Distinguishing benign from pathogenic duplications involving GPR101 and VGLL1-adjacent enhancers in the clinical setting with the bioinformatic tool POSTRE

Abstract

Background:

Structural variants (SVs) that disrupt topologically associating domains (TADs) can cause disease by rewiring enhancer-promoter interactions. Duplications involving GPR101 are known to cause X-linked acrogigantism (X-LAG) by enabling aberrant expression of GPR101 through hijacking of enhancers at VGLL1. However, not all GPR101-containing duplications are pathogenic, presenting a diagnostic challenge, especially in the prenatal setting.

Methods:

We evaluated POSTRE, a tool designed to predict the regulatory impact of SVs, to distinguish pathogenic from benign GPR101 duplications. We analyzed six non-pathogenic duplications, and 27 known X-LAG associated pathogenic duplications. Tissue-specific enhancer maps built using H3K27ac ChIP-seq and ATAC-seq data as well as gene expression data derived from human anterior pituitary samples were integrated into POSTRE to enable predictions in a X-LAG relevant tissue context.

Results:

POSTRE correctly classified all 33 duplications as benign or pathogenic. In addition, one X-LAG case with mild clinical features (e.g., severe GH hypersecretion in the absence of pituitary tumorigenesis) was found to include only 2/5 VGLL1 enhancers (also predicted to be the weakest enhancers), whereas all 26 typical X-LAG cases had ≥4 enhancers duplicated. This suggests that milder enhancer hijacking at VGLL1 could explain the different clinical features of X-LAG in this individual.

Conclusions:

These findings support the utility of POSTRE to support diagnostic pipelines when interpreting SVs affecting chromatin architecture in pituitary disease. By accurately modelling enhancer adoption in a cell type-specific context, POSTRE could help to reduce uncertainty in genetic counselling and offers a rapid alternative to performing chromatin conformation capture experiments.

Introduction

The spatial organization of the genome plays a fundamental role in gene regulation. At the sub-mega-base scale, chromatin architecture is partitioned into topologically associating domains (TADs)¹. These are self-interacting regions that constrain enhancer-promoter communications and insulate genes from regulatory elements outside their TAD^1–3. Disruption of TADs by structural variants (SVs), such as duplications, deletions, inversions, or translocations can result in aberrant gene expression by altering the regulatory landscape, a phenomenon increasingly recognized as a common mechanism in human disease^3–7. These so-called “TADopathies”⁸ have been implicated in a growing number of disorders, highlighting the importance of chromatin architecture in maintaining gene expression fidelity⁹.

One TADopathy is X-linked acrogigantism (X-LAG), a very rare condition characterized by early pediatric growth hormone (GH) excess due to pituitary tumors and/or hyperplasia¹⁰. The disease is caused by duplications at chromosome Xq26.3 involving the GPR101 gene. We demonstrated recently that these duplications can disrupt the local GPR101 TAD structure and lead to the formation of a pathological neo-TAD. The neo-TAD brings GPR101 into contact with ectopic enhancers, resulting in its pathological overexpression^11,12. A critical ectopic enhancer cluster implicated in this mechanism resides within the VGLL1 gene locus and is active specifically in pituitary cells¹². The action of this element on GPR101 likely underlies the pathogenicity of duplications in X-LAG.

Despite the growing understanding of these regulatory mechanisms, distinguishing pathogenic from benign duplications involving GPR101 remains a major diagnostic challenge. This is especially true in the context of prenatal chromosomal microarray analysis (CMA), where duplications may be detected incidentally, creating the potential for significant distress, due to the challenges inherent in fetal phenotyping. Without functional data, interpreting the clinical significance of these findings is difficult, often leading to uncertainty in genetic counseling and clinical decision-making¹³. We recently showed that 4C-seq and HiC can map chromatin interactions and assess TAD integrity in these, thereby playing a diagnostic role¹⁴. Notwithstanding their clinical utility, we are aware that these techniques are highly technically specialized, labor-intensive, and not readily available in most clinical laboratories.

