiation (S0, stem cell state; S1, definitive endoderm), with MER11 elements over-represented near activated genes at enhancer (10 kb–1 Mb) but not promoter distances (Fig. 4b and Extended Data Fig. 9a–c). We also found that the smaller group of MER11 elements that gain H3K27ac later in differentiation drive gene activation at S4 (pancreatic progenitor stage) (cluster no. 6, Fig. 3c and Extended Data Fig. 9c). These results show that the unmasking of MER11 elements in the ZNF808 KO impacts nearby gene expression in early differentiation with downstream effects as differentiation progresses.

Fig. 4: Loss of ZNF808 during pancreas differentiation leads to activation of genes in proximity to unmasked MER11 elements and induction of a liver gene expression program.
figure 4

a, Loss of ZNF808 leads to perturbed gene expression throughput pancreatic differentiation. Bar chart showing total genes activated and repressed with FDR < 0.05 and |FC| > 1.25 for each stage of pancreatic differentiation. K, thousands. b, Dysregulated genes are found in proximity to unmasked MER11 elements in the ZNF808 KO early in differentiation. Proximity enrichment (−log10 Fisher exact right-tail P value) showing an excess of dysregulated genes compared to all genes in proximity to MER11 elements losing H3K9me3 and gaining H3K27ac in the ZNF808 KO as a function of distance between genes and binding sites for genes activated (orange) and repressed (blue) on log10 scale. Enrichment peaks between 10 kb and 100 kb suggest ZNF808 repressing distal gene enhancers. A total of 43.4% and 23.9% of activated genes at S0 and S1, respectively, are within 1 MB of a MER11 element. c, Hepatic cords genes are activated and dorsal pancreas bud genes are repressed in ZNF808 KO. Top, Fisher exact enrichment between ZNF808 KO activated and repressed genes and genes more highly expressed in CS12–14 hepatic cords or dorsal pancreatic buds, respectively28. Bottom, log2 fold-change ZNF808 KO over WT for dorsal pancreas bud (left, blue) and hepatic cords (right, orange) genes. Each dot represents a single gene. d, Genes exclusively expressed in liver and activated in hepatic cords are activated in ZNF808 KO. The log2 fold-changes of the 29 genes that are expressed in CS12–14 hepatic cords, exclusively expressed in GTEx liver and activated at S2 in ZNF808 KO. Error bars give DESeq2 standard error of the log2 fold-change, n = 3 independent biological replicates. e, Immunostaining of AFP and PDX1 at S3 (posterior foregut) stage. Confirmation of RNA-seq results by immunostaining, showing activation of AFP in ZNF808 KO in PDX1 positive cells (representative of three independent differentiation experiments; scale bar, 100 μm). Independent replicate stainings from the same cell lines are given in Extended Data Fig. 3d.

Full size image

We performed gene set enrichment analysis using Enrichr27 and observed fetal and adult liver genes activated in ZNF808 KO from S2 (primitive gut tube) onwards (fetal liver enriched at stage S2, false discovery rate (FDR) < 10−14, odds ratio (OR) 12.3) (Extended Data Fig. 9d). We tested whether this reflected the divergence between pancreatic buds and hepatic cords in vivo, with comparison to Carnegie stage 12–14 human embryos that correspond to our S2–S4 in vitro stages28 (Fig. 4c, top). We found hepatic cord genes activated in our ZNF808 KO (FDR < 10−45, OR 6.4, stage S2) and conversely dorsal pancreas genes repressed (FDR < 10−12, OR 3.1, stage S3). The divergence between hepatic cord and dorsal pancreas genes peaks at posterior foregut stage S3 (Fig. 4c, bottom), coinciding with the introduction of factors modulating propancreatic signaling pathways (for example, retinoic acid and BMP and SHH inhibition); we suggest that this extinguishes the hepatic fate in the ZNF808 KO.

