Functional Assays Are Essential for Interpretation of Missense Variants Associated with Variable Expressivity

Abstract

Missense DNA variants have variable effects upon protein function. Consequently, interpreting their pathogenicity is challenging, especially when they are associated with disease variability. To determine the degree to which functional assays inform interpretation, we analyzed 48 CFTR missense variants associated with variable expressivity of cystic fibrosis (CF). We assessed function in a native isogenic context by evaluating CFTR mutants that were stably expressed in the genome of a human airway cell line devoid of endogenous CFTR expression. 21 of 29 variants associated with full expressivity of the CF phenotype generated <10% wild-type CFTR (WT-CFTR) function, a conservative threshold for the development of life-limiting CF lung disease, and five variants had moderately decreased function (10% to ∼25% WT-CFTR). The remaining three variants in this group unexpectedly had >25% WT-CFTR function; two were higher than 75% WT-CFTR. As expected, 14 of 19 variants associated with partial expressivity of CF had >25% WT-CFTR function; however, four had minimal to no effect on CFTR function (>75% WT-CFTR). Thus, 6 of 48 (13%) missense variants believed to be disease causing did not alter CFTR function. Functional studies substantially refined pathogenicity assignment with expert annotation and criteria from the American College of Medical Genetics and Genomics and Association for Molecular Pathology. However, four algorithms (CADD, REVEL, SIFT, and PolyPhen-2) could not differentiate between variants that caused severe, moderate, or minimal reduction in function. In the setting of variable expressivity, these results indicate that functional assays are essential for accurate interpretation of missense variants and that current prediction tools should be used with caution.

Introduction

Determining the phenotypic consequences of DNA variants in genes associated with disease is a major goal for genomic medicine.¹ Variants in the coding region of genes can have a variety of consequences that can affect RNA quantity or processing or can alter the sequence of the encoded protein. Missense changes account for ∼38% of variants implicated in single-gene disorders² and are particularly challenging to interpret because they can produce a broad array of effects, ranging from loss of protein due to severe instability to no discernible consequence. Accordingly, missense variants can create a spectrum of phenotypic consequences that encompass both variable expressivity and incomplete penetrance of clinical features that constitute Mendelian disorders.³ Interpreting whether missense variants are responsible for partial expressivity of single-gene disorders is a major challenge in the clinical and research setting. Indeed, genetic testing is frequently requested to help diagnose individuals with incomplete features of a Mendelian condition. However, labeling missense changes as variants of unknown significance (VUS) because of a lack of functional information does not resolve diagnostic dilemmas.

Expression of mutants in heterologous cell lines provides a versatile method for assessing the functional effect of a wide range of variants.^{4, 5} The American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) have recognized the importance and utility of heterologous expression systems and have weighted the results of well-established functional testing heavily within their recommended pipeline for variant classification for all Mendelian diseases.⁶ These groups emphasize that studies should be validated, reproducible, and robust and are most helpful when they are reflective of the biological environment in which a variant operates. Importantly, the ACMG-AMP guidelines recommend that interpretation of functional testing occur within the context of other available information to avoid placing too much emphasis on data that, although valuable, might not completely reflect a variant’s behavior in vivo.

Cystic fibrosis (CF [MIM: 219700]) is caused by variants that lead to reduced function of CF transmembrane conductance regulator (CFTR [GenBank: NP_000483.3]). Full expressivity of CF manifests as dysfunction of epithelial tissues in the lungs, pancreas, and sweat duct.⁷ However, not all features are consistently present, thereby giving rise to phenotypes (exhibiting partial expressivity) that can be difficult to differentiate from other causes of lung and/or pancreatic disease.⁸ Missense variants are commonly found in association with partial expressivity of CF, thereby providing an opportunity for researchers to evaluate the utility of functional assessment in this situation.⁹ Because protein-folding studies, which are commonly performed, are poor predictors of overall function,⁴ we chose to evaluate chloride conductance to determine the functional consequences of 48 CFTR (MIM: 602421; GenBank: NM_000492.3) missense variants reported in individuals that exhibit full and partial expressivity of CF. We show that functional assessment informs variant annotation when full or partial expressivity is present and that uncertainty of variant effect can be re-interpreted in the context of variable expression of a phenotype. Finally, we illustrate the limitations of algorithms that predict the functional consequences of missense variants.

Material and Methods

Selection of CFTR Variants

CFTR variants for study were selected from de-identified demographic and clinical data collected by the CFTR2 (Clinical and Functional Translation of CFTR) project from 88,664 individuals who received regular care at CF specialty centers in 41 countries. Data were provided by national CF patient registries and by major clinical centers in countries with no registry (Table S1 [contributors of variants reported in this manuscript] and Supplemental Acknowledgments [all CFTR2 contributors]). CFTR genotype and sweat chloride concentration were obtained from the individuals’ clinical records as previously reported.¹⁰ Missense variants of unknown effect were selected for study if they were reported in at least three individuals in the CFTR2 database and if the average sweat chloride concentration of individuals bearing the variant of interest in trans with a severe CF-causing variant fell within a range of minimal to moderate elevation of 31–102 mEq/L.

RNA Sequencing Analysis

RNA sequencing raw reads of CF bronchial epithelial (CFBE) cells expressing wild-type CFTR (WT-CFTR) and G551D (c.1652G>A [p.Gly551Asp])-CFTR¹¹ (n = 6), pancreas (n = 3), lung tissues (n = 3), nasal epithelia (n = 3), and air-liquid interface bronchial epithelia (n = 3) were obtained from the NCBI Sequence Read Archive (SRA) (study accession numbers in Table S2). Transcripts were assembled and abundances were estimated with the Tuxedo software suite. We mapped raw reads to the reference genome (UCSC Genome Browser hg19) with TopHat (v2.0.13) by using Bowtie2 (v2.1.0.0). Mapped sequences were assembled with Cufflinks (v2.2.1). CuffQuant was used to estimate the relative abundances of gene transcripts among samples, and CuffDiff was then used to determine differential expression values among samples. For comparison with RT-PCR fold-change values, CFTR expression was measured as a fold change relative to that of HPRT1 (CFTR fragments per kilobase of transcript per million mapped reads [FPKM] divided by HPRT1 FPKM).

Derivation and Analysis of CFBE Cell Lines Expressing CFTR Variants

Site-Directed Mutagenesis

Utilizing human eF1α promoter coupled to WT-CFTR cDNA sequence, site-directed mutagenesis of pEF5/Flp recombinase target plasmid was performed with mutagenic primers designed with the Agilent QuikChange primer design website. PCR products were digested with Dpn1 for the removal of parental plasmid and then ethanol precipitated, rehydrated, and used for transforming XL10-Gold Ultracompetent cells (Agilent). DNA minipreps were prepared (Denville Spinsmart Plasmid Miniprep DNA Purification Kit), and Sanger sequencing confirmed the presence of the variant of interest. A sequence-confirmed miniprep plasmid was used for transforming DH5α competent cells (Invitrogen), and DNA maxipreps were prepared (QIAGEN Plasmid Plus Maxi Kit). Sanger sequencing of the entire CFTR cDNA confirmed the presence of the variant of interest and the absence of secondary changes.