In this context, there is a pressing need for computational approaches capable of predicting the regulatory impact of structural variants using only the SV genomic coordinates and accessible omics data (e.g., HiC, ChIP-Seq, ATAC-seq, RNA-seq). One such approach is POSTRE (Prediction Of STRuctural variant Effects), a recently developed tool designed to model SV long-range pathomechanisms, including enhancer adoption and neo-TAD formation, based on tissue-specific genomic data. By integrating TAD maps with chromatin accessibility, histone modifications, and gene expression data, POSTRE enables the prediction of regulatory disruptions caused by SVs in phenotypically-relevant cellular contexts¹⁵. As originally implemented, POSTRE was only applicable to a limited set of congenital defects (i.e. neurodevelopmental, craniofacial, limb, heart) for which genomic data in disease relevant human tissues was analyzed and integrated. In the present study, we developed an expanded version of POSTRE that is enriched with anterior pituitary-specific data, thus making it compatible with X-LAG and, potentially, with other disorders involving the pituitary gland. We then assessed whether POSTRE could accurately discriminate pathogenic from benign duplications at the GPR101 locus and thereby serve as a tool to inform the clinical interpretation of TAD-disrupting SVs.

Materials and Methods

Study population

The study population consisted of 33 individuals, of whom 30 had previously reported pathogenic GPR101 duplications associated with X-LAG^{10,12,16–20} or non-pathogenic duplications¹⁴. Three newly identified individuals were also included, they harbored microduplications at the GPR101 locus on chromosome Xq26.3 that were discovered incidentally during prenatal or pediatric genetic testing at sites in three different countries. None of the three individuals exhibited signs of pituitary hyperplasia/tumor, gigantism, or other endocrine abnormalities consistent with X-LAG at the time of assessment. Individual F4A, a female, was identified via prenatal chromosomal microarray analysis (CMA) of DNA extracted from amniotic fluid that was performed for advanced maternal age. Individual F5A, a female, was identified after CMA on chorionic villus sampling for investigation of unrelated fetal ultrasound findings. Individual F6A was diagnosed postnatally as part of a clinical workup for developmental concerns without endocrine pathology. Some details of F6A were previously reported²¹.

Copy number variant (CNV) analysis in new subjects

CMA was performed in F4A on embryonic DNA derived from amniotic fluid using the Agilent Technologies G3 ISCA V2 8×60K CGH microarray platform. This revealed a 366 kb duplication at Xq26.3, corresponding to genomic coordinates chrX:136,104,659–136,470,844 (hg19). The duplicated segment included the genes RBMX and GPR101 but did not encompass the VGLL1 gene or its intronic enhancer cluster, which we previously linked to the pathogenesis of X-LAG¹². The duplication was inherited from a healthy parent, who had no history of growth or other disorders; she was born at term after a normal pregnancy and no evidence of growth disorders was noted.

DNA for F5A was extracted directly from blood and analyzed using the Illumina 850K CytoSNP v1.4 SNP microarray (50 kb mean effective resolution). F5A had a 237 kb duplication involving the region chrX:135,954,223–136,191,468 (hg19). This duplication also included RBMX and GPR101, but did not reach the VGLL1 locus. F5A had no history of growth disorders and her hormone levels were normal.

Genomic DNAs from F6A and his healthy parent were analyzed using a clinical-grade 60K CGH array (Agilent ISCA CGH 60K, AMADID 031746), which provides genome-wide coverage with an estimated effective resolution of about 100 kb for gains and losses. The aCGH array analysis identified a 444.8 kb duplication spanning chrX:135,702,382–136,147,166 (hg19). aCGH analysis performed on F6A’s parent did not detect the duplication, indicating a de novo origin. Confirmation of the CNV was performed using droplet digital PCR (ddPCR), with probes targeting four genes located within the duplication (CD40LG, ARHGEF6, RBMX, and GPR101) and two control genes outside the duplicated segment (HTATSF1 and ZIC3). DNA from a previously diagnosed X-LAG patient harboring a constitutional duplication served as a positive control for the ddPCR study. ddPCR showed that the duplication was mosaic, with an average copy number of 1.44 for five probes located within the duplicated region (Supplementary Figure 1). This indicated that approximately 44% of peripheral blood cells carried the duplication. The affected genes included CD40LG, ARHGEF6, RBMX, and GPR101, while flanking genes such as HTATSF1 and ZIC3 showed a diploid copy number.

