From incomplete penetrance with normal telomere length to severe disease and telomere shortening in a family with monoallelic and biallelic PARN pathogenic variants

2.1 Human subjects

This investigation was conducted according to the Declaration of Helsinki Principles. Informed consent was received from participants before inclusion in the study according to protocol H-17698 Genetic and Biological Determinants of Bone Marrow Failure approved by the Institutional Review Board for Baylor College of Medicine and affiliated hospitals.

2.2 Telomere length analyses

Telomere flow FISH analysis was performed on peripheral blood leukocytes by Repeat Diagnostics (Vancouver, BC) as previously described (33). Telomere restriction fragment length analysis of HCT116 PARN wild type (WT) and heterozygous cell lines was performed by Southern blot as previously described (Nelson et al., 2018). Mean telomere length was determined using TeloTool V1.3 (Gohring, Fulcher, Jacak, & Riha, 2014). In addition, telomere length of the cell lines was analyzed by quantitative telomere FISH as previously described (Ourliac-Garnier & Londono-Vallejo, 2017). Metaphase chromosomes were prepared by treatment of cells with colcemid (Sigma-Aldrich) 0.1 µg/ml for 30 min at 37°C. FISH was performed as previously described (Ourliac-Garnier & Londono-Vallejo, 2017) using TelG Alexa Fluor 647 (PNA Bio; catalog # F1014). DNA was counterstained with 4′,6-diamidino-2-phenylindole (DAPI) in SlowFade Diamond Antifade Mountant (Life Technologies, catalog # S36964). Images were captured using a GE Healthcare DVLive epifluorescence image restoration microscope with an Olympus PlanApo ×100/1.4 NA objective and a 1.9k × 1.9× sCMOS camera. Z stacks (0.24 µm) covering entire metaphases before applying a conservative restorative algorithm for quantitative image deconvolution. Exposures for DAPI and Alexa Fluor 647 were set on the control sample and maintained constant within each set of samples. Maximum intensity projections were generated and used for image analysis. Telomere length intensity was calculated using the NIH software ImageJ (https://imagej.nih.gov/ij/) with the plugin Telometer version 3 (http://demarzolab.pathology.jhmi.edu/telometer).

2.3 DNA sequencing and data analyses

Whole exome sequencing was conducted on the proband and parents according to standard protocols at Baylor Genetics. Reinterpretation of exome data was conducted in the research setting using GATK software for calling variants and custom perl scripts for interpretation of variants that annotate and prioritize variants based on multiple sources of information including SIFT, PolyPhen, GERP, PhyloP, CADD, ExAC, UniProt, OMIM, IntAct as previously described (Besse et al., 2015; Stiles et al., 2015).

For orthogonal sequence validation and genotyping of PARN alleles in additional family members, genomic DNA was isolated from whole blood and the regions corresponding to the PARN p.Y91C and p.(I274*) mutations, the most common TERT promoter variants, and hsa-mir-202, were PCR amplified. The PCR products were analyzed by Sanger sequencing (Eurofins Genomics, LLC). To determine the sequence of the insertion allele, before Sanger sequencing, the amplicon encompassing PARN p.(I274*) was cloned using the Zero Blunt™ TOPO™ PCR cloning kit. See Table S2 for oligonucleotide sequences.

2.4 Deadenylation assays

PARN deadenylation assays were performed using recombinant proteins as previously described (Henriksson, Nilsson, Wu, Song, & Virtanen, 2010). The kinetic analysis was performed using PARN WT (100 nM) or p.Y91C (30 nM) and A₃ trinucleotide substrate. For determining the rate constants under single turnover conditions, each reaction contained 10 nM RNA substrate and 30 nM PARN. The reactions were incubated at 30°C for the indicated length of time and analyzed as described above. The mathematical model of multistep reactions with intermediate products is derived in Kuriyan, Konforti, and Wemmer (2013). The k_obs1 and k_obs2 parameters were determined from time course degradation assays of di- or trinucleotide substrates with subsequent curve fitting in Prism 7 (GraphPad) with the intermediate product equation, . The given values are the mean ± standard deviation (SD) of at least three independent time-course experiments.