Generation of Human Airway Epithelial Cell Lines with Integrated CFTR Variants

CF bronchial epithelial (CFBE41o-) cells containing an Flp recombinase target integration site¹¹ were grown in complete media supplemented with 100 μg/mL Zeocin (GIBCO or ThermoFisher). Before transfection, cells were seeded in collagen-coated 6-well plates and grown to >70% confluency. We achieved collagen coating by applying a mixture of 5 mL of 0.1% bovine serum albumin (MilliporeSigma), 500 μL rat tail collagen I (Life Technologies, 3 mg/mL), and 500 μL of human fibronectin (Sigma-Adrich, 1 mg/mL) diluted in 44 mL minimal essential media to each well before aspirating the mixture and allowing the plates to dry for at least 1 hr. 0.5 μg of CFTR plasmid combined with 4.5 μg of pOG44 Flp-recombinase plasmid was transfected with Lipofectamine LTX (Life Technologies). Cells were incubated for 48 hr and then split 1:4 into collagen-coated 6-well plates. Media were changed after 48 hr to include 50 μg/mL hygromycin B for 24–48 hr and then changed again to contain 100 μg/mL hygromycin B. Cells remained under hygromycin selection until distinct clones were observed in transfection wells and all cells in mock-transfected wells had died. Individual clones were isolated with 8 × 8 mm sterile cloning cylinders (Millipore) and grown in collagen-coated 24-well plates until confluency, when they were expanded to uncoated vessels for characterization.

Verification of CFTR cDNA Integration by PCR

Genomic DNA was extracted from hygromycin-resistant cells with the QIAGEN DNeasy Blood & Tissue Kit, and full-length CFTR cDNA was PCR amplified for confirming plasmid integration. PCR of the Flp-In site was also performed to confirm disruption of the Flp recombinase target site. PCR fragments containing the sequence variant of each cell line were sequenced for confirmation of the presence of the variant in the genomic DNA of the cell line.

Quantification of CFTR RNA by Real-Time PCR

Total RNA was isolated from cultured cells on the same day that short-circuit current (I_sc) measurements were taken (see below). Cells were washed with PBS made with diethyl-pyrocarbonate-treated H₂O, lysed by the addition of 250 μL of TRIzol reagent (Life Technologies), and then centrifuged through a shredder column (Denville). 100 μL chloroform was added to cell lysates, which were centrifuged at 12,000 × g for 5 min at 4°C. Clear supernatant was kept for RNA purification with the Total RNA Mini Purification Kit (Denville). 500 ng of total RNA was used for generating first-strand cDNA libraries with the iScript cDNA Synthesis Kit (Bio-Rad). RNA and cDNA preparations were made each day that short circuit current measurements were taken. cDNA samples were amplified with SsoAdvanced Universal SYBR Green Supermix (Bio-Rad) on the CFX Connect Real-Time System (Bio-Rad). Two sets of PCR primers were designed for the generation of short CFTR cDNA fragments (121 and 102 bp) that spanned exon-exon junctions. PCR primers for housekeeping genes were purchased from Bio-Rad. Reactions were performed in triplicate for 30 cycles, and ΔCt was calculated for each sample.

Quantification of CFTR by Immunoblot

Whole-cell lysate was purified from WT-CFTR CFBE stable cell lines with 250 μL of RIPA lysis buffer (MilliporeSigma) supplemented with 1% protease inhibitor cocktail (MilliporeSigma) and 0.1% serine protease inhibitor, PMSF (MilliporeSigma). Cells were further lysed by vortexing, and lysates were collected after centrifugation. 100 μg of protein lysate was diluted to 21 μL in PBS and incubated with 7 μL dye solution containing a 1:5 dilution of DTT to 4x Laemmli Sample Buffer (Bio-Rad) at 37°C for 15 min. Samples were run in a 7.5% Tris-HCl, 1.0 mm Bio-Rad Criterion Precast Gel with running buffer composed of 25 mM Tris, 250 mM electrophoresis grade glycine (pH 8.3), and 0.1% SDS in dH₂O. Protein was transferred to a polyvinylidene fluoride membrane by electrophoresis with the Trans-Blot Turbo Transfer System (Bio-Rad) at 2.5 A and 25 V for 10 min. We cut membranes above the 100 kDa band and reserved the upper half for visualization of CFTR and the lower half for visualization of Na⁺/K⁺-ATPase. Membranes were blocked for 1 hr in 5% non-fat dry milk reconstituted in PBS containing 0.1% Tween 20 (PBST). Membranes were washed in PBST and then incubated for 1 hr at room temperature with primary antibody: anti-CFTR m596 antibody (UNC) diluted to 1:1,000 or anti-Na⁺/K⁺ antibody (abcam, ab76020) diluted to 1:100,000. We removed the primary antibody by washing the membranes with PBST for 30 min. Membranes were incubated for 1 hr at room temperature with secondary antibody: anti-mouse (CFTR) (GE Healthcare) diluted to 1:150,000 or anti-rabbit (Na⁺/K⁺) (GE Healthcare) diluted to 1:150,000. We removed secondary antibody by washing the membranes with PBST for 45 min. Membranes were imaged on high-performance chemiluminescence film (GE Healthcare) with ECL Prime Detection Reagent (GE Healthcare) and the Kodak X-Omat 2000A Processor.

CFTR was quantified by the ImageJ software, which divided the intensity of CFTR C-band by the intensity of the Na⁺/K⁺-ATPase band for exposures that had not reached saturation. CFTR quantity represents an average of three immunoblots, and 5–21 exposures were measured for each cell line.

Assessment of CFTR Function by Short-Circuit Current Measurement

1 × 10⁵ cells were plated onto Snapwell filters (12 mm filter diameter with 0.4 μm pore diameter; Corning Costar #3407) for 6 days with daily feeding, resulting in a transepithelial resistance of at least 200 Ω∙cm², although the specific resistance achieved was variable between cell lines. Filters were mounted into Ussing chambers, and I_sc was measured with a VCC MC6 or VCC MC8 multichannel voltage-current clamp amplifier (Physiologic Instruments). Asymmetric apical and basolateral buffers were used for creating a chloride gradient; the apical buffer was composed of 145 mM NaGluconate, 1.2 mM MgCl₂, 1.2 mM CaCl₂, 10 mM dextrose, and 10 mM HEPES, and the basolateral buffer was composed of 145 mM NaCl, 1.2 MgCl₂mM, 1.2 CaCl₂mM, 10 mM dextrose, and 10 mM HEPES dissolved in DI water. Buffers were maintained at 37°C, and air was bubbled in to introduce circulation. After stabilization of transepithelial current, 10 μM forskolin (Selleckchem) was added to the basolateral chamber to stimulate generation of cAMP and activation of CFTR, after which 10 μM of CFTR inhibitor-172 (Selleckchem) was administered in the apical chamber to block CFTR-mediated currents. Data were acquired with the software Acquire and Analyze (Physiologic Instruments). We calculated I_sc changes (ΔI_sc) by taking the difference in the I_sc recorded after adding Inh-172.