Using GTEx we found that the induction of hepatic gene expression in the ZNF808 KO was enriched in genes whose expression is exclusive to liver (Extended Data Fig. 10; P < 10−16, OR 7.6, stage S2). We identified a set of 29 liver-exclusive, hepatic cord genes activated at S2, including genes such as AFP, ALB, APOA1 and LIPC (Fig. 4d). This included an example of a gene (TDO2) activated in proximity to a MER11 element, suggesting that MER11 unmasking is compatible with hepatic gene expression (Extended Data Fig. 10c–e). We confirmed the transcriptomic upregulation of the early liver marker alpha-fetoprotein (AFP) by immunostaining (Fig. 4e and Extended Data Fig. 3d). Taken together, our data show that ZNF808 is essential to prevent a liver gene expression program from being aberrantly activated during pancreas differentiation, suggesting a potential mechanism for pancreatic agenesis.

We report the identification of recessive loss-of-function variants in ZNF808, a primate-specific gene, as a cause of pancreatic agenesis, a congenital developmental disorder. This confirms the crucial importance of gene discovery efforts in patients with extreme phenotypes.

Our study shows that ZNF808 is critical for human pancreatic development. Previous human genetic studies identified pancreatic developmental genes, including GATA6 (refs. 1,29,30), GATA4 (refs. 2,15,29,30,31) and HNF4A (refs. 16,32,33), with different dosage-dependent effects between mouse and humans, suggesting key differences in pancreatic development between the two species. Our identification of the role of ZNF808 and MER11 element regulation during human pancreatic development offers important insights into how mechanisms regulating pancreas development have diverged between primates and other mammals and supports a key role of ZNF808 in regulating differentiation of endoderm progenitors between liver and pancreas lineages. We provide correlative evidence that the unmasking of MER11 elements in the ZNF808 KO ultimately leads to the downstream induction of a liver gene expression program during pancreas differentiation. Future work combining targeted genome editing and 3D conformation data will be necessary to confirm which of the regulatory regions repressed by ZNF808 are responsible for disrupting pancreas development in our patients.

Our characterization of the role of ZNF808 during early development offers insights into the role of KZFPs, showing that they provide a negative layer of regulation that masks transposable element regulatory potential in cellular settings where the right transcription factors are present, yet the domesticated regulatory activity is undesired. We believe that this is an evolutionary mechanism that allows for a broader range of transposons to be domesticated as they do not have to prove beneficial in all cellular contexts if KZFPs can selectively silence them where they would have a negative impact.

The crucial role of ZNF808 and MER11 elements in human pancreas development underscores that even primate-specific genes and transposable elements can be involved in important aspects of human biology. This discovery offers important insights into the evolution of gene regulation during human development and opens new avenues of research in the fields of human genetics and diabetes.

Methods

Subjects

The study was conducted in accordance with the Declaration of Helsinki and all subjects or their parents/guardian gave informed written consent for genetic testing. DNA testing and storage in the Beta Cell Research Bank was approved by the Wales Research Ethics Committee 5 Bangor (REC 17/WA/0327, IRAS project ID 231760). For Proband 2, parents gave written consent for collection of a skin biopsy to be dedifferentiated to iPSCs to study the patient’s cause of neonatal diabetes. No participant received compensation for entering this study.

Individuals with neonatal diabetes diagnosed before the age of 6 months were recruited by their clinicians for molecular genetic analysis in the Exeter Genomics Laboratory. For one patient (case 10 in Supplementary Table 3), a ZNF808 homozygous loss-of-function variant was identified by exome sequencing analysis by the Center for Genomic Medicine, King Faisal Specialist Hospital and Research Center, Riyadh.

Proband 1 and 2 were selected from a larger cohort of individuals with pancreatic agenesis3. Within this group, they were the only two individuals without a genetic diagnosis who were born to consanguineous parents.

