Introduction

Alveolar capillary dysplasia with misalignment of pulmonary veins (ACDMPV; MIM 265380) is a lethal neonatal lung disorder caused by abnormalities in air–blood barrier structure and function (Langston 1991; Bishop et al. 2011). Affected newborns typically present with severe respiratory failure and refractory pulmonary hypertension within a few hours after birth and die in the first month of life. Histopathologically, ACDMPV is characterized by decrease in number of capillaries adjacent to the alveolar epithelium, alveolar wall thickening, hypertrophy of the muscular layer of small pulmonary arteries with abnormal muscular extensions into intra-acinar vessels, and malposition of the small pulmonary veins. In addition, the majority of patients with ACDMPV manifest extra-pulmonary anomalies of the cardiovascular, gastrointestinal, and genitourinary systems (Sen et al. 2004; Bishop et al. 2011; Galambos et al. 2015).

Heterozygous point mutations in FOXF1 (OMIM 601089) and genomic deletion copy-number variants (CNVs) at chromosomal region 16q24.1 including FOXF1 or its upstream regulatory region have been identified in the vast majority of patients with ACDMPV (Stankiewicz et al. 2009; Sen et al. 2013a, b; Szafranski et al. 2013a, b, 2014). FOXF1, expressed in lung mesenchyme and vascular endothelium, belongs to the forkhead family of transcription factors, and is a target of sonic hedgehog (SHH) signaling from lung epithelium.

Homozygous Foxf1 −/− mice die by embryonic day 8.5 because of defects in the development of extraembryonic and lateral mesoderm-derived tissues (Mahlapuu et al. 2001), whereas heterozygous Foxf1 +/− mice exhibit features resembling ACDMPV (Kalinichenko et al. 2001).

The FOXF1 promoter overlaps a CpG island, does not contain a TATA-box, and requires enhancer function for its activity (Chang and Ho 2001; Kim et al. 2005; Szafranski et al. 2013a, b). We have shown that the lung-specific enhancer region, mapping ~270 kb upstream of FOXF1, harbors genes for long noncoding RNAs (lncRNAs) that regulate FOXF1 expression (Szafranski et al. 2013a, 2014). Further, we have also identified another FOXF1 enhancer located within the FOXF1 intron (Szafranski et al. 2013b).

Interestingly, all 17 reported pathogenic genomic deletions involving FOXF1 or its upstream regulatory regions, for which parental origin was determined, arose de novo on the maternal chromosome 16, suggesting that the FOXF1 locus is imprinted (Stankiewicz et al. 2009; Sen et al. 2013b; Szafranski et al. 2013a, b; Dharmadhikari et al. 2015). Segregation analysis of a missense mutation in FOXF1 (c.416G>T; p. Arg139Leu) in a familial case of ACDMPV provided additional support for imprinting of FOXF1 in humans (Sen et al. 2013a). Furthermore, previous bioinformatics studies, aimed at identification of imprinted genes, indicated that the FOXF1 locus may be imprinted (Luedi et al. 2007).

We present 13 novel ACDMPV-causing de novo deletion CNVs in 16q24.1, 12 of which arose on maternal chromosome 16, providing statistically significant support for imprinting of the FOXF1 locus. Surprisingly, we also identified one small pathogenic genomic deletion in the upstream regulatory region and three causative variants involving FOXF1 that all arose on paternal chromosome 16, highlighting the complexity of genomic and epigenetic regulation of FOXF1 expression that underlies the etiology of ACDMPV. Lastly, we describe the results of whole exome sequencing (WES) in three unrelated ACDMPV families negative for both FOXF1 point mutations and deletion CNVs. Our results demonstrate complexity of genomic and epigenetic regulation of the FOXF1 gene in 16q24.1 and implicate the role of other genes in ACDMPV.

Methods

Subject recruitment

After informed consent, using protocols approved by the Institutional Review Board for Human Subject Research at Baylor College of Medicine (BCM) (H8712), histopathological specimens and DNA samples (peripheral blood or lung) from probands with ACDMPV: 20 postnatal [pts 114.3, 115.3, 117.3, 119.3, 120.3 (Decipher 285653), 121.3, 122.3, 123.3, 124.3, 125.3, 126.3, 127.3, 128.3, 130.3, 133.3, 134.3, 136.3, 138.5, 139.3, and 141.3] and two prenatal (135.3 and 140.3) and their family members (blood) were obtained.

Histopathological studies

Histopathological evaluations of all cases suspected of ACDMPV were performed in formalin-fixed paraffin-embedded (FFPE) specimens from lung biopsies or autopsies stained with hematoxylin and eosin.

Molecular biology

DNA isolation and sequencing, RNA isolation, RT-qPCR, cloning of the 16q24.1 region deleted in patient 122.3 and of the FOXF1 promoter, and transcriptional activity assay were performed as described in Supplemental Methods.