2.5 Lymphoblastoid cell lines

LCLs were generated by the tissue culture core laboratory within the Department of Molecular and Human Genetics, Baylor College of Medicine. LCLs were cultured in RPMI-1640 medium containing l-glutamine (Invitrogen) and 10% fetal bovine serum (FBS) in 5% CO₂.

2.6 Protein extraction and analysis

Cells were lysed in RIPA buffer (50 mM Tris-HCl pH 8.0, 150 mM NaCl, 1% Triton X-100, 0.5% Na deoxycholate, and 0.1% sodium dodecyl sulfate) with 1× protease inhibitor cocktail III (Calbiochem) and centrifuged. The protein concentration of the supernatant was determined using the bicinchoninic acid protein assay kit (Pierce). A quantity of 20–40 μg of protein were loaded per sample. Protein was detected using polyclonal rabbit anti-PARN antibody (Bethyl; A303–562A) and β-actin (Sigma-Aldrich; A5441).

2.7 RNA analyses

To determine if RNA encoding PARN p.(I274*) underwent nonsense-mediated decay, RNA was isolated from whole blood using the RNeasy® kit (QIAGEN) and complementary DNA (cDNA) synthesized using the qScript Flex cDNA Synthesis Kit (Quanta Biosciences, Beverly, MA). The region surrounding PARN p.Y91C was PCR amplified and analyzed by Sanger sequencing.

PARN gene expression levels were measured by qPCR using a QuantStudio 6 Flex (Applied Biosystem) with PowerUp™ SYBR Green Master Mix (Applied Biosystems, Austin, TX) in triplicate and normalized to glyceraldehyde 3-phosphate dehydrogenase (GAPDH). For these experiments, total RNA was extracted from LCLs using the RNeasy® kit (QIAGEN) and cDNA synthesized using the qScript Flex cDNA Synthesis Kit (Quanta Biosciences, Beverly, MA) and oligo(dT) primer.

To analyze snoRNAs and scaRNAs, total RNA, including small RNAs, was isolated using the miRNeasy® kit (QIAGEN). cDNAs were prepared with 1–2 µg of RNA using qScript and random or oligo(dT) primer. qPCR was performed using gene-specific primers and PowerUp™ SYBR Green Master Mix (Applied Biosystems™) or Maxima SYBR green master mix (Thermo Fischer Scientific) according to the manufacturer's protocol using the ABI 7300 real-time PCR instrument. The relative changes in RNA abundance were calculated by the comparative ΔC_t method with normalization to GAPDH.

To analyze rRNA by northern blot, total RNA was extracted using the RNeasy® kit (QIAGEN). A quantity of 5 µg of total RNA per sample were separated on a 1.2% agarose/formaldehyde gel and capillary transferred to a Zeta-Probe cationized nylon membrane (BioRad; 1620165). The membrane was prehybridized in ULTRAhyb™ buffer (Invitrogen; AM8663) for 45 min at 42°C and then hybridized with γ³²P-ATP end-labeled ITS1–59 probe in ULTRAhyb™ buffer overnight at 65°C. The membrane was exposed to a phosphorimager screen for 4 hr. Signals were quantified using ImageQuant TL 8.2 image analysis software (GE Healthcare). Ratio analysis of multiple precursors was performed as previously described except that the average values of three controls were used as control values for the analysis of LCLs rather than the values of a single control (Wang, Anikin, & Pestov, 2014).