Genetic analysis

Exonic sequences were enriched from genomic DNA using Agilent’s SureSelect Human All Exon kit (v.4) and then sequenced on an Illumina HiSeq 2000 sequencer using 100 base pair (bp) paired-end reads. The sequencing data were analyzed using an approach based on the GATK best-practice guidelines. GATK v.3.7 HaplotypeCaller was used to identify variants that were annotated using Alamut batch v.1.8 (human reference genome assembly hg19) and variants that failed the QD2 VCF filter or had less than five reads supporting the variant allele were excluded. Copy number variants were called by SavvyCNV36, which uses read depth to judge copy number states. SavvyVcfHomozygosity34 was used to identify large (>3 Mb) homozygous regions in the exome sequencing data (https://github.com/rdemolgen/SavvySuite).

A total of 232 patients diagnosed with diabetes before age 6 months in whom the known genetic causes of neonatal diabetes had been excluded were analyzed either by using a targeted next-generation sequencing assay, which includes baits for known neonatal diabetes genes and additional candidate genes followed up from gene discovery, such as ZNF808, or by independent genome sequencing analysis. Variant confirmation and cosegregation in family members were performed by Sanger sequencing (primers available on request).

Cell culture and in vitro differentiation of human stem cells

The hESC (WA01/H1 line, Wicell) and patient-derived iPSC were cultured on Matrigel-coated plates (BD Biosciences) in Essential 8 (E8) medium (Life technologies, A1517001) and passaged using EDTA. To carry out the differentiation experiments, hPSC (human pluripotent stem cells) were dissociated using EDTA, then seeded on new Matrigel-coated plates in E8 medium supplemented with 10 µM rho-associated kinase inhibitor (ROCKi, catalog no. Y-27632; Selleckchem catalog no. S1049) at a density of 0.21 million cells per cm2. After 24 h, the differentiation was started by washing the cells with PBS, then changing the medium to D0 medium. The differentiation was carried out using our optimized protocol as previously described17.

Genome editing

To create an in vitro model for studying the role of ZNF808 in pancreatic development, guide RNAs targeting the zinc fingers domain of the fifth exon of ZNF808 for deletion were designed using Benchling (https://benchling.com) (gRNAs sequence available in Supplementary Table 7). The gRNAs with the highest quality score and lowest off-targets score were selected and purchased, alongside the RNP components (Alt-R Cas12a (Cpf1) Ultra protein, crRNA), from Integrated DNA Technologies (IDT) and used according to the manufacturer’s recommended protocol. A total of 2 million cells were electroporated with the RNP complex using Neon Transfection system (Thermo Fisher, 1100 V, 20 ms, two pulses) and plated on Matrigel-coated plates in E8 medium containing 10 µM ROCK inhibitor overnight. Afterwards, cells were single-cell sorted, expanded and screened for the desired deletion using PCR. Positive clones were validated by Sanger sequencing at Eurofins Genomics and the sequences were aligned using Geneious Prime 2020.1.1. The KO clones were characterized for pluripotency, chromosomal integrity and the top three off-target hits predicted by the online tool CRISPOR37 were checked with no off-target indels found.

Formaldehyde crosslinking

To fix the cells for ChIP–seq samples preparation, cells were incubated with TrypLE for 5–10 min at 37 °C and gently homogenized with the pipette, then pooled in a 15 ml Falcon tube containing warm DMEM. Afterwards, cells were spun down at 250g at room temperature for 3 min, then resuspended in DMEM at a concentration of 5 million cells per ml and incubated with 333 mM fresh 16% methanol-free formaldehyde at room temperature for precisely 10 min. Formaldehyde was quenched using 250 mM Tris pH 8.0 for another 10 min at room temperature, then cells were spun down at 250g at 4 °C for 5 min. Cell pellet was resuspended gently in PBS, aliquoted into 1.5 ml Eppendorf tubes and spun down at 250g at 4 °C for 5 min. Supernatant was removed and samples were stored at −80 °C until further processing.