Genomic data processing (RNA-seq, ChIP-seq, ATAC-seq and HiC data) and incorporation into POSTRE

To enable accurate modeling of enhancer adoption and regulatory disruption in pituitary tissue, we generated a tissue-specific dataset (NormalPituitary) and supplemented this with publicly available transcriptomic data (NormalPituitary2) for cross-validation.

Normal adult anterior pituitary tissue samples (two males, one female) were obtained post-mortem through Cureline Inc, as previously reported¹². These were used to generate RNA-seq, ATAC-seq, and H3K27ac ChIP-seq data. RNA sequencing was performed on total RNA extracted from these tissues (for further details refer to Franke et al.¹²). Prior to its incorporation into POSTRE, gene expression levels were converted to RPKMs and aggregated. In this regard, for each gene, the average expression across the three samples was taken as reference. In parallel, RNA-seq data normalized in the form of Transcripts Per Million (TPM)s, from normal whole pituitary glands were obtained from the Genotype-Tissue Expression (GTEx) Portal. In particular, the gene_tpm_v10_pituitary.gct.gz file was downloaded in December 2024 from the “Bulk tissue expression” section in the downloads page. Next, the median expression value for each gene across all available samples was taken as reference. The GTEx data formed the basis for the NormalPituitary2 condition in POSTRE, providing an additional independent gene expression match to the enhancer maps.

The TAD coordinates used by POSTRE, in NormalPituitary and NormalPituitary2, were derived from the brain prefrontal cortex boundary map provided from Schmitt et al.²² as previously done for other tissues (for more details please visit the POSTRE user guide). ChIP-seq profiling of the H3K27ac histone mark was performed on an additional normal adult anterior pituitary sample using the ChIP-IT High Sensitivity kit (Cat. No. V13–53040, Active Motif,) and a polyclonal anti-H3K27ac antibody (Cat. No. 39135, Active Motif). Library preparation was conducted using the Novogene NGS DNA Library Prep Set (Cat. No. PT004), and sequencing was performed on the Illumina NovaSeq X Plus platform with 150 bp paired-end reads.

ATAC-seq was performed on two anterior pituitary samples using about 50,000 cells per preparation. The ATAC-seq Kit (Cat. No. C01080006, Diagenode) was used according to the manufacturer’s protocol, including Illumina-compatible library generation and indexing (UDI Index Set I, Diagenode). Sequencing was carried out at Azenta Life Sciences using the Illumina NovaSeq platform (2×150 bp), yielding about 350 million paired-end reads (~105 GB) per sample.

The resulting H3K27ac and ATAC-seq datasets were integrated to define tissue-specific active enhancers. In this regard, enhancers were defined as ATAC peaks (fold change>4 and q value <0.05) located more than 5 kb from any protein coding transcription start site (TSS) and within 500 bp of H3K27ac peaks (fold change >2 and p value <0.05). Finally, the enhancer maps were matched, in parallel, with the NormalPituitary and NormalPituitary2 gene expression datasets, giving rise to the two different conditions evaluated in POSTRE in relation with Endocrine abnormalities.

In silico modeling with POSTRE

The pathogenic potential of multiple CNVs was assessed using the POSTRE software, a computational tool, available at https://github.com/vicsanga/Postre, designed to predict the regulatory impact of SVs that alter enhancer-promoter interactions¹⁵. To do so, POSTRE utilizes gene expression and enhancer maps from disease relevant cell types and combines them with available TAD maps. Since the first version of POSTRE did not present genomic data relevant for the interpretation of pituitary related diseases, an expanded version has been created in this project, as described in the previous Method section. The expanded version, which includes the newly incorporated pituitary data, is currently available on request and will be directly available through POSTRE GitHub page upon peer-reviewed publication of this manuscript.

Regarding the details of POSTRE usage, it was run specifying the “Endocrine” phenotype category and with the rest of default parameters. A potential limitation of the default running mode is that it only considers as disease relevant those genes already associated with the specified phenotypical category (in this case endocrine-related diseases) in established databases, such as OMIM. In this regard, higher sensitivity analyses allowing broader gene inclusion were also explored by disabling the patient phenotype filter (i.e., Gene-PatientPheno option set to No).