In Silico Prediction Models and ACMG-AMP Classification Criteria

To determine the predicted effect of missense variants on protein function, we applied four in silico prediction models: CADD,¹² REVEL,¹³ SIFT,¹⁴ and PolyPhen-2 (HumVar model).¹⁵ These programs were accessed via their respective online tools. CADD and REVEL had specific thresholds for deleteriousness applied (CADD PHRED score of 15, as recommended by the program developers, and REVEL score of 0.659, as determined by a subset of previously defined CF-causing variants). SIFT and PolyPhen-2 distinguish between damaging and tolerated and between benign, possibly damaging, and probably damaging, respectively. All missense variants (total 122), including those reported in this manuscript and those previously published on the CFTR2 website, were also characterized according to the ACMG-AMP variant classification guidelines as pathogenic variants, likely pathogenic variants, VUS, likely benign variants, or benign variants.⁶ Interpretations were assigned with and without the inclusion of functional testing data and were reviewed by the director of a clinical laboratory accredited by the College of American Pathologists (CAP) and certified by the Clinical Laboratory Improvement Amendments (CLIA).

Results

Establishing Wild-Type CFTR Standards for mRNA Quantity, Protein Quantity, and Function

To approximate the native context of CFTR in the lungs and to enable repeated measures of the chloride channel function of CFTR, we studied CFTR variants that were stably expressed in a well-established human airway cell line (CF Bronchial Epithelia; CFBE41o-).¹⁶ To determine the degree to which variants altered function relative to function in wild-type CFTR (WT-CFTR), we independently derived ten CFBE41o- cell lines expressing a single copy of wild-type CFTR cDNA. The expression of CFTR mRNA in each of the ten cell lines varied, as was previously reported in cell lines with single-site integrations,⁴ as did the quantity of CFTR and the magnitude of its function (Figure 1A–1C; Table S3). Because mRNA expression in each CFBE cell line was stable over time,¹¹ we assessed the degree of correlation between RNA quantity, determined by quantitative reverse-transcriptase PCR (qRT-PCR), and mature (band C) protein quantity, assessed by quantitative immunoblotting; the correlation was linear and robust (R = 0.92; p = 1.59 × 10⁻⁴; Figure 1D, left panel). Likewise, CFTR quantity and CFTR function, determined by short-circuit measurement for each of the ten lines, showed excellent linear correlation (R = 0.99, p = 1.5 × 10⁻⁷; Figure 1D, middle panel). Finally, mRNA quantity and CFTR function correlated well (R = 0.94; p = 7.06 × 10⁻⁵; Figure 1D, right panel), indicating that CFTR mRNA quantity could be used for normalizing CFTR chloride currents among independent cell lines.

Figure 1.

CFTR mRNA, Protein Quantity, and Function Are Variable but Correlated

(A) Mean and standard deviations of CFTR mRNA transcript quantity relative to that of HPRT1 (n = 3 for each cell line) for ten independent cell lines expressing WT-CFTR.

(B) Immunoblot detecting varying quantities of mature (band C) CFTR from whole-cell lysates of ten cell lines expressing WT-CFTR. Controls include cell lines expressing the CF-causing variants F508del, which causes a folding defect (band B only), and G551D (band C); non-transfected CFBE cells (no signal); and HEK293 cells transiently expressing WT-CFTR (band C). Loading controls for protein quantity (Na⁺/K+ ATPase) are shown below. The plot on the right shows the mean and standard deviations of CFTR quantities for each cell line relative to the WT-CFTR 5 cell line assessed from at least three immunoblots.

(C) Representative recordings of CFTR function measured by I_sc for ten WT-CFTR cell lines. Forskolin (10 μM) activates CFTR chloride current, and the amount of current inhibited by the CFTR-specific inhibitor inh-172 (10 μM) determines the level of CFTR function. The plot on the right shows the mean and standard deviations for I_sc derived from at least three measurements for the ten WT-CFTR cell lines.

(D) Correlations of the quantity of CFTR mRNA with quantity of mature CFTR (left), quantity of mature CFTR with CFTR function (center), and quantity of CFTR mRNA with CFTR function (right) for ten independent cell lines expressing WT-CFTR.

Determining the Function of CFTR Mutants Relative to the Wild-Type

To establish the relationship between mRNA quantity and function across the full range of WT-CFTR expression, we combined data from 14 additional cell lines with the prior ten WT-CFTR cell lines (total 24) to derive the slope of the linear correlation between CFTR mRNA quantity and CFTR function (R = 0.84; p = 3.37 × 10⁻⁷; Figure 2A and Table S3). We made a minor correction so that the slope intersected with the origin by assuming that zero mRNA should correspond to zero CFTR chloride current, given that parental CFBE cells express no detectable CFTR mRNA and generate no CFTR chloride current. Using the slope function, chloride channel function generated by a variant (Isc_var), and mRNA level of the variant (mRNA_var), we calculated a variant's percentage of CFTR function (F_var) relative to WT-CFTR function as F_var = 100 × (Isc_var/(242.61 × mRNA_var)). With the slope set at 100% WT-CFTR, we then calculated slopes representing 25%, 10%, and 1% WT-CFTR function across the range of observed mRNA expression in CFBE cells (Figure 2A). These functional levels were chosen because they reflect approximate thresholds that transition from fully expressive to partially expressive forms of CF, where the 10%-to-25% threshold represents the range at which most individuals escape life-limiting lung disease.^{17, 18, 19, 20, 21, 22, 23}

Figure 2.

Independently Derived Cell Lines of CFTR Missense Variants Yield Consistent Interpretation

(A) Standard curve (dashed line) for 100% WT-CFTR function derived with CFTR mRNA and CFTR function from 24 independent cell lines expressing WT-CFTR. Predicted CFTR function corresponds to 25% (green), 10% (gold), and 1% (red) of WT-CFTR function across the range of mRNA expression observed in WT-CFTR cell lines.

(B) Plot of log CFTR function against CFTR mRNA quantity derived from the correlation shown in (A). After normalization for mRNA quantity, CFTR variants expressed in multiple independent cell lines show consistent levels of residual CFTR function. Four variants illustrate the range of CFTR function observed.

To ensure that expression levels of CFTR variants were comparable to those of native tissues, we compared the mean RNA levels of CFTR with those of the housekeeping gene HPRT1 by using qRT-PCR in the 24 WT-CFTR cell lines or by extracting data from RNA sequencing of a different WT-CFTR and a G551D-CFTR cell line (0.49 versus 0.57; p = 0.44; Figure S1A). Using the entire set of RNA sequencing data, we determined CFTR transcript quantity relative to those of all other RNA transcripts in the CFBE cell lines and compared these amounts with CFTR mRNA quantity in primary tissues affected by CF (study accession numbers in Table S2). The mean level of CFTR mRNA in the CFBE cell lines was higher than in bronchial and nasal epithelia but lower than its expression in the pancreas (Figure S1B). By extrapolation to the qRT-PCR data, we concluded that CFTR mRNA quantity in the WT-CFTR cell lines were within physiologic ranges of endogenous CFTR mRNA quantity in airway and pancreatic epithelia.