Flow cytometry analysis

The hESC-derived cells were dissociated into single cells by incubation with TrypLE for 5–10 min at 37 °C and resuspended in cold FBS/PBS (5% v/v). For surface marker staining of CXCR4, 1 million cells were incubated with the directly conjugated antibody CD184/CXCR4 APC at a final dilution of 1:10 for 30 min at room temperature. For intracellular markers, 1 million cells were first fixed in 350 µl of Cytofix/Cytoprem Buffer (BD, no. 554722) for 20 min at 4 °C, then washed twice with BD Perm/Wash Buffer Solution (BD, no. 554723). Cell pellet was resuspended in 80 µl of FBS/BD Prem/Wash buffer (4% v/v) and incubated with the corresponding directly conjugated antibody at a final dilution of 1:80 overnight at 4 °C. After incubation with the antibody, cells were washed twice and analyzed using FACSCalibur cytometer (BD Bioscience), BD CellQuest Pro software v.4.0.2 and FlowJo software v.9 (Tree Star). Details of the antibodies are listed in Supplementary Table 6.

Immunocytochemistry

For adherent cultures, cells were fixed in 4% PFA for 15 min at room temperature, permeabilized with 0.5% Triton X100 in PBS, then blocked with UltraV block (Thermo Fisher) for 10 min and incubated with primary antibodies diluted in 0.1% Tween in PBS overnight at 4 °C. After incubation, cells were washed twice with PBS and incubated with corresponding secondary antibodies diluted in 0.1% Tween in PBS for 1 h at room temperature. Details of the antibodies are listed in Supplementary Table 6.

Chromatin immunoprecipitation

The following steps were performed with ice-cold samples and buffers containing a protease inhibitor cocktail (cOmplete ULTRA Tablets EDTA-free, Roche). In 1.5 ml DNA LoBind tubes, 4 million fixed cells were resuspended in 1 ml of lysis buffer 1 (50 mM HEPES-KOH pH 7.4, 140 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, 10% glycerol, 0.5% NP40, 0.25% Triton X100, proteinase inhibitor 1×) incubated at 4 °C on a rotating wheel at 10 r.p.m. for 10 min. Cells were then centrifuged at 1,700g for 5 min at 4 °C. Supernatant was discarded and pellets were resuspended in 1 ml of lysis buffer 2 (10 mM Tris HCl pH 8.0, 200 mM NaCl, 1 mM EDTA, 0.5 mM EGTA, proteinase inhibitor 1×) and incubated at 4 °C on a rotating wheel at 10 r.p.m. for 10 min. After centrifugation at 1,700g for 5 min at 4 °C, supernatant was discarded and pellets were washed with 500 µl of SDS shearing buffer (1 mM Tris HCl pH 8.0, 1 mM EDTA, 0.15% SDS, proteinase inhibitor 1×), without disturbing the pellets, followed by centrifugation at 1,700g for 5 min. Washing was repeated twice and pellets were resuspended in 1 ml of SDS shearing buffer and transferred into Covaris milliTUBE 1 ml AFA Fiber. Chromatin was sheared on a Covaris E220 for 6 min at 5% duty cycle, 140 W, 200 cycles. The sheared chromatin was then transferred into 1.5 ml DNA LoBind tubes and centrifuged at 10,000g for 5 min at 4 °C. Supernatant was then used immediately for immunoprecipitation. Chromatin quality control was performed on Bioanalyzer 2100 (Agilent) to verify that most fragments ranged between 200 and 600 bp.