All duplications were analysed in POSTRE through the Single SV and Multiple SV Submission interfaces. Regarding the results provided by each mode, the Single SV analysis report included: details of the predicted enhancer adoption events, redefined TAD boundaries (neo-TADs), and additional information for the impacted target genes (illustrated in Figure 2). With respect to the Multiple SV analysis, after uploading all the SVs information through a single txt file (Supplementary Table 1), a set of tables with the aggregated results of the predictions were obtained. POSTRE also allows users to upload their own TAD map to perform the interpretation of the SVs analyzed but this feature was not used, since the TAD map already selected was valid for the locus of interest.

Figure 2. Example of POSTRE output for a pathogenic (X-LAG case S22) duplication at the GPR101 locus.

Panels A-E show part of the results obtained with POSTRE for the analysis of the pathogenic duplication S22. A) Screenshot of POSTRE submission menu for the Single SV analysis mode with the X-LAG case S22 SV coordinates introduced. The duplication was evaluated using POSTRE in Standard mode with the “Endocrine” phenotype selected. This phenotype selection triggers the consideration of anterior pituitary-specific enhancer and expression data. B) Visualization of S22 duplication at the UCSC genome browser. The image was obtained through the adjustment (zoom in) of a UCSC link available from the POSTRE output. The duplicated area is highlighted in red. The session also depicts bigwigs of H3K27ac ChIP-seq and ATAC-seq data from anterior pituitary, TAD maps from the brain prefrontal cortex, and highlights POSTRE predicted active enhancers with green vertical lines. C) Main POSTRE prediction summary table. D) Graphical representation of the SV impact with respect to the GPR101 regulatory domain. In the control scenario, the existence of eight enhancers can be observed at the TAD containing the VGLL1-adjacent enhancers. These eight enhancers include the five VGLL1-adjacent enhancers (Figure 4) and three additional ones located at the centromeric side of that TAD (according to the prefrontal cortex TAD map used as reference). However, these latter three enhancers (located >250Kb away from the closest duplication breakpoint) were excluded from downstream considerations as they are never duplicated in any of the reported X-LAG duplications (Figure 1). E) Barplot highlighting the changes in the enhancer number surrounding GPR101 in the control versus patient alleles. Overall, for S22, POSTRE predicts enhancer adoption and neo-TAD formation involving the VGLL1-adjacent enhancers, resulting in a high pathogenicity score and a likely mechanism of GPR101 misexpression.

Results

Thirty three duplications involving GPR101 (Figure 1) were analyzed using the POSTRE tool, with the phenotype set to “Endocrine” and the running mode set to “Standard” (Figure 2). This configuration integrates enhancer maps derived from ATAC-seq and H3K27ac ChIP-seq data generated from human anterior pituitary cells, alongside RNA-seq profiles specific to this tissue (see Methods). The analysis also relies on prefrontal cortex-derived TAD maps and disease-gene annotations obtained from different databases, such as OMIM²³ and the Mouse Genome Database²⁴. As we established previously, owing to the overall conservation of TADs across cell types, TAD maps from other human tissues can provide informative data on gene regulatory domains and the prediction of long range pathological mechanisms related to SVs¹⁵.

Figure 1. Overview of duplications at the Xq26.3 locus involving GPR101, classified by pathogenicity.

The genomic region surrounding GPR101 (chrX:135,500,000–136,700,000, hg19) is shown with annotated protein-coding and OMIM genes (blue and dark green, respectively), CTCF binding sites (orange ChIP-seq track from GM12878 cells), and putative CREs (six we previously described¹² based on publicly available data^39,40 given as light green bars and a subset predicted by POSTRE located within or distal to VGLL1 given as purple bars). The centromeric GPR101-TAD boundary is marked by a red vertical bar. Colored bars below represent the extent of individual duplications: yellow for three non-pathogenic duplications reported by Daly et al.¹⁴ (F1A-F3A), orange for the three newly identified non-pathogenic duplications presented in this study (F4A-F6A), and blue for X-LAG-associated pathogenic duplications, including both continuous and discontinuous (patients S4 and I) rearrangements. Only pathogenic duplications span the VGLL1 intronic enhancer cluster. The light blue vertical bar highlights the smallest region of overlap of all duplications partially encompassing that CRE. All non-pathogenic duplications, despite partial TAD disruption or inclusion of other CREs such as the RBMX enhancer¹², were predicted by POSTRE to be neutral.