2.8 CRISPR-Cas9 guide design and heterozygous PARN cell line generation

The guide RNAs (GGAGGAGCTGAATGATGCTG and CTCACTTACCGAATTAGCAA) were identified using the online software Benchling (www.benchling.com/). The PARN gene sequence (ENSG00000140694) was uploaded and the search for possible guides was performed around exon 12. The guides in plasmid pSpCas9 BB-2A-GFP (Ap311 and Ap312, respectively) and the scrambled synthetic guide RNA (sgRNA) in pSpCas9 BB-2A-GFP (Ap314) were obtained from GenScript. HCT116 cells were grown in McCoy's 5A medium with 10% FBS and seeded in a six-well plate at a concentration of 90,000 cells/well. Cells were transfected at 50% confluency with 0.5 µg of Ap311 and Ap312 using Lipofectamine Plus Reagent (Invitrogen) following the manufacturer's protocol. After 72 hr, the cells were harvested for analysis using the Guide-it Mutation detection Kit (Takara Clontech; catalog # 631443), which uses a resolvase to determine cutting efficiency. Once the efficiency of cleavage was confirmed, the experiment was repeated as described. After 72 hr, green fluorescent protein-positive cells were sorted via flow cytometry and single cells collected in 92-well plates. Clones from single-cell growth were split into two sister plates. One was cultured while the other was used to identify heterozygous PARN clones by Sanger sequencing. Clones with abnormal chromatograms underwent Zero Blunt™ TOPO™ PCR cloning and Sanger sequencing to determine the genotype. Three clones that underwent transfection but still maintained WT genotype (WT-C1, -C2, and -C3) and one PARN-het clone were used in subsequent experiments.

2.9 Transfection with miR mimic RNA

HEK293T were grown in modified Eagle's media-α with 10% FBS and 5% CO₂. Cells were seeded in six-well plates and, when at 60% confluence, transfected with 0.1 µM Pre-miR™ miRNA Precursor hsa-miR-202-5p mimic (Ambion; AM17100), 0.1 µM pre-miR™ miRNA Precursor random miR mimic (Ambion; AM17111), or transfection reagent alone (DharmaFECT #1; 6 µl in 2 ml). The cells were harvested at 72 hr for western blot analysis.

2.10 Statistical analyses

The data are represented as mean ± SD of the values. A p < .05 was considered statistically significant and indicated by an asterisk. For the PARN western blot and qPCR, hTR RH qPCR, and hTR dT qPCR, each biological replicate was normalized to one control individual (128S2 for PARN western and qPCR, RQ4115 for hTR RH and dT qPCRs). One-way analysis of variance with Dunnet's comparisons were performed on log transformed data for those data sets having a lognormal rather than normal distribution. The log-transformed data were normally distributed. Brown–Forsythe tests showed that the SDs were not significantly different between individuals for the PARN western blot or qPCR, or the hTR RH qPCR but were for the hTR dT qPCR. Dunnet's post-hoc test was performed and p values adjusted for multiple comparisons. For the H/ACA snoRNA profile, the average of the control individuals (RQ4115, 101F, and 128S2) was used for normalization and a t test was performed to determine significant differences from the average of controls. For the assays using HCT116 clonal cell lines, the Shapiro–Wilk test was applied to check for normal distribution and the Bartlett test to check for equal variance across samples.

3.1 Compound heterozygous PARN variants segregating with HH in family BMF22

The proband of this study was a male diagnosed with HH, based on clinical features of bone marrow failure, intellectual disability, cerebellar atrophy, and oral leukoplakia, and lymphocyte telomere length <1st percentile (Figures 1a,b and S1 and Table S3). He was 17-years-old at the time of this report. Whole-exome-sequencing (WES) identified two variants in PARN, a variant of uncertain clinical significance, which mapped to the CAF1 ribonuclease domain (NM_002582.3: c.272A>G; p.Y91C; Figures 1c and S2A), and a likely pathogenic 20 bp exonic insertion resulting in an in-frame translation stop codon (NM_002582.3: c.819_820insTAGAAATCATTTCTAGAGTC; p.(I274*); Figures 1c and S2). PCR amplification of the region, blunt-ended cloning, and Sanger sequencing confirmed the sequence of the insertion (Figure S2B). A portion of the insertion shared microhomology with a nearby 5′ region, suggesting that it was caused by a duplication event.

Concurrent with these investigations, his younger sister, S2, was also diagnosed with HH based on developmental delays, microcephaly, cerebellar atrophy and telomere length <1st percentile (Figures 1a,b, S1 and Table S3). She was 4-years-old at the time of this report, at which time she had normal peripheral blood counts and indices, except for a slightly decreased white blood cell count (4.91 × 10³/µl; lower limit of the normal range for age 5 × 10³/µl). She was also found to carry both PARN variants (Figures 1a and S2A). Targeted sequencing of the parents and grandparents demonstrated the inheritance of the missense p.Y91C variant through the paternal grandmother (GP2), and inheritance of the p.(I274*) insertion through the maternal grandmother (GP4; Figures 1a and S2A). The proband's other sibling, S1, inherited only the missense variant. Only those family members who carried both mutant alleles (the proband and S2) were symptomatic (Figures 1a and S2A). Siblings of F and M, who were reportedly healthy, are not noted in the pedigree as they were unavailable for analysis with the exception of one sister, BMF22-A, who had limited studies performed (mutation analysis demonstrating the absence of the familial PARN variant and telomere length analysis, which was within normal range; data not shown).