For H3K9me3 IP, chromatin corresponding to 1 million cells was put in a new 1.5 ml DNA LoBind tube and topped to 900 µl with SDS shearing buffer. For H3K27ac IP, chromatin corresponding to 3 million cells was put in a new 1.5 ml DNA LoBind tube and topped to 900 µl total with SDS dilution buffer (1 mM Tris HCl pH 8.0, 1 mM EDTA, 0.026% SDS, proteinase inhibitor 1×) so that the SDS in the final IP buffer would be 0.1%. IP conditions were further adjusted to 150 mM NaCl and 1% Triton final, 1 ml final volume. Either 1 µl of H3K9me3 antibody (catalog no. 39685, Active Motif) or 5 µg of H3K27ac antibody (catalog no. 39161, Active Motif) was added. The IP was then incubated on rotating wheel at 10 r.p.m. at 4 °C overnight. The next day, Dynabeads Protein G (5 µl for the H3K9me3 IP or 25 µl for the H3K27ac) were put on magnet and supernatant was removed. Beads were then resuspended in the full volume of the IP and incubated for 2 h on a rotating wheel at 10 r.p.m. at 4 °C.

For H3K9me3, low-salt washing buffer (10 mM Tris HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 1% Triton X100, 0.15% SDS, 1 mM PMSF) and high-salt washing buffer (10 mM Tris HCl pH 8.0, 500 mM NaCl, 1 mM EDTA, 1% Triton X100, 0.15% SDS, 1 mM PMSF) were used.

For H3K27ac, low-salt washing buffer (20 mM Tris HCl pH 8, 150 mM NaCl, 2 mM EDTA, 1% Triton X100, 0.1% SDS, 1 mM PMSF) and high-salt washing buffer (20 mM Tris HCl pH 8, 500 mM NaCl, 2 mM EDTA, 1% Triton X100, 0.1% SDS, 1 mM PMSF) were used.

All washes took place while IPs and buffers were ice cold. PMSF was always added in the buffers immediately before each wash. The IPs were placed on a magnetic rack and supernatant was discarded. Beads were resuspended in low-salt washing buffer and transferred into a clean DNA LoBind tube. Beads were then placed on a magnetic rack; supernatant was removed and beads were resuspended in low-salt washing buffer. The mixture was placed again on a magnetic rack, supernatant was discarded and beads were washed with high-salt washing buffer. Once more, samples were placed on the magnetic rack, supernatant was removed and beads were resuspended in LiCl buffer (10 mM Tris HCl pH 8.0, 1 mM EDTA, 0.5 mM EGTA, 250 mM LiCl, 1%NP40, 1%NaDOC, 1 mM PMSF). The mixture was placed on the magnetic rack, supernatant was removed and beads were washed with 10 mM Tris HCl pH 8.0 and transferred to a clean DNA LoBind tube. Finally, with the samples on the magnetic rack the supernatant was completely removed and beads were resuspended in elution buffer (10 mM Tris HCl pH 8.0, 1 mM EDTA, 1% SDS and 150 mM NaCl).

RNase A was added to the elution buffer at a final concentration of 0.5 µg µl−1 and samples were incubated at 37 °C for 1 h in a shaking incubator at 1,100 r.p.m. Subsequently proteinase K was added at a concentration of 400 ng µl−1 and chromatin was decrosslinked at 65 °C overnight. The supernatant was collected and purified using Serapure beads before library preparation.

To control for the efficiency of the IP, we used qPCR with primers targeting negative and positive regions in the genome for the histone marks H3K9me3 and H3K27ac, respectively (primer sequences available in Supplementary Table 7).

Library preparation

Libraries were prepared using NEBNext Ultra II DNA Library Prep Kit for Illumina following the manufacturer’s instructions. Adapters and indexed primers design, resuspension and annealing were as previously described38. The adapters used were iTrusR2-stubRCp andiTrusR1-stub.

The library was quantified by qPCR using KAPA SYBR FAST and the set of primers itru7_101_01 and itru5_01_A. After quantification the library was amplified and double indexed (primers details in Supplementary Table 7). Amplified libraries were double size selected using home-made Serapure beads to enrich for fragments between 200 and 600 bp.