In all three of the new non-X-LAG cases, POSTRE predicted the duplications as non-pathogenic (Table 1 and Figure 3). The duplication in individual F4A is entirely contained within the GPR101 TAD and, accordingly, is not predicted to result in neo-TAD formation—similar to the non-pathogenic duplications we previously reported¹⁴. By contrast, the duplications in F5A and F6A spanned the centromeric TAD boundary downstream of GPR101, resulting in the formation of a neo-TAD. However, unlike in X-LAG associated duplications neither F5A nor F6A included the intronic VGLL1 enhancer cluster, which is thought to be critical for driving aberrant GPR101 expression. Both duplications encompassed a putative enhancer located near RBMX (eRBMX)¹², but the POSTRE enhancer calling pipeline does not predict this as one, indicating that this cis-regulatory element (CRE) alone might not be sufficient to induce GPR101 misexpression. As a result, POSTRE predicted these rearrangements to be neutral for GPR101 regulation despite their inter-TAD configuration. This finding refines the mechanistic interpretation we recently outlined¹⁴, where the pathogenic importance of TAD boundary disruption was emphasized, but the differential impact of distinct enhancer inclusions was not examined in detail. POSTRE predictions were also benchmarked against the three non-pathogenic GPR101 duplications that we recently reported (Table 1)¹⁴. All three were correctly predicted to be non-pathogenic, in line with the original experimental observations using chromatin conformation capture techniques.

Table 1. POSTRE output for the Multiple SV analysis.

Note: pathogenicity scores for the F3A duplication were also computed by POSTRE for the genes located in the predicted TAD centromeric to GPR101. Although the duplication is reported to lie entirely within the GPR101-TAD (Figure 1), the TAD map used by POSTRE considers that it encroaches on this centromeric TAD but does not overlap the enhancers located within. This discrepancy reflects a limitation of TAD-calling softwares, as precise annotation of TAD boundaries is often lacking.

SV ID	Phenotype	Pathogenic Score	Pathogenic (Yes/No)	Causative genes	Candidate genes (Pathogenic Score)
F1A	endocrine	0.5	No		GPR101(0.5),ZIC3(0.25)
F2A	endocrine	0.5	No		GPR101(0.5),ZIC3(0.25)
F3A	endocrine	0.5	No		GPR101(0.5),ARHGEF6(0.38),HTATSF1(0.37),RBMX(0.37),FHL1(0.37),SLC9A6(0.33),CD40LG(0.25),ZIC3(0.25),MMGT1(0.25),VGLL1(0.19),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
F4A	endocrine	0.5	No		GPR101(0.5),ZIC3(0.25)
F5A	endocrine	0.5	No		GPR101(0.5),RBMX(0.49),ARHGEF6(0.38),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),CD40LG(0.25),ZIC3(0.25),MMGT1(0.25),VGLL1(0.19),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
F6A	endocrine	0.5	No		GPR101(0.5),ARHGEF6(0.5),RBMX(0.49),HTATSF1(0.37),FHL1(0.37),CD40LG(0.33),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),VGLL1(0.19),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
F1	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
F2	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),HTATSF1(0.62),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),BRS3(0.43),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),MAP7D3(0.16),ADGRG4(0)
F3	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.5),VGLL1(0.44),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S1	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S2	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S5	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S6	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S7	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S8	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.5),VGLL1(0.44),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S9	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S10	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S11	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S13	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),HTATSF1(0.62),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),BRS3(0.43),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),MAP7D3(0.16),ADGRG4(0)
S14	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S15	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S16	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),HTATSF1(0.62),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),BRS3(0.43),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),MAP7D3(0.16),ADGRG4(0)
S18	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S19	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S20	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),HTATSF1(0.62),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.23),MAP7D3(0.16),ADGRG4(0)
S21	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.5),VGLL1(0.44),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
S22	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
II	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),VGLL1(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
III	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
IV	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
V	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
VI	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),HTATSF1(0.5),VGLL1(0.44),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)
VII	endocrine	0.83	Yes	GPR101	GPR101(0.83),ARHGEF6(0.63),RBMX(0.62),CD40LG(0.5),VGLL1(0.44),HTATSF1(0.37),FHL1(0.37),SLC9A6(0.33),ZIC3(0.25),MMGT1(0.25),BRS3(0.18),MAP7D3(0.16),ADGRG4(0)

Figure 3. POSTRE output for a non-pathogenic (F6A) duplication at the GPR101 locus.