Array CGH analyses

Genomic CNVs in the FOXF1 locus were analyzed using customized 16q24.1 region-specific (1 Mb region flanking FOXF1) high-resolution 4 × 180 K microarrays, manufactured by Agilent Technologies (Santa Clara, CA, USA), as previously described (Szafranski et al. 2013a) [pts 114.3, 115.3, 117.3, 119.3, 120.3 (Decipher 285653), 121.3, 122.3, 125.3, 126.3, 127.3, 128.3, 133.3, 135.3, 136.3, 139.3, and 140.3].

Characterization of deletion breakpoints

Deletion junction fragments were amplified using long-range PCR with LA Taq DNA polymerase (TaKaRa Bio, Madison, WI, USA) and primers designed by Primer3 software (http://frodo.wi.mit.edu/primer3) as described (Szafranski et al. 2013a, b). The sequence of a complex genomic rearrangement in exon 1 of FOXF1 has been deposited in the GenBank database (http://ncbi.nlm.nih.gov/genbank) under accession number KT963011.

Parental origin of deletions and point mutations

Parental origin of the deletions was determined using informative microsatellites or single nucleotide polymorphisms (SNPs) mapping to the deleted genomic interval. For point mutations, PCR product containing the pathogenic variants and the neighboring SNPs were cloned into pGEM-T vector (Promega, Madison, WI, USA) and 10 clones were used for plasmid isolation and sequencing.

Bioinformatic analysis of the distant upstream FOXF1 enhancer region

Reference DNA sequences, chromatin modification, location of CpG islands, and ChIP-seq data for the selected transcription regulators were accessed using the UCSC Genome Browser (http://genome.ucsc.edu, GRCh37/hg19). High-throughput chromosome conformation capture (Hi-C) analyses were performed as described in Supplemental Methods.

Whole exome sequencing

Three family trios with sporadic ACDMPV, negative for FOXF1 mutations and deletions (using Sanger sequencing and custom-designed region-specific high-resolution array CGH) (pts 114.3, 121.3, and 128.3), were analyzed using WES as described in Supplemental Methods.

Results

Clinical characterization

Histopathological examination of lung specimens revealed the characteristic pathognomonic features of ACDMPV in all cases analyzed. Identified anomalies involving other organs are listed in Table 1 and Supplemental Table S1.

Table 1 Additional non-lung clinical features in 37 ACDMPV patients with the reported and novel pathogenic deletions involving the FOXF1 locus in chromosome 16q24.1
Full size table

Genomic deletions

We identified and characterized novel different-sized genomic deletions at 16q24.1 in 13 unrelated patients with ACDMPV [pts 115.3, 117.3, 119.3, 120.3 (Decipher 285653), 122.3, 125.3, 126.3, 127.3, 133.3, 135.3, 136.3, 139.3, and 140.3] (Fig. 1; Supplemental Table S1). Twelve out of 13 deletions (pts 115.3, 117.3, 119.3, 120.3, 125.3, 126.3, 127.3, 133.3, 135.3, 136.3, 139.3 and 140.3), for which the parental origin could have been determined, arose de novo on the maternal chromosome 16 (Supplemental Table S1). In one of those cases (pt 115.3), the CNV deletion involved only FOXF1, in five cases (pts 118.3, 120.3, 125.3, 133.3, 135.3, and 140.3), deletions encompassed FOXF1 and its upstream enhancer region, and in seven cases (pts 117.3, 119.3, 122.3, 126.3, 127.3, 136.3, and 139.3) the deletion CNVs harbored only the upstream enhancer, leaving FOXF1 intact. We did not find any evidence of somatic mosaicism in the parental DNA samples from peripheral blood using PCR with primers specific for the patients’ deletion junction fragment. Recently, we showed that new mutations that occur on the maternal allele are more likely to be recurrently transmitted to offspring (Campbell et al. 2014). Given that all published and all but one described here new deletions in the 16q24.1 FOXF1 region arose de novo on the maternal chromosome 16, the recurrence risk for ACDMPV may potentially be elevated in comparison to that observed for other sporadic diseases.

Fig. 1
figure 1

A compilation of chromosome 16q24.1 deletions pathogenic for ACDMPV. Deletions which occurred on maternal chromosome 16, are shown in red, the deletion on paternal chromosome is shown in blue (pt 122.3), and deletions, for which parental origin could not be determined, are shown in black. Numbers refer to ACDMPV cases. Locations of deletion breakpoints (BPs) are indicated by names of flanking repetitive elements. SRO smallest deletion overlap delineating upstream enhancer region, unk unknown sequence, uniq unique sequence. Note that most of the deletions are flanked by retrotransposons, mainly by Alu repetitive segments. The LIPA2 element (chr16:86,266,902–86,272,916) within LINC01081 at three distal deletion breakpoints in patients 57.3, 60.4, and 127.3 is bolded (Supplemental Table S1). Thirteen novel deletions are dashed and their number bolded to distinguished from the published cases

Full size image

In patient 122.3, we identified a small ~4.1 kb de novo deletion in the centromeric portion of the upstream FOXF1 enhancer region, mapping ~9.1 kb upstream to LINC01082 (Fig. 2). Surprisingly, three informative SNPs, mapping to the deleted region, consistently showed that the deletion arose de novo on the paternal chromosome 16 (Supplemental Table S1).