To better discern CFTR variants with low residual function, we plotted chloride channel current on a semi-log chart against CFTR mRNA quantity (Figure 2B). To determine the function of CFTR bearing putative CF-causing variants as a fraction of WT function, we normalized currents generated by CFTR mutants by using the level of mRNA expression in each CFBE cell line. Four examples are presented. Independently derived cell lines stably expressing F508del (c.1521_1523delCTT [p.Phe508del])-CFTR generated currents ranging from 0.3% to 1.4% of WT-CFTR with a mean of 0.7% WT-CFTR after normalization for the mRNA quantity (Figure 2B). This estimate of function is consistent with the association between F508del and full expressivity of CF and measurement of F508del-CFTR function in Fischer rat thyroid (FRT) cell lines (0.2% WT) and in primary bronchial airway cells (range 0.5%–3.4% in F508del/F508del individuals).^{10, 24, 25, 26, 27} CFBE cell lines stably expressing T338I (c.1013C>T [p.Thr338Ile])-CFTR generated currents estimated at 6.4% of WT currents. The residual function of T338I-CFTR was consistent with the full CF expressivity observed in individuals with this variant, albeit a less severe phenotype than observed with F508del.^{10, 28} Variant G622D (c.1865G>A [p.Gly622Asp]) permitted higher amounts of CFTR function at 18.2% of WT-CFTR function, consistent with its partial expressivity.²⁹ Finally, the substitution of cysteine for phenylalanine at codon 508 (F508C [c.1523T>G (p.Phe508Cys)]) had no reduction of CFTR function in two CFBE cell lines (mean 114% WT). F508C has been shown to have a minimal effect on CFTR folding and function,³⁰ consistent with evidence that F508C does not cause disease when found in healthy CF carriers of F508del.³¹

Functional Assessment Distributes Variants according to Expressivity and Informs Assignment of Disease Liability by Using Expert Annotation Criteria

48 missense variants ranging from 0.002% to 0.042% frequency in the CF population were selected from the CFTR2 database (Table 1 and Table S1). 29 variants were associated with clinically diagnostic elevations in sweat chloride concentration (≥60 mEq/L) and life-limiting lung disease, consistent with full expressivity of CF. The remaining 19 variants were associated with partial expressivity (elevated but non-diagnostic sweat chloride concentration [31–59 mEq/L] and variably present life-limiting lung disease). We included four variants (F508del, G551D, I336K [c.1007T>A (p.Ile336Lys)], and T338I) that have been extensively studied in other cell lines and/or primary cells to validate our functional assay (total 52 variants). None of the variants were predicted to cause aberrant mRNA splicing by the algorithms CryptSplice and NNSplice, each of which demonstrated >80% sensitivity.^{32, 33, 34} CFTR function of the 52 missense variants was determined in CFBE stable cell lines that were normalized for mRNA quantity as shown above (Table 1; data for individual clones are reported in Table S4).

Table 1.

Functional Results and Disease Liability Determination for Missense Variants

Variant (Legacy)	Variant (cDNA)	Variant (Protein)	Alleles in CFTR2 (n)	Allele Frequency in CFTR2	Mean Sweat [Cl−] in CFTR2	% WT Function ± SD	CFTR2 Final Determination
Variants with Function < 10% WT-CFTR
F508del	c.1521_1523delCTT	p.Phe508del	98,735	69.856%	≥60	0.7 ± 0.4	CFa
W57G	c.169T>G	p.Trp57Gly	10	0.007%	≥60	1 ± 0.2	CF
L558S	c.1673T>C	p.Leu558Ser	34	0.024%	≥60	1.2 ± 1.6	CF
Y563D	c.1687T>G	p.Tyr563Asp	7	0.005%	≥60	1.3 ± 0.7	CF
Y563N	c.1687T>A	p.Tyr563Asn	33	0.023%	≥60	1.9 ± 1	CF
H609R	c.1826A>G	p.His609Arg	10	0.007%	≥60	2.2 ± 0.4	CF
A613T	c.1837G>A	p.Ala613Thr	6	0.0042%	≥60	2.3 ± 0.9	CF
L1335P	c.4004T>C	p.Leu1335Pro	19	0.013%	≥60	2.4 ± 1.2	CF
I336K	c.1007T>A	p.Ile336Lys	55	0.0389%	≥60	2.4 ± 0.7	CFa
L165S	c.494T>C	p.Leu165Ser	21	0.015%	≥60	2.7 ± 1.3	CF
G551D	c.1652G>A	p.Gly551Asp	2,984	2.111%	≥60	2.9 ± 1.6	CFa
P574H	c.1721C>A	p.Pro574His	25	0.018%	≥60	3 ± 0.5	CF
A1006E	c.3017C>A	p.Ala1006Glu	8	0.0057%	≥60	3.4 ± 1.5	CF
R334L	c.1001G>T	p.Arg334Leu	15	0.011%	≥60	3.6 ± 0.9	CF
P99L	c.296C>T	p.Pro99Leu	7	0.005%	≥60	3.6 ± 0.8	CF
V456A	c.1367T>C	p.Val456Ala	27	0.019%	≥60	4.1 ± 1.4	CF
S1159F	c.3476C>T	p.Ser1159Phe	13	0.009%	≥60	4.7 ± 0.5	CF
D513G	c.1538A>G	p.Asp513Gly	7	0.005%	≥60	4.8 ± 1.7	CF
Q98R	c.293A>G	p.Gln98Arg	16	0.011%	≥60	5.4 ± 0.7	CF
S1118F	c.3353C>T	p.Ser1118Phe	7	0.005%	≥60	5.8 ± 3.1	CF
T338I	c.1013C>T	p.Thr338Ile	52	0.0368%	≥60	6.4 ± 0.8	CFa
R1283M	c.3848G>T	p.Arg1283Met	7	0.005%	≥60	6.7 ± 3.9	CF
E116K	c.346G>A	p.Glu116Lys	8	0.006%	≥60b	6.7 ± 2	CF
D979V	c.2936A>T	p.Asp979Val	3	0.002%	≥60	7 ± 3.7	CF
F311L	c.933C>G	p.Phe311Leu	9	0.006%	≥60	7.6 ± 3.3	CF
Variants with Function 10% to <25% WT-CFTR
T1246I	c.3737C>T	p.Thr1246Ile	23	0.016%	≥60	12.9 ± 4.1	VCC
F1099L	c.3297C>A	p.Phe1099Leu	7	0.005%	≥60	15.1 ± 6.4	VCC
F575Y	c.1724T>A	p.Phe575Tyr	7	0.005%	<60	17.1 ± 3.1	VCC
G622D	c.1865G>A	p.Gly622Asp	8	0.006%	<60	18.2 ± 0.4	VCC
V1153E	c.3458T>A	p.Val1153Glu	6	0.004%	<60	18.7 ± 5.6	VCC
M265R	c.794T>G	p.Met265Arg	7	0.005%	<60	19.1 ± 1.9	VCC
D110E	c.330C>A	p.Asp110Glu	14	0.010%	≥60	19.9 ± 4.6	VCC
Y1032C	c.3095A>G	p.Tyr1032Cys	16	0.011%	<60	20.6 ± 4.7	VCC
P5L	c.14C>T	p.Pro5Leu	60	0.042%	≥60	22.4 ± 3.1	VCC
Variants with Function 25% to 75% WT-CFTR
R334Q	c.1001G>A	p.Arg334Gln	8	0.006%	<60	26.6 ± 4.2	VCC
E588V	c.1763A>T	p.Glu588Val	6	0.004%	≥60	27.5 ± 6	VCC
R117G	c.349C>G	p.Arg117Gly	8	0.006%	<60	34.7 ± 4.7	VCC
A349V	c.1046C>T	p.Ala349Val	11	0.008%	<60	44.9 ± 10.3	IND
V201M	c.601G>A	p.Val201Met	11	0.008%	<60	47.5 ± 15.1	IND
P750L	c.2249C>T	p.Pro750Leu	13	0.009%	<60	48.6 ± 6.6	VCC
D443Y	c.1327G>T	p.Asp443Tyr	8	0.006%	<60	53.2 ± 10.8	VCC
R31L	c.92G>T	p.Arg31Leu	7	0.005%	<60	56.3 ± 16	IND
Q1291R	c.3872A>G	p.Gln1291Arg	9	0.006%	<60	62.3 ± 43.3	VCC
S912L	c.2735C>T	p.Ser912Leu	6	0.004%	≥60	71.2 ± 18.2	IND
Y1014C	c.3041A>G	p.Tyr1014Cys	6	0.004%	<60	73.8 ± 27.7	IND
L967S	c.2900T>C	p.Leu967Ser	20	0.014%	<60	74.4 ± 12.2	VCC
Variants with Function > 75% WT-CFTR
T1053I	c.3158C>T	p.Thr1053Ile	9	0.006%	<60	78.8 ± 27.2	non-CF
F508C	c.1523T>G	p.Phe508Cys	8	0.006%	<60	114 ± 20.1	non-CF
I807M	c.2421A>G	p.Ile807Met	9	0.006%	<60	115.2 ± 27	non-CF
V562I	c.1684G>A	p.Val562Ile	20	0.014%	≥60	116.4 ± 30.5	non-CF
D836Y	c.2506G>T	p.Asp836Tyr	9	0.006%	≥60	122.1 ± 36.4	non-CF
R170H	c.509G>A	p.Arg170His	11	0.008%	<60	150.4 ± 97.1	non-CF