A) Screenshot of POSTRE submission menu for the Single SV analysis mode with the F6A SV coordinates introduced. The duplication was evaluated using POSTRE in Standard mode with the “Endocrine” phenotype selected. This phenotype selection triggers the consideration of anterior pituitary-specific enhancer and expression data. B) Main POSTRE prediction summary table. It predicts the SV as non-pathogenic.

We analyzed 27 X-LAG-associated contiguous duplications at the GPR101 locus, and POSTRE predicted all 27 to be pathogenic, with each receiving a GPR101 score of 0.83 (Table 1 and Figure 2). Interestingly, one X-LAG case (patient II) had been previously described as having some unique clinical features^12,17. Typically, X-LAG is associated with pituitary gigantism due to GH hypersecretion from anterior pituitary adenoma/hyperplasia. Patient II had early-onset GH hypersecretion and overgrowth, but despite many years of detailed clinical follow-up never developed a pituitary tumor. When compared with the other X-LAG cases, the duplication in patient II is also unique. As shown in Figure 4, based on the ATAC peaks enriched in H3K27ac in pituitary cells, POSTRE predicts five enhancers (e1-e5) in the vicinity of VGLL1: four in the VGLL1-intronic enhancer (located in introns 2, 3, and 4 of the gene) and one in a more distal telomeric peak (Figure 4 A). These two enhancer loci were previously identified using publicly available, pituitary-specific, human and mouse H3K27ac ChIP-seq and ATAC-seq datasets, respectively¹². In all the analyzed X-LAG cases, except for patient II and patient S5, all five of the VGLL1 enhancers were duplicated. The duplication in patient II only involves e4 and e5, which are also the weakest enhancers based on H3K27ac levels (Figure 4B). Regarding S5, which has no atypical features of X-LAG, the duplication includes enhancers e2-e5, some of which are strong by H3K27ac levels, and only excludes the most centromeric enhancer, e1 (Figure 4). These data further emphasize the key importance of the VGLL1 intronic enhancer cluster and for the first time suggests that the X-LAG phenotype is modulated according to the cumulative strength of these ectopic enhancers acting on the GPR101 neo-TAD. In the case of patient II, the duplication might lead to a milder enhancer adoption mechanism and, thereby explain the modified clinical phenotype^17,25.

Figure 4. VGLL1-adjacent enhancers overview.

A) Overview of the VGLL1-adjacent enhancers. According to the POSTRE enhancer calling pipeline, five enhancers, that we named here e1-e5, are predicted. B) Barplot of the mean H3K27ac levels (calculated from H3K27ac ChIP-Seq bigwigs) at e1-e5 enhancers. Higher H3K27ac levels are indicators of stronger enhancer activity⁴¹.

Discussion

Structural variants altering gene-enhancer communication through TAD disruption are implicated in an emerging group of genomic disorders^3,5–7. These disorders, which alter regulatory landscapes, present a significant and growing diagnostic challenge in clinical genomics. Their investigation often requires advanced techniques like 4C-seq and HiC, which are not yet optimized for routine diagnostic workflows. Furthermore, integrating these data with other omics datasets (e.g., ChIP-Seq, RNA-Seq) demands specialized genomic and bioinformatics expertise for processing and analysis. This complexity ultimately limits diagnostic yields²⁶. In the case of X-LAG, duplications near the GPR101 locus are known to cause pathogenic rewiring of enhancer-promoter interactions^11,12. However, as shown previously and expanded on in this study, not all GPR101-spanning duplications result in disease, and data on the frequency of 3D genome rearrangements in the general population is unavailable¹⁴. In parallel, the use of genomic techniques for prenatal screening leads to the identification of SVs of uncertain significance. As not all GPR101-related duplications lead to X-LAG, this creates a challenge for clinical geneticists when counselling patients, and misclassification can lead to anxiety and potential misdiagnosis. The ability to reliably distinguish benign from pathogenic duplications at the GPR101 locus is, therefore, highly clinically relevant, especially in prenatal and pediatric contexts. Our data provide further support for the concept that pathogenicity is not solely determined by TAD disruption or duplication size, but critically depends on inclusion of specific regulatory elements, particularly the pituitary-active enhancer cluster located in the vicinity of the VGLL1 gene¹².