Fig. 2
figure 2

UCSC browser display of the enhancer region upstream to FOXF1. a An ~60 kb SRO including LINC01081 and LINC01082, CpG islands, H3K27Ac mark, DNaseI hypersensitivity clusters, and transcription factor ChIP-Seq data. b The ~4.1-kb deletion CNV identified in pt 122.3 maps to the FOXF1 upstream enhancer region. Location of the transcription factor and CTCF-binding sites, and a diagram of histone modification in the deleted region is shown

Full size image

Bioinformatic analyses of the FOXF1 locus on 16q24.1

Hi-C analysis of chromatin interactions around the FOXF1 gene revealed that FOXF1 and its upstream enhancer reside within the same topologically associated domain (TAD), spanning ~400 kb upstream of FOXF1 (Fig. 3). Intriguingly, FOXF1 maps at the distal edge of the TAD (Fig. 3), consistent with previous suggestion by Parris et al. (2013) based on distribution of the CTCF-binding sites. This domain exhibits stronger or more frequent chromatin interactions within its boundaries in fetal lung fibroblasts IMR-90 compared to non-lung cell lines: GM12878, HUVEC, NHEK, and HMEC. The strength of these interactions positively correlated with FOXF1 expression, which was stronger in IMR-90 cells than in, e.g., HUVEC, and was around the threshold level in peripheral blood lymphocytes (GM12878).

Fig. 3
figure 3

Genomic in-cis interactions at chromosome 16q24.1. a Hi-C contact maps are visualized as heatmaps for IMR-90, GM12878, HUVEC, NHEK, and HMEC cell lines. b Tiles mark active transcription start sites (red), transcribed (green), enhancer (yellow), low (gray) and heterochromatin (purple) state calls with ChromHMM within the represented locus. Shaded region highlights the active chromatin region in IMR-90 cells, whereas the same locus is not enriched for active chromatin states in other cell types, especially on the left-side of TAD boundary. c Zoomed-in view of Hi-C contact map around the FOXF1 locus in IMR-90 cells at 5 kb resolution. d Tiles mark genomic locations of SRO-enhancer region (yellow) and FOXF1 (red), FOXC1 (blue) and FOXL1 (green) genes. e Histogram represents CTCF ChIP-seq enrichment levels within the visualized locus. f Histogram represents RNA PolII ChIP-Seq enrichment levels. g Positions of two strong CTCF-binding sites (chr16:86,551,417–528 and 86,553,812–944; hg19) at the TAD boundary, 3.3 and 5.7 kb downstream of the 3′ end of FOXF1 (chr16: 86,548,070)

Full size image

In silico analysis of the FOXF1 upstream enhancer region identified an 880-bp-long CpG island (71 % GC-rich) located 0.3 kb from the known transcription start site of LINC01081 (Supplemental Fig. S1). We have previously shown that LINC01081 positively regulates FOXF1 expression (Szafranski et al. 2014). The methylation status of this CpG island is unknown. However, if this region is differentially methylated, it might contribute to the proposed epigenetic regulation of LINC01081 and thus also FOXF1 expression.

We also found that the ~4.1-kb region of the FOXF1 upstream enhancer, adjacent upstream to another lncRNA gene, LINC01082 and deleted in patient 122.3, contains the TATA-box (TTATAAATAGGAATT; chr16:86,220,297–86,220,311) and the binding sites for several transcription factors including myocyte enhancer factor-2 (MEF2), RING finger protein LUN1, hepatic leukemia factor (HLF), and myeloblastosis proto-oncogene protein (MYB) (Fig. 2a). Importantly, LUN1 and MEF2 are relatively highly expressed in the human lungs (Brand 1997; Chu et al. 2001), and might be involved in long distance interaction with the FOXF1 promoter.

Reporter assay

Because the 4.1-kb region, deleted in patient 122.3, contains the TATA-box located close to the lncRNA gene, LINC01082, we tested whether this region exhibits promoter activity. We found that the deleted fragment did not exhibit any promoter activity, even when compared with the residual activity of the FOXF1 promoter (Supplemental Fig. S2), suggesting instead its function as a scaffold for LUN1, MEF2 and TATA-binding TFIID.