Abbreviations are as follows: CF, CF causing; VCC, varying clinical consequences; non-CF, non-CF-causing; IND, indeterminate.

^aVariant disease liability previously determined and published in Sosnay et al.¹⁰

^bSweat chloride value is based on n = 2, which is below CFTR2 standards (minimum requirement of sweat chloride is n = 3) for use as clinical evidence for disease. Consequently, this variant will not be published on the CFTR2 website.

21 of the 29 variants associated with full expressivity generated less than 10% WT-CFTR function, a conservative threshold for the development of life-limiting lung disease, and we used expert annotation criteria to assign them as CF causing as previously described.¹⁰ Another four variants (P5L [c.14C>T (p.Pro5Leu)], D110E [c.330C>A (p.Asp110Glu)], F1099L [c.3297C>A (p.Phe1099Leu)], and T1246I [c.3737C>T (p.Thr1246Ile)]) reduced function to 10%–25% of WT-CFTR, a range consistent with their less severe phenotype. Because expert annotation uses a highly conservative 10% threshold to define CF causing, these four variants were assigned as having varying clinical consequences (VCC). Of the four remaining variants observed in individuals with full expressivity of CF, one (E588V [c.1763A>T (p.Glu588Val)]) had 27.5% ± 6% WT-CFTR function, which overlaps the 10%–25% WT-CFTR functional range described above and is consistent with a less severe phenotype. However, the remaining three (V562I [c.1684G>A (p.Val562Ile)], D836Y [c.2506G>T (p.Asp836Tyr)], and S912L [c.2735C>T (p.Ser912Leu)]) allowed 71.2%–121.1% WT-CFTR function (Table 1). Even though these variants were presumed to be disease causing and were reported as such by clinicians, functional evidence indicates that they are not deleterious. Notably, all three variants have been reported to occur as part of complex alleles (Table S5) involving other known or likely deleterious variants in cis,^{35, 36, 37, 38} hence potentially explaining their presence in individuals with CF.

None of the 19 variants associated with partial expressivity of CF had less than 10% WT-CFTR function, as expected (Figure 3). Five variants generated 10%–25% WT-CFTR function, consistent with variable disease presentation, and were assigned as VCC. However, ten variants generated between 25% and 75% WT-CFTR function, which should be sufficient to escape life-limiting lung disease. Six of these ten variants were reported in individuals with clinical features consistent with a diagnosis of CF and were characterized as VCC. The remaining four in this group were reported in individuals who did not have clinical features consistent with a diagnosis of CF; however, because they reduce CFTR function by more than 25%, their role in the development of CF disease processes is unclear, and their disease liability is indeterminable. A further four variants associated with partial expressivity of CF had minimal to no effect on CFTR function (75%–100% WT-CFTR) and were assigned as non-CF causing. These included one variant (F508C) with previously published evidence of a lack of CF phenotype when present in trans with F508del.^{31, 39} Of the 14 variants associated with >25% WT-CFTR function, eight are associated with previously reported complex alleles (Table S5), which complicates our ability to directly correlate variant function with phenotype. We suspect that the identification of these “indeterminate” or “non-CF-causing” missense variants in individuals with CF might have led to the erroneous assumption that they were deleterious.⁴⁰ Thus, functional testing was informative for the assessment of variants associated with full expressivity and essential for the interpretation of variants found in individuals with partial expressivity of CF.

Figure 3.

Distinct Distributions of the Residual CFTR Function of Variants Associated with Full or Partial Expressivity of CF

The majority of variants associated with full expressivity of CF allow less than 10% WT-CFTR function, and the remainder distribute across three higher ranges of function. None of the variants associated with partial expressivity of CF have less than 10% CFTR function.

Functional Data Inform the Assignment of Disease Liability according to ACMG-AMP Criteria

Disease assignment using ACMG-AMP criteria, which incorporate evidence from a variety of sources, demonstrated good correlation with expert annotation (Table 2; individual variant annotations in Table S6). Inclusion of functional data in the ACMG-AMP algorithm enabled the assignment of four VUSs as likely pathogenic, and 17 likely pathogenic variants could be assigned as pathogenic (Table 2, upper panel). At the other end of the spectrum, two of the six variants assigned as non CF causing by CFTR2 criteria could be moved from VUS to likely benign or from likely benign to benign. The ACMG-AMP criteria also distributed the 16 VCC variants more precisely when applying a 25% threshold for defining a deleterious variant. Thus, functional data were particularly useful for verifying and excluding pathogenicity, thereby improving the assignment of variants as fully expressive or benign.

Table 2.

Comparison of Variant Annotation according to ACMG-AMP Criteria (with and without Inclusion of Functional Data) and Expert Annotation with Functional Data (CFTR2)

CFTR2 Interpretation	Inclusion of Functional Data in Classificationa	ACMG-AMP Classification
		Pathogenic	Likely Pathogenic	Variant of Uncertain Significance	Likely Benign	Benign
Variants Reported in This Manuscript
CF causing (n = 21)	–	–	17	4	–	–
	+, 10%	17	4	–	–	–
	+, 25%	17	4	–	–	–
VCC (n = 16)	–	–	7	9	–	–
	+, 10%	–	7	9	–	–
	+, 25%	5	6	5	–	–
Non-CF causing (n = 6)	–	–	–	5	1	–
	+, 10%	–	–	4	1	1
	+, 25%	–	–	4	1	1
IND (n = 5)	–	–	–	5	–	–
	+, 10%	–	–	5	–	–
	+, 25%	–	–	5	–	–
Variants Previously Reported on the CFTR2 Website
CF causing (n = 50)	–	1	47	2	–	–
	+, 10%	46	4	–	–	–
	+, 25%	46	4	–	–	–
VCC (n = 12)	–	–	5	7	–	–
	+, 10%	2	3	6	1	–
	+, 25%	4	1	6	1	–
Non-CF causing (n = 11)	–	–	–	3	7	1
	+, 10%	–	–	1	2	8
	+, 25%	–	–	1	2	8
IND (n = 1)	−	–	–	1	–	–
	+, 10%	–	–	–	1	–
	+, 25%	–	–	–	1	–

Abbreviations are as follows: VCC, varying clinical consequence; IND, indeterminate.