In this study, we demonstrate the utility of POSTRE¹⁵, an in silico tool designed to model the regulatory consequences of SVs, at the GPR101 locus. POSTRE is built upon the integration of several experimentally derived datasets, including ATAC-seq, H3K27ac ChIP-seq, and RNA-seq. These datasets, when combined with available TAD maps, allow the creation of gene-enhancer maps in disease relevant tissues and the possibility of predicting the consequences of their alteration through SVs. While chromatin conformation capture techniques such as 4C-seq and HiC provide direct evidence of three-dimensional genome organization, they are resource-intensive and impractical in routine diagnostic workflows, particularly those that are time-sensitive like prenatal testing. POSTRE complements these experimental approaches by enabling rapid, reproducible predictions of pathogenic enhancer adoption requiring only, from the user, the SV coordinates. When equipped with anterior pituitary-specific datasets, POSTRE correctly classified all known pathogenic X-LAG and non-pathogenic GPR101 duplications. Crucially, all six duplications—including three from this study and three from our recent work¹⁴—that were not associated with pituitary dysfunction were classified as benign. This predictive accuracy, coupled with ease of use, highlights its potential for robust diagnostic modeling, particularly when functional assays may be unfeasible or unavailable.

The clinical implications of these findings are especially relevant in time-sensitive contexts such as prenatal diagnosis. In multiple individuals described here and previously, chromosome Xq26.3 duplications were detected through prenatal microarray testing, and were flagged for review based on their proximity to GPR101 and its association with X-LAG. However, absence of the VGLL1-adjacent enhancers, as confirmed by POSTRE modeling, provides strong evidence against a pathogenic interpretation. This not only could aid genetic counseling but also could help to avoid unnecessary monitoring or interventions. In this context, tools like POSTRE provide an objective framework to move toward mechanistically informed variant classification of SVs.

The new pituitary-specific multi-omic dataset used in this study also uncovered important details regarding the VGLL1-adjacent enhancers that we identified previously¹². According to POSTRE enhancer calling criteria, the enhancer cluster identified within VGLL1 introns is predicted to be formed by four independent enhancers, given the existence of four independent ATAC-seq peaks enriched in H3K27ac. Further research will be needed to assess their independent contributions to GPR101 expression and their cooperation mode (e.g., additive, synergistic, hierarchical or redundant)^27,28. In addition, this cluster of VGLL1 intronic enhancers is accompanied by a single distal enhancer (Figure 4). This information supports the crucial importance of duplication of these enhancers for the generation of the X-LAG phenotype. Since early studies on the genetics of X-LAG, the VGLL1 locus, along with GPR101, formed the smallest regions of overlap (SROs) that were shared by all affected individuals^17,20 (Figure 1). Since then, we have also identified and tested a series of putative enhancers including one at RBMX (eRBMX)¹². Using POSTRE, the current study indicates the centrality of VGLL1-adjacent enhancer elements in the pathogenicity of duplications in X-LAG patients. Interestingly, a partial duplication of a subset of enhancers in patient II led to an atypical form of X-LAG, with severe excess GH hypersecretion in the absence of pituitary tumor formation. No prolactin hypersecretion occurred, which is rare in X-LAG^10,17,29, and GHRH hypersecretion, that is thought to be responsible for pituitary hyperproliferation in many X-LAG cases³⁰, was absent. This case shares many features with the pituitary somatotrope-specific transgenic mouse model that we described in Abboud et al.³¹. That model had gigantism driven by pituitary GH and IGF1 excess, that was driven by an up to 30-fold increase in pituitary Gpr101 expression (in contrast, GPR101 is increased 1000s-fold in the tumors of X-LAG patients)^12,32. Importantly, the Gpr101 transgenic mouse had a normal pituitary morphology and no evidence of increased proliferation, hyperplasia or tumorigenesis. Taken together, partial duplication of the intronic VGLL1 enhancer cluster could lead to an incomplete form of X-LAG, hypothetically due to modest, somatotrope-specific elevations in GPR101 expression. Definitive proof will require further characterization of the enhancer sequences and their functional interactions among themselves (i.e. enhancer cooperation mode), as well as their actions on the GPR101 promoter. This information would help to further refine the pathogenicity classification of SVs in this region by tools like POSTRE. These observations demonstrate that POSTRE, when equipped with cell-type specific epigenomic and transcriptomic data, can effectively discriminate between pathogenic and benign SVs, even among inter-TAD duplications. Moreover, they reinforce the functional significance of the VGLL1-adjacent enhancers in X-LAG and argue against a pathogenic role for the eRBMX CRE, despite its duplication and transcriptional enhancing activity shown in HEK293 cells¹².