Molecular mechanisms of formation of 16q24.1 deletions

To infer the molecular mechanism of formation of the identified deletions and to better assess their recurrence risk, we mapped and sequenced 18 breakpoint junctions of nine 16q24.1 deletion CNVs (Fig. 1; Supplemental Table S1). Twelve breakpoints map to Alu repeats and three within LINE1 elements. In total, 53 % (16 out of 30) of all characterized 16q24 deletions were Alu–Alu-mediated. Interestingly, in three cases (pts 57.3, 119.3, 127.3), the deletion breakpoints map to the same LINE element L1PA2 (chr16:86,266,902–86,272,916) within LINC01081, indicating that it is a genomic recombination hotspot (Fig. 1; Supplemental Table S1). Microhomologies between 7 and 41 bp were found in six out of 10 cases analyzed, suggesting that those six deletions might have arisen by a template switching replicative mechanism such as fork stalling and template switching (FoSTeS), or microhomology-mediated break-induced replication (MMBIR) (Hastings et al. 2009).

Familial cases of parentally transmitted FOXF1 mutations

We have identified three ACDMPV families with novel pathogenic variants in FOXF1 transmitted from the reportedly healthy carrier fathers (Fig. 4). Sequencing of FOXF1 and its flanking regions in patient 124.3 (Fig. 4a) revealed a complex genomic rearrangement within a noncoding portion of the FOXF1 first exon, mapping six base pairs upstream of the FOXF1 ATG codon (Supplemental Fig. S3a). A copy of a portion of the first intron of the lncRNA FENDRR (chr16:86,540,260/271–86,540,607/610), encoded upstream and opposite to FOXF1, was inserted into the untranslated portion of the FOXF1 first exon, replacing 5–15 bp in the position chr16:86,544,155/165–86,544,169. In addition, 265 bp of this inserted fragment (chr16:86,540,342–86,540,607/610) was inverted. This small insertion occurred within the centromeric portion of (CGG)n simple repeat, which is normally divided by a unique sequence containing the FOXF1 AUG codon (Supplemental Fig. S3b). Importantly, it also introduced five alternative AUG initiation codons into the 5′ untranslated part of the FOXF1 first exon, none of which is in frame with FOXF1, generating potential translation start sites for five novel peptides. This complex rearrangement within FOXF1 was inherited from the reportedly healthy father, in whom it arose de novo on the grandmaternal chr16. No evidence of somatic mosaicism of this rearrangement was detected.

Fig. 4
figure 4

Unusual inheritance and manifestation of pathogenic variants in the FOXF1 locus. Pedigrees of ACDMPV families with reportedly healthy parent transmitting FOXF1 pathogenic variant to their affected children. a Family 124 with the complex genomic insertion rearrangement (ins) within the 5′ noncoding part of the FOXF1 first exon. b Family 130 with the frameshift mutation within the FOXF1 first exon. c Family 138 with the pathogenic missense mutation in the FOXF1 first exon transmitted by healthy father, 70 % mosaic in peripheral blood. A twin sister (shaded symbol) of the deceased ACDMPV patient has partial anomalous venous return (Reiter et al. 2016). d Family 123 with the frameshift mutation transmitted by healthy mother, likely germline mosaic for the mutation

Full size image

In family 130 (Fig. 4b) with two children affected by ACDMPV, we found a deletion of five nucleotides in the coding region of the FOXF1 first exon, resulting in a translational frameshift (c.90_96del; p.Ser31fs). This pathogenic variant occurred de novo in the father who was found to be approximately 20 % mosaic for the mutation in the peripheral blood nucleated cells.

We found a missense FOXF1 mutation c.C231A, (p.Phe77Leu) in family 138 (Fig. 4c) with four children, one of them affected by ACDMPV, that was inherited from the healthy father who was approximately 70 % mosaic for the mutation in the peripheral blood nucleated cells (Reiter et al. 2016).

In family 134 (pedigree not shown), the affected newborn had an in-frame duplication c.54_59dup (p.Gly22_Gly23dup) in the first exon of FOXF1. The variant was located within a stretch of 11 Gly residues. There are few known in-frame deletions and duplications within this Gly repeat (e.g., p.Gly19del, p.Gly13_Gly17del, p.Gly17dup), listed in the ExAC database (http://exac.broadinstitute.org), interpreted as non-pathogenic. Moreover, the number of Gly in this repeat varies between species (e.g., there are six residues in rabbit, 12 in macaque). The healthy father is heterozygous for this variant. Therefore, we concluded that this Gly duplication is unlikely to be pathogenic for ACDMPV. Thus, the causative factor for the disease in this case remains unknown.

In addition, in family 123 (Fig. 4d), two siblings died of ACDMPV 6 years apart. Sequencing of FOXF1 from one of the siblings revealed the presence of a deleterious frame-shift mutation c.849_850del (p.I285fs) in the first exon. This variant was the cause of ACDMPV also in the other sibling, and must have been transmitted from the healthy mother who is likely germline mosaic or low-level somatic mosaic.