3.2 Both PARN p.(I274*) and p.Y91C are damaging

Next, we sought further evidence for the pathogenicity of these variants. We predicted that the RNA encoding the p.(I274*) would undergo nonsense-mediated decay (NMD) because the premature termination codon introduced by the insertion was well upstream of the final exon–exon junction (Figure 1c). To test this, we compared Sanger sequencing of the genomic DNA (gDNA) and cDNA at position c.272, within the codon for residue Y91, in the proband, S1, S2, and F, all of whom were heterozygous for the c.272A>G transition. As expected, we found that the reference (A) and variant (G) bases were both clearly demonstrated in the chromatograms of F and S1's gDNA and cDNA, and the proband and S2's gDNA (Figure 2A). In contrast, the reference base (A) was barely demonstrated in the chromatograms of the cDNA of the proband and S2, consistent with a specific depletion of the p.(I274*) transcript from the maternally inherited allele by NMD.

To investigate if the p.Y91C mutation, which mapped near the dimerization interface (Figure 2b), affected the deadenylation activity of PARN, we generated a corresponding mutant PARN polypeptide, PARN p.Y91C, by site-directed mutagenesis, and functionally characterized its deadenylation activity in vitro by incubating it with a 20 nucleotide long homopolymeric adenine substrate (A₂₀). Figure 2c shows that its deadenylation activity was at least 10-fold lower than for the WT PARN polypeptide. We also noted that the deadenylation pattern, that is, accumulation of A₄ and A₂ deadenylation intermediates (see Henriksson et al., 2010), was not affected when using the mutant PARN p.Y91C polypeptide.

To further analyze the PARN p.Y91C mutant protein, we determined its kinetic rate constants, k_obs1 and k_obs2, for the complete degradation of an A₃ trinucleotide substrate. The k_obs1 rate constant describes the release of the 3′-end located adenine residue from the A₃ substrate, while the k_obs2 rate constant describes the final hydrolysis of the A₂ reaction intermediate (see Materials and Methods section for definitions of k_obs1 and k_obs2 rate constants). From this analysis, we found that the cysteine substitution reduced both kinetic rate constants, k_obs1 and k_obs2, approximately 30-fold (Table 1, catalytic efficiency ratio). This drop was observed both in reactions performed with Mg²⁺ or Mn²⁺ as the divalent metal ion. This showed that the Y91C mutation affected the entire hydrolytic cycle of PARN, which encompasses both one translocation and one hydrolytic event for each round of hydrolysis (Virtanen et al., 2013). Together these studies showed that the Y91C mutated PARN polypeptide was perturbed in its deadenylation activity and suggest that the PARN-mediated deadenylation activity in individuals in this pedigree carrying the mutation was affected.

3.3 The same monoallelic PARN variants do not always associate with short telomeres and may be associated with telomere lengthening over time

Having established that both variants were damaging and cosegregated with disease (Figure 1a), we evaluated their effects on telomere length through telomere flow-FISH analysis of peripheral blood leukocytes in BMF22 family members. In Figure 1b, we present the length of the total lymphocyte population, although the trends were similar in granulocytes and subpopulations of lymphocytes (Figure S3). In the initial set of analyses, among the p.Y91C variant monoallelic carriers (F, S1, and GP2), the lymphocyte telomere lengths in F and S1 were between the 1st and 10th percentiles, whereas the lengths in GP2 were at the 50th percentile, apparently unaffected. Strikingly, M, who carried the p.(I274*) allele subject to NMD, had lymphocyte telomere lengths around the 50th percentile. Thus, the monoallelic variants were not always associated with short telomeres, consistent with prior reports (Dhanraj et al., 2015; Moon et al., 2015; Stuart et al., 2015; Tummala et al., 2015).