^aFunction was incorporated with different thresholds for pathogenicity: 10% (more conservative and considered the threshold for life-limiting lung disease) and 25% (less conservative and could be the threshold for CFTR-related symptoms).

To confirm that ACMG-AMP classifications maintained good correlation with expert annotation for more common variants, we also applied the guidelines to the 74 missense variants (many of which occur more frequently than those reported in this manuscript and some of which have been widely studied by other groups) previously characterized by the CFTR2 team. ACMG-AMP annotations were reviewed by a CAP-accredited and CLIA-certified clinical laboratory director with multiple years of experience in variant classification (M.B.S). Excellent correlation between ACMG-AMP and CFTR2 annotation was again observed, such that all 50 CF-causing variants were assigned as pathogenic or likely pathogenic and 10 of 11 non-CF-causing variants were assigned as benign or likely benign according to ACMG-AMP guidelines when functional data were considered (Table 2, lower panel; individual variant annotations in Table S6). Inclusion of functional data again enabled distribution of the variants annotated as causing varying clinical consequences and confirmed its utility in determining pathogenicity.

Algorithms Fail to Distinguish Variants Associated with Variable Expressivity

To assess the ability of algorithms to predict the effect of the missense variants upon protein function, we evaluated four methods (CADD, REVEL, SIFT, and PolyPhen-2) commonly used in diagnostic and research settings. To ensure that variant scores for CADD and REVEL could reliably predict well-studied, relatively common, fully penetrant missense variants as deleterious, we first tested six CFTR missense variants (A455E [c.1364C>A (p.Ala455Glu)], G551D, G85E [c.254G>A (p.Gly85Glu)], N1303K [c.3909C>G (p.Asn1303Lys)], R334W [c.1000C>T (p.Arg334Trp)], and R347P [c.1040G>C (p.Arg347Pro)]) that are included in the ACMG-recommended panel of CF-causing variants. All six variants exceeded the recommended threshold score for assignment as deleterious by CADD (PHRED score of 15) and scored highly (mean score of 0.88 out of 1) when evaluated by REVEL, which does not have a specific recommended cutoff for deleteriousness. Next, an additional 30 variants previously classified as CF causing¹⁰ were scored as deleterious with CADD, confirming its ability to correctly predict CF-causing variants even at lower frequencies. This group of 30 and the previously described six ACMG variants were used for determining the appropriate REVEL cutoff for deleterious, which was set at 0.659 (two standard deviations below the mean score of all 36 variants) and which corresponds to 62% specificity and 95% sensitivity¹³ (Table S7).

We next scored the 48 missense variants functionally studied here by using REVEL and CADD and used the thresholds described to assign pathogenicity. Two other commonly used methods (SIFT and PolyPhen-2) that provide categorical assignments were also employed (Table 3). Each method classified the majority of variants as deleterious (Table 4). Consequently, accuracy was quite high when calling variants that reduced function below 10% of WT, the level at which full expressivity is present. Likewise, accuracy remained high if 25% of WT-CFTR function was used, although there is less evidence that reduction in function to between 10% and 25% is fully expressive for CF (see above). However, all four methods over-called as deleterious variants that were shown to have greater than 25% WT-CFTR function, leading to high numbers of false positives and low specificity, which ranged from 28% (REVEL) to 6% (CADD). Finally, all four methods were inaccurate in predicting the consequences of the six variants that allowed CFTR to function in the normal range (>75% WT-CFTR) (Table 4). Two programs (CADD and SIFT) predicted all six variants to be deleterious, whereas REVEL and PolyPhen-2 predicted five of the six to be deleterious. Conversely, whereas false-negative rates were low, SIFT was a notable exception: the algorithm called six of 30 variants tolerated, even though functional testing found that four of the six had less than 10% and the remaining two had less than 25% WT-CFTR function.

Table 3.