Whole exome sequencing

Three ACDMPV cases (families 114, 121, and 128), negative for FOXF1 mutations and CNVs in FOXF1 or its upstream enhancer region, were analyzed by WES. From the list of variants obtained for each proband, we filtered out synonymous or non-exonic SNVs/indels and variants with minor allele frequency >1 % in 1000 Genomes, Exome Variant Server, or in-house exome databases, which left 762 (pt 114.3), 681 (pt 121.3), and 709 (pt 128.3) variants. Analysis of WES data in family 114 revealed one de novo missense mutation c.1564T>C (p.Trp522Arg) in ZMYND11 validated by Sanger sequencing (Table 2). In family 121, we identified four de novo variants confirmed by Sanger sequencing: missense mutations c.463C>T (p.Arg155Cys) in SLC50A1 and c.881A>G (p.Tyr294Cys) in ESRP1, a non-frameshift deletion c.533_542delinsC (p.Ser179_Ser173del) in MPRIP, and a frameshift insertion c.2819_2820insT (p.Gly941fs) in DOCK8 (Table 2). None of those de novo variants were present in ESP or 1000 Genomes databases. By exploring ExAC database we found that the variant in SLC50A1 was reported in two healthy individuals (of Latino and European ancestry) and the variant in MPRIP was present in 152 individuals (MAF = 0.0015). However, given the complexity of the FOXF1 locus and its epigenetic regulation, none of these variants should be definitively excluded as a potential contributor to ACDMPV. In family 128, we did not find any de novo variants.

Table 2 WES-detected SNVs identified in three families with ACDMPV patients negative for FOXF1 mutations and CNVs on chromosome 16q24.1
Full size table

Moreover, in families 121 and 128, we identified inherited missense variants c.631C>G (p.Leu211Val) in pt 121.3 and c.3256G>A (p.Ala1086Thr in pt 128.3 in PLXNB2 encoding plexin B2 (Table 2). Further, we identified homozygous and compound heterozygous variants in all affected probands from 114, 121 and 128 families. In patient 114.3, we found 66 homozygous variants in 63 genes and 27 compound heterozygous in 10 genes, in patient 121.3, we identified 56 homozygous and 24 compound heterozygous variants in 49 and 7 genes, respectively, and in patient 128.3, we identified five homozygous variants in five genes and five compound heterozygous variants in two genes.

CNV analyses of WES data did not reveal any non-polymorphic deletion or duplication in probands 114.3 and 121.3 and no genomic imbalances were identified by whole-genome array CGH in pt 128.3. By analyzing B-allele frequency in WES data in probands 114.3, 121.3, and 128.3, we did not identify any AOH region larger than 5 Mb.

Discussion

To date, approximately 150 imprinted genes clustering in 16 genomic loci have been described in mice (Barlow and Bartolomei 2014). In humans, disease-related genomic imprinting has been well defined only for a few loci: 15q11.2 in Prader–Willi (PWS, OMIM 176270) and Angelman (OMIM 105830) syndromes, 11p15 in Beckwith–Wiedemann syndrome (BWS, OMIM 130650) and Silver-Russell syndrome (OMIM 180860), 14q32 in Kagami–Ogata syndrome (OMIM 608149), 20q13.32 in McCune–Albright syndrome (OMIM 174800), and 6q25.3 in transient neonatal diabetes mellitus 1 (OMIM 601410) (Bartolomei and Ferguson-Smith 2011). However, recent analyses have shown that the occurrence of non-canonical imprinting, with a biased allele-specific gene expression as opposed to complete allele silencing, is likely underestimated (Gregg 2014). These partial allelic imbalances may result, e.g., from incomplete methylation of imprinting control regions (ICRs) or from an ICR being epigenetically modified only in a subpopulation of cells (Gregg 2014).

We have accumulated the largest collection of ACDMPV samples worldwide (N = 141 families), in which we have identified 86 pathogenic variants in the FOXF1 locus: 38 deletion CNVs, one complex rearrangement and 47 point mutations. In the vast majority of the remaining 55 families, DNA was not of sufficient quality for genetic testing. We have previously reported pathogenic genomic deletions involving FOXF1 or its upstream regulatory region in 17 patients with histopathologically verified ACDMPV that arose de novo on the maternal chromosome 16. Based on these results, we proposed that the FOXF1 locus in chromosome 16q24.1 is imprinted in the human lungs (Stankiewicz et al. 2009; Sen et al. 2013a, b; Szafranski et al. 2013a, b; Dharmadhikari et al. 2015). We previously suggested that this imprinting is incomplete (~35 % expression from one parental allele vs ~65 % expression from the other allele) (Szafranski et al. 2013a). Together with 13 novel deletions reported here, 30 out of 31 CNV deletions involving FOXF1 or its upstream enhancer in patients with ACDMPV arose de novo on the maternal chromosome 16, now providing statistically significant evidence for genomic imprinting at this locus (p < 4E−06). However, we have now also identified an upstream deletion CNV (pt 122.3) and three non-deletion variants in FOXF1 (pts 124.3, 130.3, 138.3), pathogenic for ACDMPV, that arose on paternal chromosome 16q24.1, indicating complex genomic and epigenetic regulation of the FOXF1 locus.