Given these results, we next sought to determine leukocyte telomere lengths in the monoallelic PARN variant carriers over time (Figures 1b, S3 and Table S3). Telomere flow FISH was performed on M, 8 years after her initial testing. Telomere lengths shortened in pan-lymphocytes and each of the lymphocyte subsets, as anticipated, with the percentiles declining to varying extents, although in all cases remaining greater than the 10th percentile and in the memory T and B cells above the 50th percentile.

The results of repeat telomere length analyses of S1 and F, both of whom carried the p.Y91C allele, were different (Figures 1b and S3A and Table S3). F's second sample, drawn 8 years after the first, surprisingly showed increased telomere length in four of the six subsets (pan-lymphocytes, granulocytes, CD45RA+ [naïve T] cells, and CD57+ [NK] cells), with average increases of 63–239 bp/year. Despite the increases, F's telomere lengths remained in the 1st–9th percentile range except in the naïve T and NK cell populations, which incurred the greatest increases, resulting in lengths in the >10th and <50th percentile range. S1's granulocyte and NK cell telomere lengths also increased over a 7 year interval, although to a lesser extent than F's (averages of 66 and 69 bp/year in granulocytes and NK cells, respectively), and his naïve T cell telomere lengths declined only 50 bp total in the 7 year interval compared to a decline of 610 bp in the median (87 bp/year). Nonetheless, S1's telomere length percentiles remained slightly greater than F's in all subsets except the NK cells.

We next asked if M's normal telomere length and F and S1's increases in telomere lengths were due to a previously reported phenomenon: somatic TERT promoter mutations in individuals with germline monoallelic PARN or TERT variants (Maryoung et al., 2017). These TERT promoter mutations (NC_000005.10(TERT_v002):c.-146C>T and NC_000005.10(TERT_v002):c.-124C>T) are associated with increased TERT expression, telomerase activity, and proliferative capacity (Borah et al., 2015; Chiba et al., 2015). However, there was no TERT promoter variant detected in M, F or S1 (Figure S3B).

3.4 The same biallelic PARN variants may be associated with accelerated telomere shortening or stable, very short telomere lengths over time

The proband and S2 shared biallelic PARN variants and both had severely short telomeres (<1st percentile for age), however, the proband had much shorter telomeres than S2, either objectively or adjusted for age at the time of diagnosis (Figures 1b, S3A and Table S3). This, too, was surprising, so, as with other family members, we performed repeat testing. The proband's lymphocyte telomeres were stably short (maximum 4.4 and minimum 3.8 kb over an approximately 9-year span), while S2's lymphocyte telomeres shortened dramatically from 7.5 to 5.9 kb in 3 years, a rate greater than expected based on the normal curves. At each of her time points, a complete blood count was obtained, which remained within normal range. Additionally, she reported no significant infections during the time frame. Subsequent testing at 4 years-of-age yielded telomere lengths at 6.1 kb, suggesting a potential stabilization of the rate of attrition.

3.5 PARN target RNAs are altered in one individual with biallelic variants but not the other

In light of the variable telomere length of variant carriers, we proceeded to assess the cellular phenotypes of LCLs derived from BMF22 family members and multiple unique LCL controls. PARN protein level was variable upon biological replicates, even in control LCLs (Figures 3a and S4). Nonetheless, when we compared all other controls and all family members to control BMF128-S2, PARN was significantly reduced only in the proband, M, and GP4 (p < .0001, .0002, and <.0001, respectively), as expected since they each carried the insertion (Figure 3b). Surprisingly, despite carrying the insertion, S2's PARN level was not consistently nor significantly reduced among repeated experiments (p = .7199; Figures 3 and S4). The higher PARN protein level in S2 as compared with the other family members carrying the p.(1274*) could not be attributed simply to a higher PARN mRNA level as, while PARN mRNA level, assessed by quantitative reverse-transcriptase PCR (qRT-PCR), was significantly lower in M and G4 compared to S2, the level in the proband was comparable to that of S2 (Figure 3c). Sequence analysis of the 5′- and 3′-untranslated regions (UTR) of the PARN locus for all members of the BMF22 family revealed no variants to account for these differences (data not shown).