After amplification, confirmation that the IP remained efficient was carried out using qPCR with primers targeting genomic regions negative and positive regions for the histone marks H3K9me3 and H3K27ac, respectively, as mentioned above. Libraries were sent for 150 bp paired-end sequencing at Novogene.

ChIP–seq analysis

Reads from each library were mapped on hg19 using Bowtie2 v.2.4.4 (ref. 39) with the ‘very-sensitive-local’ setting. Mapped reads were compressed in BAM files and indexed using SAMtools v.1.18 (ref. 40). We used MACS2 v.2.2.7.1 (ref. 41) to call peaks using the corresponding input datasets as background controls. Peaks were excluded if the average read had MAPQ < 20. Reads intersecting intervals were counted with HTSeq 0.13.5, with a filter of MAPQ > 20 for properly paired reads applied. To identify MER11 elements losing H3K9me3 and gaining H3K27ac, we took peak regions defined by an overlap with H3K9me3 (pybedtools 0.8.1 and bedtools v.2.30.0, f = 0.5, F = 0.5; e, True; u, True; wa, True) in WT cells with no overlap in the KO and the opposite for H3K27ac to derive a final list of 220 MER11 elements activated in ZNF808 KO. To identify epigenetic responses over differentiation, we quantified depth normalized reads in each region and standardized H3K9me3 and H3K27ac signals separately by applying a z-score normalization (zero-meaned and standard deviation scaled) to each region over the differentiation stages. To assign regions to clusters of similar behavior, we performed k-means clustering with parameters n_clusters = 6, init = ‘random’, max_iter = 3,000, n_init = 100 using Scipy. To visualize these, we also performed hierarchical clustering using fastcluster v.1.2.4 with optimal leaf ordering which we mapped onto our k-means clustering to provide a heatmap with consistent within cluster and between cluster ordering. ChIP–seq tables and data analyzed in Python used pandas 1.3.5 and NumPy 1.12.5. Enrichment of transcription factors was by downloading hg19 ChIP peak data from the repository ChIP-Atlas (https://chip-atlas.org14). We used bedtools fisher v.2.30.0 with options –f 0.5 -F 0.5 -e to calculate intersection contingency table between MER11 elements of distinct subfamilies (MER11A, MER11B and MER11B) and each ChIP-Atlas dataset. We then calculated right-tail Fisher exact P values for over-representation and ORs using the Julia Distributions.jl package and applied Benjamini–Hochberg FDR control using MultipleTesting.jl.

Analysis of chromatin accessibility at MER11 elements was performed using the chromatin atlas25. Peak calls of chromatin accessibility for each cell type deconvoluted from single-cell signal clustering were downloaded and overlapped with bedtools with f = 0.5, F = 0.5; e, True (at least 50% overlap from either peak).

Analysis of the epigenetic status of MER11 in various cell types was done using the expanded NIH Roadmap analysis by ref. 26. Only datasets with measured H3K9me3 and at least one of H3K27ac or H3K4me1 were retained. Chromatin states were assigned by overlapping with precalculated HMM states from ref. 26 and were collapsed by cell type using the provided metadata—overlap in at least one cell type with a particular chromatin state was counted as present and Enhancer/TSS status were given precedence over heterochromatin if both were found in a cell type at a particular locus in two different biosamples.

Gene expression analysis by RNA-seq

Stranded, poly-A selected RNA-seq libraries were prepared and sequenced from three independent replicates of our differentiation time course (paired-end 150 bp reads) by Novogene.