Predicted Effects of 48 Missense Variants by Four Algorithms

Variant (Legacy)	Variant (cDNA)	Variant (Protein)	CFTR2 Determination	% WT Function	CADD PHRED	REVEL Score	SIFT	PolyPhen-2
Variants with Function <10% WT-CFTR
W57G	c.169T>G	p.Trp57Gly	CF	1.0	29.9	0.834	damaging	benign
L558S	c.1673T>C	p.Leu558Ser	CF	1.2	30	0.98	damaging	probably damaging
Y563D	c.1687T>G	p.Tyr563Asp	CF	1.3	31	0.975	damaging	probably damaging
Y563N	c.1687T>A	p.Tyr563Asn	CF	1.9	31	0.969	damaging	probably damaging
H609R	c.1826A>G	p.His609Arg	CF	2.2	18.85	0.809	tolerated	probably damaging
A613T	c.1837G>A	p.Ala613Thr	CF	2.3	24.9	0.83	tolerated	probably damaging
L1335P	c.4004T>C	p.Leu1335Pro	CF	2.4	29	0.948	damaging	probably damaging
L165S	c.494T>C	p.Leu165Ser	CF	2.7	27.4	0.929	damaging	probably damaging
P574H	c.1721C>A	p.Pro574His	CF	3.0	32	0.98	damaging	probably damaging
A1006E	c.3017C>A	p.Ala1006Glu	CF	3.4	23.2	0.58	damaging	benign
R334L	c.1001G>T	p.Arg334Leu	CF	3.6	23.2	0.726	damaging	probably damaging
P99L	c.296C>T	p.Pro99Leu	CF	3.6	34	0.945	damaging	probably damaging
V456A	c.1367T>C	p.Val456Ala	CF	4.1	25.4	0.916	damaging	possibly damaging
S1159F	c.3476C>T	p.Ser1159Phe	CF	4.7	33	0.919	damaging	probably damaging
D513G	c.1538A>G	p.Asp513Gly	CF	4.8	24.2	0.963	damaging	probably damaging
Q98R	c.293A>G	p.Gln98Arg	CF	5.4	27.1	0.946	damaging	probably damaging
S1118F	c.3353C>T	p.Ser1118Phe	CF	5.8	25	0.085	damaging	probably damaging
R1283M	c.3848G>T	p.Arg1283Met	CF	6.7	33	0.962	damaging	probably damaging
E116K	c.346G>A	p.Glu116Lys	CF	6.7	26.5	0.782	tolerated	probably damaging
D979V	c.2936A>T	p.Asp979Val	CF	7.0	28.7	0.985	damaging	probably damaging
F311L	c.933C>G	p.Phe311Leu	CF	7.6	22.8	0.712	tolerated	possibly damaging
Variants with Function 10% to <25% WT-CFTR
T1246I	c.3737C>T	p.Thr1246Ile	VCC	12.9	32	0.925	damaging	probably damaging
F1099L	c.3297C>A	p.Phe1099Leu	VCC	15.1	23.3	0.619	tolerated	probably damaging
F575Y	c.1724T>A	p.Phe575Tyr	VCC	17.1	29.6	0.868	damaging	probably damaging
G622D	c.1865G>A	p.Gly622Asp	VCC	18.2	29.7	0.964	damaging	probably damaging
V1153E	c.3458T>A	p.Val1153Glu	VCC	18.7	33	0.939	damaging	possibly damaging
M265R	c.794T>G	p.Met265Arg	VCC	19.1	24.3	0.659	damaging	benign
D110E	c.330C>A	p.Asp110Glu	VCC	19.9	16.39	0.733	tolerated	possibly damaging
Y1032C	c.3095A>G	p.Tyr1032Cys	VCC	20.6	25.1	0.887	damaging	probably damaging
P5L	c.14C>T	p.Pro5Leu	VCC	22.4	33	0.887	damaging	probably damaging
Variants with Function 25% to 75% WT-CFTR
R334Q	c.1001G>A	p.Arg334Gln	VCC	26.6	19.78	0.692	tolerated	probably damaging
E588V	c.1763A>T	p.Glu588Val	VCC	27.5	29.3	0.97	damaging	possibly damaging
R117G	c.349C>G	p.Arg117Gly	VCC	34.7	26.4	0.773	damaging	probably damaging
A349V	c.1046C>T	p.Ala349Val	IND	44.9	27.5	0.632	tolerated	possibly damaging
V201M	c.601G>A	p.Val201Met	IND	47.5	27.1	0.676	damaging	possibly damaging
P750L	c.2249C>T	p.Pro750Leu	VCC	48.6	22.9	0.655	damaging	benign
D443Y	c.1327G>T	p.Asp443Tyr	VCC	53.2	26	0.83	damaging	possibly damaging
R31L	c.92G>T	p.Arg31Leu	IND	56.3	22.9	0.546	tolerated	benign
Q1291R	c.3872A>G	p.Gln1291Arg	VCC	62.3	23.2	0.871	damaging	benign
S912L	c.2735C>T	p.Ser912Leu	IND	71.2	9.977	0.543	tolerated	benign
Y1014C	c.3041A>G	p.Tyr1014Cys	IND	73.8	28.5	0.889	damaging	probably damaging
L967S	c.2900T>C	p.Leu967Ser	VCC	74.4	24.7	0.659	damaging	possibly damaging
Variants with Function >75% WT-CFTR
T1053I	c.3158C>T	p.Thr1053Ile	non CF	78.8	26	0.899	damaging	possibly damaging
F508C	c.1523T>G	p.Phe508Cys	non CF	114.0	29.4	0.865	damaging	probably damaging
I807M	c.2421A>G	p.Ile807Met	non CF	115.2	21.9	0.738	damaging	probably damaging
V562I	c.1684G>A	p.Val562Ile	non CF	116.4	23.8	0.637	damaging	benign
D836Y	c.2506G>T	p.Asp836Tyr	non CF	122.1	32	0.924	damaging	probably damaging
R170H	c.509G>A	p.Arg170His	non CF	150.4	34	0.829	damaging	probably damaging

Abbreviations are as follows: CF, CF causing; VCC, varying clinical consequence; non-CF, non-CF causing; IND, indeterminate.

Table 4.

Predicted Functional Consequences of Missense Variants

Functional Grouping	N	CADD		REVEL		SIFT		PolyPhen-2
		Del	Not Del	Del	Not Del	Dam	Tol	Pr/Po Dam	Ben
<10%	21	21	–	19	2	17	4	19	2
10% to <25%	9	9	–	8	1	7	2	8	1
25% to 75%	12	11	1	8	4	8	4	8	4
>75%	6	6	–	5	1	6	–	5	1

Abbreviations are as follows: Del, deleterious; Not Del, not deleterious; Dam, damaging; Tol, tolerated; Pr/Po Dam, probably damaging or possibly damaging; Ben, benign.

To further assess the reliability of the predictive algorithms with variants known to be benign for CF, we also evaluated five relatively frequent CFTR variants that have minimal effect on function and multiple lines of evidence of non-penetrance. All four tools predicted at least two variants to be deleterious, meaning that they over-called 40% (SIFT and PolyPhen-2) or 60% (CADD and REVEL) of variants as deleterious although they are accepted as benign (Table S8). Moreover, three of the five variants (G576A [c.1727G>C (p.Gly576Ala)], R668C [c.2002C>T (p.Arg668Cys)], and S1235R [c.3705T>G (p.Ser1235Arg)]) were inconsistently predicted by the four tools; only one variant (V470M [c.1408G>A (p.Val470Met)]) was consistently and correctly deemed non-deleterious, and another (R75Q [c.224G>A (p.Arg75Gln)]) was consistently but incorrectly deemed deleterious by all four tools.

Discussion

As the number of identified variants continues to increase, interpretation of missense variants and their contribution to expressivity of Mendelian diseases presents a growing challenge to both researchers and clinicians. To inform this process, broader use of laboratory-based functional assays has been advocated.^{6, 41} Wide application of functional tests could increase the accuracy of variant interpretation in genes with both known and unknown association with disease, generate information and reagents necessary for testing therapeutic agents, and inform the development of analytic tools for predicting variant effect.¹ Each of the aforementioned benefits was evident from our analysis of 48 CFTR missense variants associated with variable expressivity of CF.

Assaying effects of missense variants upon protein function is a well-established approach to interpreting disease liability.^{42, 43} A desirable situation is testing mutant function in primary cells obtained from subjects, but this is challenging because (1) samples are not easily collected from difficult-to-access tissues; (2) extremely rare variants might exist only in a few individuals worldwide; (3) primary cells undergo few cell divisions, limiting culturing and expansion; and (4) individual factors (e.g., background variation in the cellular genome) could confound the interpretation of variant effect. Together, these factors limit the number of variants that can be reliably measured in primary cells. High-throughput assays that utilize uniform measures of function, such as assessments of enzyme activity or protein binding, are useful for testing many variants but might not provide detailed functional information for non-soluble proteins or those with complex functions.^{44, 45} Most non-secreted proteins can be analyzed in cell-based systems that discern the severity of defects through comparison of the mutant function to wild-type function.⁴⁶ Thus, heterologous cell-based systems provide a viable option for variant interpretation in the research and potentially the clinical laboratory.^{38, 47} Such systems are not without challenges; transient expression of mutants might vary in transfection rates, cell growth, and other factors, leading to significant differences in expression level and complicating the assessment of mutant function. As shown here, many of these problems can be addressed in systems that enable stable expression of a mutant.

Numerous CFTR mutants have been functionally assessed by stable expression in FRT cells; this system has garnered Food and Drug Administration approval for drug-label expansion to new variants. Although FRT cells have provided a useful platform, they are from a different species and tissue type than human airway epithelia, in which CFTR defects cause life-shortening lung disease. For this reason, we elected to use a CF airway epithelial cell line shown to be a viable substitute for primary airway cells.^{11, 48, 49} By using isogenic cell lines stably expressing CFTR, we were able to normalize for the level of expression among cell lines. This step was critical to interpreting partially functional variants expressed at different levels among individual clones. Without this normalization, minimally functional variants expressed at high levels and high-functioning variants expressed at low levels could appear to have the same effect on function. Our approach to functional studies can be applied to almost any protein of interest because many different cell lines with targetable integration sites are available from academic and commercial sources.