Next, we assessed various RNA targets of PARN using qRT-PCR. It was reported previously that DC/HH patients with biallelic PARN variants had increased hTR adenylation and decreased total hTR (Dhanraj et al., 2015; Moon et al., 2015; Tummala et al., 2015). In particular, LCLs derived from a DC/HH patient had reduced hTR level (Tummala et al., 2015). We, therefore, expected to find alterations in the proband and S2. Compared with control RQ4115, only the proband had significantly increased adenylated hTR as measured by qPCR (p = .0003; Figure 4a), whereas both the proband and S2 had significantly reduced hTR (p = .001 and .0248, respectively; Figure 4b).

In addition to hTR, PARN deadenylates multiple snoRNAs and scaRNAs (Berndt et al., 2012) and a patient, reported by Dhanraj et al. (2015), with biallelic PARN variants and DC/HH was found to have an aberrant snoRNA expression profile. To determine if this was also the case in LCLs derived from BMF22 family members, we estimated the abundance of the total and adenylated levels of H/ACA type and C/D box type snoRNAs and scaRNAs previously found to be impacted by PARN knockdown (Berndt et al., 2012) or in the patient with biallelic PARN variants reported by Dhanraj et al. (2015). When compared with an average of controls, the proband had multiple H/ACA snoRNAs demonstrating a marked enrichment of the adenylated form (up to approximately 90-fold) and a slight reduction of abundance (none with less than approximately 0.9-fold; Figure 4c). In contrast, the snoRNA profile of other family members, including the biallelic carrier S2, was similar to that of controls. The overall snoRNA expression profile of the proband was consistent with the previously reported patient. Of the snoRNAs analyzed, the three with the greatest enrichment of adenylated species were the same three with the highest enrichment in the previously reported patient, and each was significantly increased compared to an average of controls: SNORA9, SNORA63, and SCARNA8 (p = .003, .02093, and .0036, respectively).

The above patient studied by Dhanraj et al. (2015) was also suspected to have a defect in ribosome biogenesis because H/ACA snoRNAs are involved in the chemical modification of pre-rRNA and a reduction of the mature 80S ribosome was found. To further explore the impact of PARN mutation on ribosome biogenesis, we specifically looked at the levels of pre-rRNAs in BMF22 family member LCLs. A single pre-RNA molecule undergoes multiple steps of cleavage, releasing precursors to ultimately generate the mature rRNAs, including 18S rRNA (reviewed in Henras, Plisson-Chastang, O'Donohue, Chakraborty, & Gleizes, 2015). These precursors undergo trimming during maturation. The pre-rRNA 18SE-FL is trimmed to 18SE and then to 18S. Analysis of siRNA PARN knockdown cells demonstrated that PARN trims 18SE-FL, facilitating 18S maturation (Ishikawa et al., 2017; Montellese et al., 2017; Tafforeau et al., 2013). Therefore, we performed northern blot analyses using the ITS1-59 probe and Ratio Analysis of Multiple Precursors (RAMP; Wang et al., 2014) to assess rRNA precursors in LCLs derived from BMF22 family members as compared to control LCLs. The proband LCLs, but not other family member LCLs, had an increase of untrimmed 18SE-FL/PTP and 18SE-FL/21 S ratios compared to an average of controls (Figure 4d–f). These results support the finding that PARN trims 18SE-FL and could contribute to a ribosome defect in patients.