Stranded paired-end 150 bp RNA-seq reads were aligned to the hg19 genome using STAR v.2.7.3a (ref. 42) and quantified against Gencode v.36 release liftover to hg19 by RSEM v.1.3.2 (ref. 43) using the RSEM-STAR pipeline, with further options --seed 1618 --calc-pme --calc-ci --estimate-rspd --paired-end. RSEM estimated read counts per sample were rounded for use with DESeq2 v.1.30.1 (ref. 44). We perform differential expression analysis for each stage WT versus ZNF808 KO for all genes with at least ten raw counts in all replicates of one condition. Gene expression varies substantially over the differentiation time course with subsets of genes only expressed at early or late stages, therefore when testing each stage, we supply DESeq2 with samples from adjacent stages for information sharing in estimating dispersions. We have three independent replicates of our differentiation time course and we perform a paired analysis between WT and KO pairs within DESeq2 with the model ∼ExpNum + Genotype, where ExpNum is a factor indicating the replicate and GenoType is a factor indicating WT or ZNF808 KO, we then perform a contrast between the two levels of the GenoType factor and calculate differential expression with independentFiltering=FALSE. We consider genes with FC (fold change) > 1.25 and FDR < 0.05 differentially expressed.

To determine enrichments of differentially expressed genes in proximity to clusters of epigenetic response in ZNF808 loss, we used the package ProximityEnrichment.jl (https://github.com/owensnick/ProximityEnrichment.jl). Specifically, we calculate hypergeometric right-tail P values for the association of differentially expressed genes within x bp of ZNF808-bound region cluster against a background of genes for x in [1, 1 × 106]. For proximity-enrichment heatmaps, we take the maximal enrichment from this interval.

To determine gene set enrichments for sets of differentially expressed genes, we use the Enrichr API27, with the package (https://github.com/owensnick/Enrichr.jl) to recover enrichments for BioPlanet 2019, GO_Biological_Process 2018, GO_Cellular_Component 2018, GO_Molecular_Function 2018, Human_Gene_Atlas and KEGG_2019_Human gene sets. As Enrichr calculates enrichments using a generic background, we downloaded all definitions of Terms and Gene sets and recalculated Fisher exact right-tail P values and ORs (using HypothesisTesting.jl) for terms over-represented in the dysregulated genes against a background of all genes tested for differential expression, we then corrected these for multiple testing using the Benjamini–Hochberg method (using MultipleTesting.jl). We report all enrichments in Supplementary Table 5, we selected the most prominent and relevant. To assess the intersection between our data and the laser capture of human embryo hepatic cords and dorsal pancreas28, we took the set of genes detected in each stage of our data and hepatic cords versus dorsal pancreas and calculated the association between direction of dysregulation in our data with the direction of differentially expressed genes in the hepatic cords versus dorsal pancreas comparison with FDR < 0.05.

To identity genes exclusively expressed in the adult liver, we downloaded v8 gene level transcripts per million over all GTEx samples from https://gtexportal.org (file GTEx_Analysis_2017-06-05_v8_RNASeQCv1.1.9_gene_tpm.gct.gz). We calculated the IQR for all genes in all tissues and took a pragmatic definition of liver-exclusive expression: we took those genes for which the lower quartile of liver expression exceeded the upper quartile of expression in all other tissues, yielding 357 genes (Extended Data Fig. 10a). The genotype-tissue expression (GTEx) project was supported by the of the Office of the Director of the National Institutes of Health (NIH) and by NCI, NHGRI, NHLBI, NIDA, NIMH and NINDS. The data used for the analyses described in this manuscript were obtained from the GTEx Portal on 6 January 2023.

Data and code to perform transcriptomic analysis and to generate figure panels are available from https://github.com/owensnick/ZNF808Genomics.jl.

Statistics and reproducibility

No statistical method was used to predetermine sample size for the patient cohort as the aim of the study was to identify the genetic cause of pancreatic agenesis, which is a rare monogenic condition.

Three independent replicates of H1 embryonic stem cell-derived WT and ZNF808 KO differentiation time courses were collected. For transcriptomic analysis we used n = 3 replicates for WT and ZNF808 KO and n = 1 for the patient-derived iPSC. From each of WT, ZNF808 KO and patient-derived iPSC, one replicate per stage was assayed for H3K9me3 and H3K27ac epigenomics.

No data were excluded from any of the experiments described.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.