Despite its clear advantages, POSTRE does have limitations. The resolution of predictions is inherently constrained by the quality and granularity of the enhancer and TAD maps it relies upon. In the current version, TAD boundaries are derived from brain prefrontal cortex data and even accounting for strong conservation of TAD boundaries across tissues, may not fully reflect the three-dimensional chromatin architecture of anterior pituitary cells. Future versions will benefit from ongoing efforts to collect and integrate HiC or Micro-C data from pituitary tissues. Additionally, although enhancer annotations were generated for this study using primary adult human pituitary epigenomic datasets, new enhancer maps derived from genomic data at other developmental stages (bulk/single-cell resolution), and improved artificial intelligence strategies to predict enhancer-promoter interactions, could further enhance sensitivity and specificity.

Beyond X-LAG, this study illustrates POSTRE’s broader potential in evaluating SVs across TADopathies affecting the endocrine system, on top of its already proven capabilities to handle this type of alterations in retinal, limb, craniofacial and neurodevelopmental disorders^15,33,34. Many such conditions are increasingly recognized as the result of disrupted regulatory domains rather than coding mutations^35–38. Incorporating tissue-relevant data into prediction algorithms allows for more nuanced interpretation of SVs that may otherwise remain unclassified when using in silico predictions tools that are agnostic to the cellular context^15,26. As databases of chromatin and enhancer landscapes expand, and as tools like POSTRE could eventually be integrated into diagnostic pipelines, this will help the field move closer to precision genomic medicine.

In conclusion, the findings presented here establish POSTRE as an effective, interpretable, and scalable tool for the clinical interpretation of SVs at the GPR101 locus. Moreover, they reinforce the mechanistic model whereby duplications need to span both the GPR101 TAD boundary and include the VGLL1-adjacent enhancers to drive GPR101 misexpression in X-LAG.

Ethics approval and consent to participate

Subjects were recruited under the University of Liège Ethics committee approved protocol B707201420418; under the Bioethics Committee of Nicolaus Copernicus University, Toruń, Poland (NCU Committee of Bioethics KB 61/2021); and under the Eunice Kennedy Shriver National Institute of Child Health and Human Development, National Institutes of Health protocol 97-CH-0076 (ClinicalTrials.gov: NCT00001595). The study was approved by the Independent Ethics Committee of the IRCCS Humanitas Research Hospital in Rozzano (Milan, Italy) and conformed to the ethical guidelines of the Declaration of Helsinki (approval no. 642/20). Written informed consent was obtained from all the subjects/guardians.

Funding Statement

The work was supported in part by the following funding sources: Fondazione Telethon, Italy grant no. GGP20130 (to GT, supporting AG); grants from the Fonds d’Investissment pour la Recherche Scientifique 2018-2023 of the Centre Hospitalier Universitaire de Liège and grant number FSR-F-2023-FM from the Faculty of Medicine, University of Liège; Intramural Research Program, Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), National Institutes of Health (NIH) Research project Z1A HD008920 (to CAS), USA. The project that gave rise to these results received the support of a fellowship from “La Caixa” Foundation (ID 100010434). The fellowship code is LCF/BQ/PR22/11920006 (to MF). GT would like to acknowledge financial support by the Italian Ministry of University and Research (grant #MSCA_0000055).

Availability of data and materials

The datasets generated during this study are available on request.