3.6 Editing of a single PARN allele is sufficient to shorten telomeres but not to reduce hTR

We were surprised that LCLs derived from monoallelic variant carriers, including M with reduced PARN protein level, did not have altered PARN function as measured by hTR and H/ACA snoRNA adenylation and rRNA maturation. Therefore, we used CRISPR-Cas9 to generate HCT116 cells with a loss-of-function (LOF) mutation in one allele of PARN to model monoallelic PARN variants in an isogenic setting. Initially, we planned to knock-in the insertion allele, and we, therefore, identified guides that efficiently targeted exon 12 of PARN (Figure S5). Ultimately, we isolated a clone harboring a single c.824_840delinsTTCC mutation following transfection with the sgRNA-expressing plasmid Ap312 (see Materials and Methods section). Sanger sequencing demonstrated that the other allele was WT at this locus. This heterozygous PARN LOF cell line had a reduction in PARN protein level as expected compared with untransfected cells and WT clones that underwent the same transfection but were unedited (Figure 5a,b). Consistent with haploinsufficiency, telomeres shortened in the heterozygous PARN LOF cell line as analyzed by telomere restriction fragment analysis and quantitative FISH (qFISH; Figure 5c,d). Notably, the level of adenylated hTR increased in the heterozygous PARN LOF cell line, however, the level of hTR was unchanged (Figure 5e,f). Thus, a single LOF mutation can result in altered hTR processing and impaired telomere length maintenance as observed in some, but not all PARN variant carriers. Notably, single LOF mutation did not impact the trimming of 18SE-FL as the untrimmed 18SE-FL/PTP and 18SE-FL/21S ratios were unchanged in the heterozygous PARN LOF cell lines compared with the controls. These results suggest differential effects on hTR versus ribosomes in patients with monoallelic PARN variants.

3.7 miR-202-5p as a regulator of PARN

A possible etiology of the incomplete penetrance in PARN heterozygous variant carriers is a relatively common genetic modifier that influences PARN protein level. One such modifier could be a variant in a miRNA. To investigate this possibility, we utilized TargetScan (http://www.targetscan.org/, release 7.2, default parameters and program options; Agarwal, Bell, Nam, & Bartel, 2015) to identify miRNAs that might target PARN mRNA. This analysis yielded a single, conserved miRNA, miR-202-5p, with positions 2–8 of the miR-202-5p seed region complementary to nucleotides 1,010–1,016 of the PARN 3′-UTR (Context++ score percentile 96; Figure 6a). To determine if miR-202-5p could downregulate PARN protein levels, we transfected HEK293T cells with a double-stranded RNA (Ambion Pre-miR™ miRNA Precursor) designed and chemically modified such that the strand representing miR-202-5p is preferentially utilized in the RNA-induced silencing complex, a Pre-miR™ negative, random sequence control, or transfection reagent alone. In contrast to mock transfected and cells transfected with the negative control Pre-miR™, PARN protein level was significantly lower in cells transfected with mir-202-5p precursor when compared with the untransfected control (p = .02; Figure 6b,c), suggesting miR-202-5p could influence protein levels in humans.

The SNP rs12355840 (NC_000010.11:g.133247608C>T) maps to the stem-loop structure of the hsa-pre-mir-202 and is removed during maturation to the mature miR202-5p (Hoffman et al., 2013; Figure 6d). Both the C (reference) and T (variant) alleles are commonly observed in the Genome Aggregation Database (gnomAD; http://gnomad.broadinstitute.org/; Lek et al., 2016), with frequencies of 0.215 and 0.785, respectively. Interestingly, the presence of C negatively impacts miR-202-5p levels compared to T, likely by destabilizing the stem-loop structure and impairing the maturation of the precursor mir to the mature miR-202-5p (Hoffman et al., 2013). Therefore, we considered the possibility that the MIR202 genotype at this position might be a genetic modifier, with the C allele lessening the impact of a given PARN variant on telomere length by lowering miR-202-5p levels and, secondarily, increasing PARN protein levels relative to the T allele. However, because miR-202-5p appears to be predominantly expressed in the male reproductive system (Landgraf et al., 2007), a given MIR202 genotype may primarily impact the telomere length in sperm or during very early development and this may be reflected solely in a father's offspring. Because the proband and S2 shared the same father, it is unlikely that mir202-5p contributed to the variation in their telomere lengths. Similarly, as GP1 (F's father) and GP3 (M's father) both carried the CT alleles, it is unlikely that miR202-5p contributed to the shorter telomeres observed in F than M, even though M had an NMD-inducing mutation.

4.1 Normal hTR levels in LCLs derived from monoallelic PARN variant carriers and in a heterozygous PARN LOF cancer cell line model

Total hTR levels remained intact in LCLs derived from the monoallelic PARN variant carriers with not only normal but also short telomere length, which was surprising. It is possible that hTR could be reduced in hematopoietic stem cells, lymphoid progenitors, or peripheral blood lymphocytes and LCLs may not be representative of this or that a biologically relevant but small difference in RNA level is not detectable. However, it is also notable that, in the cancer cell line with a single PARN allele mutated, HCT116 PARN LOF, hTR adenylation was increased but hTR total level was normal. Nonetheless, telomeres shortened. It is possible that monoallelic loss of PARN causes telomere shortening through an RNA other than hTR in HCT116 cells, such as that encoding TPP1, as observed in PARN-deficient HT1080 cells (Benyelles et al., 2019) or that increased hTR adenylation is sufficient to shorten telomeres.