The 4.1-kb de novo deletion within the upstream FOXF1 enhancer on paternal chromosome 16 in pt 122.3 removed the binding sites for a number of transcription factors, including MEF2 and LUN1 that exhibit increased expression in the lungs. MEF2 regulates cell proliferation as a target of several growth factor signaling pathways and has been shown to play an important role in myogenesis, including morphogenesis of visceral muscles (Brand 1997; Black and Olson 1998). We hypothesize that loss of MEF2 binding in patient 122.3 might have contributed to the abnormal development of lung vasculature, a feature typical for ACDMPV. LUN1, which is highly expressed in alveolar epithelium (Chu et al. 2001) has binding sites located also in regulatory region of genes for E-cadherin and talin, which regulate cell motility (Oyanagi et al. 2004). Gene expression profiling in ACDMPV lungs and Foxf1 +/− mouse lungs showed that E-cadherin is one of the Foxf1 targets (Sen et al. 2014). Thus LUN1 might control cell motility during alveolar development, targeting E-cadherin expression directly and indirectly through FOXF1. The deleted interval in pt 122.3 also includes the TATA-box, present in 25 % of eukaryotic promoters; however, we found no promoter activity within this region. Since the upstream regulatory region becomes juxtaposed with the FOXF1 promoter, we propose that the identified transcription factor binding sites, in particular those for MEF2, LUN1, and TBP (TFIID), similarly as the relatively closely located GLI2-binding sites (Szafranski et al. 2013a), may interact with the FOXF1 promoter following chromatin looping to directly up-regulate FOXF1 promoter activity.

In patient 124.3, we found a complex genomic rearrangement within the noncoding portion of FOXF1 exon 1, potentially interfering with initiation of transcription and translation. The most plausible explanation of the absence of ACDMPV phenotype in the father is that he could be a mosaic for this rearrangement.

The other two identified paternally transmitted pathogenic variants in families 130 and 138, mapping within the coding portions of FOXF1, were inherited from the reportedly healthy fathers who are mosaics for the mutations in the blood. These two cases exemplify incomplete penetrance of ACDMPV likely due to somatic mosaicism for a pathogenic variant manifesting as non-mosaic in their affected children. The clinical relevance of somatic mosaicism has become more evident only recently (Campbell et al. 2015). Family 138 also illustrates variable expressivity of ACDMPV with two siblings presenting with severe pulmonary hypertension after birth, one of whom had hypoxemia, but survived beyond infancy, and the other had partial anomalous pulmonary venous return. The third sibling died in the neonatal period from ACDMPV (Reiter et al. 2016).

Given that all but one of the ACDMPV variants that arose on paternal chromosome 16 mapped within the FOXF1 gene, whereas all but one deletions that included the upstream enhancer arose on the maternal chromosome, we propose that the FOXF1 locus is imprinted through epigenetic modification of its distant lung-specific enhancer. In support of this notion, we have found that whereas the FOXF1 promoter is not methylated, the CpG island overlapping cluster of GLI-binding sites within the upstream enhancer is differentially methylated, and this methylation reduces regulatory function of the enhancer (Szafranski et al. 2013a). Moreover, the promoter region of the lncRNA gene, LINC01081, that positively regulates FOXF1 expression, located within the enhancer, overlaps with another CpG island, suggesting the possibility of the regulation of the expression of LINC01081 by allele-specific differential methylation of its promoter. Most recently, Dello Russo et al. (2015) reported a de novo ~2.6-Mb deletion (chr16:83,676,990–86,292,585) at maternal chromosome 16q23.3q24.1, encompassing LINC01082 and disrupting LINC01081, in a patient with a rare developmental lung disease, pulmonary capillary hemangiomatosis, providing further evidence for genomic imprinting and demonstrating allelic affinity of this genomic locus.

Whether epigenetic modification of the upstream enhancer occurs on paternal or on maternal chromosome 16 is currently unclear and requires further studies. Unlike ICRs of maternally imprinted genes that are typically located at the imprinted gene promoter, the intergenic localization of the FOXF1 locus ICR suggests paternal imprinting (Ferguson-Smith 2011). Also the location of all but one pathogenic deletion on maternal chromosome suggests paternal imprinting. In the paternal imprinting model (Fig. 5a), deletion of the strong enhancer on maternal chromosome 16 leads to ACDMPV, whereas deletion of the weak enhancer on paternal chromosome 16 is benign. In the maternal imprinting model (Fig. 5b), deletion of the weak enhancer on maternal chromosome 16 slightly decreases FOXF1 expression, resulting in ACDMPV whereas deletion of the strong enhancer on paternal chromosome 16 reduces expression of FOXF1 more dramatically and is embryonic lethal.