Functional characterization revealed expected and unexpected results. 30 variants associated with full or partial expressivity of the CF phenotype generated 25% or lower WT-CFTR function. This result is consistent with our current understanding of the level of CFTR function associated with a CF phenotype.^{20, 21, 22, 23} Notably, the 21 variants with less than 10% WT-CFTR function were associated with full expressivity of CF. Conversely, 6 of 48 variants did not affect the assayed function of CFTR despite being reported as putative disease-causing variants in individuals with CF. With no meaningful reduction in function, these variants are most likely in cis with other deleterious variants that might not be identified by the genotyping methodology used. Indeed, five of the six variants with function > 75% WT-CFTR exist in cis with known or likely deleterious CFTR variants.^{36, 37, 38, 50} This observation emphasizes that individuals who have an unambiguous clinical diagnosis of a recessive loss-of-function disorder and who carry high-function variants should have sequencing and deletion and duplication analysis to search for deleterious in cis variants. Evaluation of a variant found to exist as part of a complex allele might warrant further functional studies to accurately assess overall allele contribution to disease.

In the absence of other deleterious in cis changes, alternative mechanisms of action not detectable by our methodology should be considered for missense variants with minimal reduction in chloride conductance. It is possible that these variants affect an untested function of CFTR, such as bicarbonate transport, as seen in individuals presenting with pancreatitis.⁵¹ However, the clinical features of some individuals bearing these variants are consistent with CF, and all of these individuals receive regular care from a CF specialty care center. We also believe that all variants should demonstrate the same relationship between mRNA quantity and CFTR function as WT-CFTR (as suggested by Figure 2) and that none of the missense variants tested had a significant impact on mRNA stability, though this mechanism has been suggested by others studying a synonymous change at codon 507.⁵² Another possibility is an effect on translation speed by the use of rare tRNAs, as shown with the synonymous change T854T (c.2562T>G [p.Thr854=]),⁵³ or up- or downregulation, resulting from changes within CFTR, of other genes;⁵⁴ however, these effects are rarely reported to result from amino acid substitutions. Given the relative frequency of in cis changes within CFTR, complex alleles remain the most likely explanation for the observation of variants with high function in individuals with CF.^{36, 37, 38, 50}

12 variants caused a modest decrease in CFTR function (25% to 75% of WT-CFTR), a level generally viewed as not causative for CF but that might be influenced by other factors and result in partially or fully expressive CF in some individuals. Emerging data, such as the presence of additional in cis variants reported in the literature for 5 of these 12 variants or that genetic and/or environmental modifiers conspire to the development of a range of CF symptoms when CFTR function is modestly compromised,⁵⁵ could aid in further refinement of disease liability in the future. These findings illustrate that functional studies can disclose unexpected discrepancies that require further investigation to explain the mechanism of disease and prevent mis-assignment of disease liability, even in a well-studied disorder such as CF. Functional analysis is likely to be of equal or potentially greater utility for more recently discovered disease-associated genes.^{42, 43}

The ACMG-AMP guidelines for the interpretation of sequence variants generally had good correlation with expert annotation, particularly for variants with <25% WT-CFTR function. However, one must carefully define the phenotype of interest so that the terms “pathogenic” and “benign” are used appropriately. For example, a variant associated with a CFTR-related disorder such as isolated male infertility could be deemed pathogenic for that phenotype, but it could be benign in the context of fully expressive CF.⁵⁶ Clinical laboratories are often not provided detailed phenotype information, making it challenging to decide in what disease context to evaluate a variant.⁴⁷ Indeed, an investigation into annotation discrepancies identified functional data and population data as frequent contributors to differing interpretations.⁵⁷ Our study also revealed that correlation between ACMG-AMP and CFTR2 annotation is incomplete because the ACMG-AMP guidelines have no defined category for variants causing partial expressivity. CFTR2 uses the term VCC for variants that cause CF in some people but not others but that typically have enough of a functional deficit that they cannot be considered non-CF causing. As shown here, these variants have reduced CFTR function and are associated with CFTR-related symptoms that do not meet the diagnostic criteria for CF in all individuals bearing them. If CF is the defined phenotype for which ACMG-AMP guidelines are being applied, it might be difficult to assign these variants as pathogenic or likely pathogenic (without clear evidence of CF in some people), but the term VUS is not appropriate because, in many cases, clinical significance is not uncertain. A similar dilemma is faced in annotation of BRCA2 (MIM: 600185) variants in which a probability threshold and annotation for deleterious variants associated with only moderate risk for cancer have yet to be determined.⁵⁸ The ongoing challenge of annotation of variants with moderate functional effects lends itself well to the incorporation of quantitative functional results and their associated endophenotypes (continuous-valued quantitative traits)⁵⁹ into a variant classification scheme.

Considerable effort has been devoted to the development of computational tools that predict variant effect. Most tools consider evolutionary conservation, properties of the native and variant amino acids and their possible impact on structure and stability, or both.⁶⁰ Machine learning using a subset of annotated variants can incorporate more and varied types of information but is limited by data quality, data quantity, and relevance of the training set, often leading to low sensitivity, low specificity, and incongruent variant calls.^{58, 61, 62, 63} The in silico tools evaluated here, selected because they are either commonly used individual predictors (SIFT and PolyPhen-2)⁶⁴ or more recently developed ensemble predictors (CADD and REVEL), had good performance for functionally deleterious variants that lead to full expressivity of CF. However, they could not discern functional levels that distinguish full from partial expressivity of CF. The main issue lies within the binary output of most models, which predict whether or not a variant has an effect but not its magnitude.⁶⁵ This most likely leads to the over-calling of benign variants, some of which might minimally affect protein function but at a level that does not affect phenotype, as deleterious. When we compared functional results and disease liability assignments with the variant-effect predictions of four in silico tools, we found that the tools overestimated deleteriousness, a trend also seen in other genes with well-annotated or functionally assessed variants.^{58, 66, 67} In the current exome- and genome-sequencing era, clinicians and researchers using these in silico tools to assess the pathogenicity of missense variants discovered in putative disease-causing genes in the absence of other supporting data should be cautious.^{47, 68}

In summary, investigation into the effects of 48 missense variants in CFTR has filled a gap in knowledge regarding the function associated with partial expressivity of a recessive Mendelian phenotype. This work will also inform the testing of newly developed therapeutics on rare CFTR variants that are not well suited for clinical trials.

Published: May 24, 2018

Web Resources

• Agilent QuikChange primer design, http://www.genomics.agilent.com/primerDesignProgram.jsp

• CADD, http://cadd.gs.washington.edu/

• CFTR2, https://cftr2.org

• GenBank, https://www.ncbi.nlm.nih.gov/genbank/

• OMIM, http://www.omim.org

• PolyPhen-2, http://genetics.bwh.harvard.edu/pph2/

• REVEL, https://sites.google.com/site/revelgenomics/

• SIFT, http://sift.bii.a-star.edu.sg/