4.2 PARN deficiency impacts rRNA maturation

PARN deficiency was initially proposed to alter ribosome maturation in patients through its impact on snoRNAs (Dhanraj et al., 2015), which are involved in the chemical modification of pre-rRNA. More recently, it was found that PARN directly trims rRNA precursors and that RNAi depletion of PARN causes their accumulation (Montellese et al., 2017). In addition, while this manuscript was in revision, Benyelles et al. (2019) reported accumulation of 18S pre-rRNAs in samples derived from HH individuals, as we do here in our proband (Figure 4d–f), and in a cell line with biallelic PARN knockout. Interestingly, we observed that neither cells with a knockout of a single allele (Figure 5g) nor individuals heterozygous for pathogenic PARN variants (Figure 4d–f) had an accumulation of 18S pre-rRNA, suggesting a dose-dependent effect. Short telomeres alone appear to be sufficient to lead to an HH phenotype given that among the genes that are associated with HH are multiple involved in telomere maintenance but not firmly implicated in ribosome biogenesis or function (e.g., TERT, TINF2, and RTEL1). Nonetheless, these data contribute to the larger question in the field of inherited bone marrow failure syndromes of whether altered ribosomes contribute to DC and HH (Nakhoul et al., 2014).

4.3 Regulation of PARN levels

PARN protein was higher in S2 LCL's than in those of the proband, M, or GP4, all of whom also carried the p.(I274*) variant (Figure 3a,b) and this could not be attributed to elevated PARN mRNA specifically in S2. Indeed, PARN mRNA levels were comparable in S2 and the proband but reduced in M and GP4 (Figure 3c). The mechanisms underlying these differences remain unclear. Nonetheless, these results underscore the complexity of PARN regulation and the potential impact of PARN variants on telomere length.

We found that exogenous miR-202-5p resulted in a decrease in PARN protein, consistent with miR-202-5p target sequence in the PARN mRNA 3′ UTR (Figure 6a–c) and suggesting this miRNA might contribute to the regulation of PARN levels in vivo. Because hsa-mir-202 is predominantly expressed in the male reproductive system (Landgraf et al., 2007), it may have its greatest impact on PARN and telomere lengths in male gametes and during the earliest stages of development. While the limited analysis afforded by the small BMF22 family suggests rs12355840 is not a genetic modifier of telomere lengths in the context of PARN mutations, further analysis of larger, multigeneration families may be needed to draw stronger conclusions. Incorporation of paternal age at conception, which has been shown to impact telomere length in offspring (Eisenberg & Kuzawa, 2018), is another variable worth consideration in such larger studies.

4.4 A final cautionary note

The PARN insertion encoding p.(I274*) was initially missed in clinical WES analyses of the proband and S2, but was ultimately uncovered due to a high degree of suspicion that biallelic PARN variants were the culprit. Clinicians and investigators with unsolved cases of patients for which disease genes are known must remain cognizant of the limitations of WES, including the challenge of calling in/dels. In addition, the allele frequency for the PARN c.272A>G; p.Y91C variant in gnomAD is as high as 0.0002115 in Africans and 0.00007992 for the entire cohort. While still very low, the number of individuals with this variant in gnomAD, 22, was notably higher than that for any of the 64 unique variants reported in a 115 member IPF cohort, which included 43 TERT, six TERC, seven RTEL1, and eight PARN variants (Newton et al., 2016). In that cohort, the number of individuals in gnomAD carrying a given variant ranged from 0 to 7 (median 0), excluding the TERT p.K1050N variant (NM_198253.2:c.3150G>C), which was present in 23 individuals, 20 of which were of Ashkenazi Jewish ancestry. Therefore, functional analyses should be considered for PARN variants of uncertain clinical significance even if present at higher than expected allele frequencies.