Fig. 5
figure 5

Proposed model of ACDMPV and UPD(16) etiologies due to (a) paternal or (b) maternal imprinting of the FOXF1 upstream enhancer on chromosome 16q24.1. Epigenetic modification of the enhancer reduces its ability to stimulate FOXF1 transcription. In paternal imprinting model (a), the majority of FOXF1 transcription depends on the strong enhancer located on the maternal chromosome 16. Thus deletion of the enhancer on this chromosome would significantly reduce FOXF1 expression and cause ACDMPV whereas the deletion on the paternal chromosome 16q24.1 would reduce FOXF1 expression less with no clinical consequences. Consequently, maternal UPD(16) would increase FOXF1 expression level (pathogenic) whereas paternal UPD(16) would slightly decrease FOXF1 expression level (benign). In the maternal imprinting model (b), the upstream deletion of the weak enhancer on the maternal chromosome 16 would reduce FOXF1 expression, resulting in ACDMPV. The deletion of the strong enhancer on the paternal chromosome would reduce FOXF1 expression would be embryonic lethal. According to this model, in maternal UPD(16), the level of FOXF1 would be insufficient to prevent development of some ACDMPV features and other anomalies typical for maternal UPD(16). In contrast, point mutations in FOXF1 are found on both parental alleles, suggesting in-trans function of the FOXF1 enhancer, likely using lncRNAs (Szafranski et al. 2013a)

Full size image

Interestingly, multiple congenital malformations, including pulmonary hypoplasia, heart defects, tracheoesophageal fistula, gut malrotation, absent gall bladder, renal agenesis, hydronephrosis, imperforate anus, and single umbilical artery (SUA), seen in the vast majority of children with ACDMPV, are also observed in patients with maternal uniparental disomy 16, UPD(16). In stark contrast, a relatively normal phenotype was reported in few patients with paternal UPD(16) (Kohlhase et al. 2000; Hamvas et al. 2009), and the presence of imprinted gene(s) on chromosome 16 was suggested as causative for maternal UPD(16) phenotype (Yong et al. 2002). Differences in the clinical features observed in maternal UPD(16) cases compared to paternal UPD(16) cases and similarities between ACDMPV and maternal UPD(16) phenotypes, indicate that the identified genomic imprinting at the FOXF1 locus may be responsible for some phenotypic features of maternal UPD(16) (Dharmadhikari et al. 2015). The underlying mechanism could be similar to that in patients with maternal (but not paternal) duplication/triplication of the imprinted PWS/AS region in 15q11.2, e.g., due to inv dup(15). In the paternal imprinting model (Fig. 5a), in maternal UPD(16), FOXF1 expression is increased, manifesting typical features of UPD(16) including organs involved also in patients with ACDMPV, whereas in paternal UPD(16), FOXF1 expression only slightly decreases and is benign. In the maternal imprinting model (Fig. 5b), the level of FOXF1 in maternal UPD(16) is reduced, whereas paternal UPD(16) increases FOXF1 expression and is benign. However, given that there are only two reports of apparently benign paternal UPD16 (isodisomy) cases vs. high prevalence of maternal UPD16 (heterodisomy) cases due to common trisomy 16 (>1 % of all pregnancies), it is also possible that paternal UPD(16) is early embryonically lethal.

The observed enhancer-dependent regulation and proposed genomic imprinting in the FOXF1 locus are likely mediated by genomic insulator sites binding CTCF as was shown for the BWS region on chromosome 11p15. Supporting this model, in silico Hi-C analyses of the chromosome 16q24.1 genomic structure showed that this region is organized into ~400 kb TADs with FOXF1 being located at the TAD boundary (Fig. 3). TAD boundaries exhibit conservation across species and remain largely constant across multiple cell types (Dixon et al. 2012), suggesting that variation in intra-domain interactions, such as chromatin looping, may be crucial for dynamic regulation of gene expression in a cell type-specific fashion. Using a chromosome conformation capture-on-chip (4C) analysis, we have previously shown that the region upstream of FOXF1, including its promoter, comes in contact with the upstream enhancer sequences in a time and tissue-specific manner (Szafranski et al. 2013a); these interactions are stronger or more frequent in fetal lung fibroblasts than in cells of tissues other than lungs (Fig. 3).

Recently, disruption of TADs, resulting in “enhancer adoption”, has been shown as a novel disease-causing mechanism in patients with limb anomalies (Lupiáñez et al. 2015). We suggest that deletion CNVs or balanced paracentric inversions (Parris et al. 2013) removing or replacing, respectively, the FOXF1 TAD boundary with CTCF-binding sites (Guo et al. 2015), would expose genes neighboring TAD to a non-physiological environment deregulating their expression. This mechanism could explain our observation that in contrast to FOXF1 point mutations (Sen et al. 2013b) and upstream deletion CNVs (Stankiewicz et al. 2009; Szafranski et al. 2013a, 2014), genomic deletions of FOXF1 at its TAD boundary, and the flanking genes were associated with severe congenital heart defects, including hypoplastic left heart syndrome (HLHS) and SUA (Table 1; Supplemental Table S1). Co-existence of HLHS and SUA has been well documented (Tasha et al. 2014; Araujo Júnior et al. 2015). We have previously suggested that HLHS may result from variants in the neighboring FOXC2 and FOXL1 genes; however, screening for mutations in patients with HLHS revealed no pathogenic variants in those genes (Iascone et al. 2012). Alternatively, disruption of lncRNA FENDRR that maps 1.7 kb upstream of FOXF1 in the opposite orientation, and likely utilizes the same bi-directional promoter as FOXF1 could lead to HLHS and SUA. Corroboratively, Grote et al. (2013) reported that homozygous loss of Fendrr in mice led to hypoplasia of the myocardium affecting ventricular walls and the interventricular septum and ventral body wall (omphalocele), likely due to in trans deregulation of the cardiac master transcriptional regulators Gata6 and Nkx2-5. Further, Sauvageau et al. (2013) and Lai et al. (2015) independently demonstrated defects in lungs and heart in the Fendrr −/− mouse neonates.

Interestingly, most of deletion CNVs in 16q24.1 were flanked by retrotransposons with greater than 50 % being Alu-mediated and many representing different Alu families (Szafranski et al. 2013a). Recently, Alu-mediated genomic rearrangements were shown to be products of replication and not recombination errors (Gu et al. 2015). Of note, Jacques et al. (2015) demonstrated that transposable elements have contributed hundreds of thousands of novel regulatory elements to the primate lineage and reshaped the human transcriptional landscape. In vertebrates, transposable elements occur in more than two-thirds of mature lncRNAs, whereas they seldom occur in protein-coding transcripts. Moreover, transposable elements were found in biased positions and orientations within lncRNAs, particularly at their transcription start sites, which suggests a role in the regulation of lncRNA transcription (Fatica and Bozzoni 2014). We suggest that the high rate of the retrotransposon-mediated CNVs in 16q24.1 may result from replication–transcription collisions due to their residual transcriptional activity.

Besides variants in the FOXF1 locus, CNVs or SNVs in other genes involved in SHH or other signaling pathways essential for lung development might be also causative for ACDMPV. Our WES analyses in three unrelated ACDMPV families (114, 121, and 128) revealed de novo variants in five genes: DOCK8, ESRP1, MPRIP, SLC50A1, and ZMYND11. All these genes are involved in cell signaling or transcription regulation in general and their variants may contribute to development of ACDMPV. ESRP1 (pt 121.3) is particularly interesting in this context. It encodes endothelial splicing regulatory protein 1 (ESRP1) functioning as an epithelium-specific regulator of FGFR2 splicing into FGFR2-IIIb isoform] (Warzecha et al. 2009). Various isoforms of FGFR2 are involved in epithelial–mesenchymal crosstalk during embryonic development, and they also play a role in epithelial–mesenchyme transitions during lung and heart development. Thus, loss of ESRP1 or its function might contribute to ACDMPV by affecting FGF signaling.

In families 121 and 128, we identified two inherited missense variants in PLXNB2 encoding plexin B2. Plexins function as receptors of semaphorins and were shown to play a crucial role in lung branching morphogenesis (Kagoshima et al. 2001). Interestingly, loss of class 3 semaphorins (SEMA3) was attributed to dysmorphic vascularization during mouse lung development, resembling features of ACDMPV (Joza et al. 2012). Since both variants were inherited from healthy carrier father, they may function as modifiers of the pathway(s) contributing to ACDMPV. Nevertheless, other variants in PLXNB2 might still be causative for ACDMPV.

In aggregate, our data highlight complexity of genomic architecture of the FOXF1 locus at chromosome 16q24.1 and regulation of FOXF1 expression through epigenetic modification of its upstream enhancer. Unlike the 31 identified CNVs that all arose de novo and, with one exception, on maternally inherited chromosome 16, FOXF1 pathogenic SNVs can be inherited from either parent, who may be a mosaic carrier. We propose that genomic imprinting of the FOXF1 locus is due to parent- and tissue-specific activity of the FOXF1 enhancer regulated by lncRNAs LINC01081 and LINC01082 and chromatin folding within a defined TAD, with FOXF1 being located at its boundary. The FOXF1 promoter is presumably activated by the enhancer-bound transcription factors (e.g., GLI2, MEF2, LUN1, and TFIID). This complex gene regulation in 16q24.1, in particular a non-canonical mode of FOXF1 locus imprinting, likely contributes to variable expressivity and incomplete penetrance of ACDMPV. We also suggest that variants in two other genes could be causative (ESRP1) or function as modifiers (PLXNB2) of the ACDMPV phenotype.