To ensure comprehensive coverage of known CF variants, we first expanded our recently described DMS library by incorporating 106 additional plasmids encoding missense or small insertion/deletion variants contained within the CFTR2 database (McKee et al., 2023). Briefly, we used site-directed mutagenesis to introduce each mutation into a previously described DMS expression vector containing a randomized stretch of 10 bases in the plasmid backbone and a CFTR cDNA bearing three extracellular hemagglutinin (HA) epitope tags in extracellular loop four. We then used whole plasmid sequencing to confirm the sequence of one clone for each variant and determined its corresponding 10-base unique molecular identifier (UMI). We then stoichiometrically pooled the vectors encoding these variants with the 129 other clones in our recently described variant library. The final library contains identifiable UMIs that correspond to 232 CF variants. The variants contained within this expanded library and their associated DMS measurements can be found in Supplementary files 1 and 2.

To evaluate the fidelity of this expanded library and measure the PME of these new variants, we employed DMS to profile the surface immunostaining of these variants as previously described (McKee et al., 2023). Briefly, we generated a pool of stably recombined HEK293T cells in which each individual cell expresses a single CFTR2 variant (see Methods). We then fractionated these cells according to the relative surface immunostaining of their expressed CFTR variants, extracted the genomic DNA from each cellular fraction, and applied deep sequencing to track the relative abundance of each variant within each fraction. PME levels were then calculated from the sequencing results as previously described (Penn et al., 2020). PME measurements determined from this expanded library were highly correlated with our recently reported measurements for the 128 variants included in the first generation DMS library (Pearson’s R² = 0.9975, Supplementary file 2, Figure 1—figure supplement 1), which confirms the integrity of the scoring is maintained in the second-generation library (McKee et al., 2023). The PME intensities of the new 106 variants are generally in line with our previously reported trends: a greater proportion of the mutations in membrane spanning domain 1 (MSD1) compromise CFTR expression relative to those in C-terminal domains of CFTR (Supplementary file 2). Overall, this expanded library offers a more comprehensive coverage of the CFTR2 database with no noticeable difference in the fidelity of DMS scoring.

We previously showed that successful correction of certain misfolded CF variants often coincides with the suppression of CANX interactions. To determine how CANX impacts this wider spectrum of rare CF variants, we first utilized CRISPR to generate a stable CANX knockout (CANX KO) cell line that can be used to identify changes in variant PME by DMS (Methods). Briefly, we transfected the DMS-compatible HEK293T cells with a mixture of Cas9 enzyme, a fluorescent tracrRNA, and a CANX-specific guide RNA, then used fluorescence-activated cell sorting (FACS) to isolate individual clones from the population of transfected cells. We then used genomic sequencing and western blotting to identify an edited CANX KO clone that lacks the CANX protein (Figure 1—figure supplement 2). Consistent with previous reports (Glozman et al., 2009), the loss of CANX significantly reduces the expression of both ΔF508 and WT CFTR in a manner that depends on the N-linked glycosylation of the CFTR protein (Figure 1—figure supplement 3). Indeed, the CFTR surface immunostaining intensity of a recombinant pool of CANX KO cells expressing the CFTR2 variant library is ~69% lower than that of the corresponding pool of parental HEK293T cells (Figure 1A). Moreover, CANX KO cells expressing these variants appear to exhibit less sensitivity to treatment with the combination of VX-661 and VX-445 (Figure 1B, C)—two correctors contained within the CF therapeutic Trikafta. Consistent with previous observations, these results suggest CANX plays a major role in CFTR maturation and its response to correctors.

To identify specific variants that are particularly sensitive to CANX, we used DMS to survey the PME of the CFTR2 library, as previously described, in the context of CANX KO cells. A linear fit of the CF variant surface immunostaining intensities in CANX KO cells against their corresponding values in parental HEK293T cells has a slope of 0.26 and all CF variants exhibit decreased PME in CANX KO cells (Figure 2A, Figure 2—figure supplement 1A). Notably, there are variants throughout the primary structure that exhibit large decreases in PME in CANX KO cells (Figure 2A). However, these effects are not evenly distributed across the structural domains. While CF variants within MSDs 1 and 2 exhibit similar sensitivities to CANX overall, NBD2 variants fare consistently worse than those that fall within NBD1 (Figure 2B). Indeed, a projection of these variant effects onto the three-dimensional structure of CFTR suggests many of the variants that are most adversely affected are located within the C-terminal structural domains, including the domain-swapped TMDs in MSD1 that primarily form tertiary contacts within MSD2 (Figure 2C). Together, these trends suggest CANX is a particularly important modulator of CF variants that perturb the assembly of the C-terminal domains of CFTR. This vectorial bias potentially stems from the position of the N-linked glycosylation site within ECL4, which is likely to recruit CANX during the synthesis and assembly of NBD2.

We previously found that the PME of many misfolded C-terminal variants could be rescued with the type III corrector VX-445 (McKee et al., 2023). To determine whether the enhanced CANX dependence of CF variants within the C-terminal subdomains could be reversed by VX-445, we repeated our DMS in the presence of VX-445. To quantitatively compare the response of these variants to VX-445 in each cell line, we calculated the ratio of the variant PME in the presence and absence of the drug in the context of the knockout cells and subtracted the corresponding ratio in the parental HEK293T cells (see Methods). A plot of the resulting change in VX-445 response for each variant against its basal expression level in the parental HEK293T cells suggests the differences in sensitivity are closely tied to basal expression (Figure 3A). Poorly expressed variants are generally less sensitive in the knockout lines, while those with measurable basal expression are more sensitive in the knockout line. This observation parallels findings from our previous investigation suggesting residual expression is a key factor in corrector-mediated rescue (see Discussion). We note that these gains in sensitivity do not reflect higher expression in the presence of the corrector but rather a larger change relative to lower basal expression in the CANX KO line (Supplementary file 2).

As the majority of poorly expressed CF variants reside within MSD1 and NBD1, many of the variants that exhibit diminished VX-445 sensitivity fall within the N-terminal portions of the CFTR protein (Figure 3B). However, there is a notable cluster of C-terminal variants that also exhibit diminished sensitivity to VX-445 (see the green stripe in Figure 3B). These variants fall within TMDs 10 and 11—a domain-swapped portion of MSD2 that makes structural contacts within MSD1 and forms part of the VX-445-binding site (Figure 3C). While many of these variants preferentially respond to VX-445 in parental cells (McKee et al., 2023), their rescue is compromised in the absence of CANX (Supplementary file 2). Computational docking studies for certain CFTR variants of interest reveal that VX-445 binding has little impact on G85E but should generally stabilize P5L, V232D, L1077P, W1098R, and N1303K CFTR to a similar degree (Figure 4A). While impacts on stability appear to be quite similar, modeling suggests the conformational effects of the mutations and of VX-445 binding are more nuanced. For instance, P5L, L1077P, and W1098R generate conformational defects that are allosterically propagated to NBD2 (Figure 4B). In each of these cases, the binding of VX-445 allosterically suppresses these defects (Figure 4B, D). These observations potentially suggest a means by which the loss of CANX could compromise the pharmacological rescue of variants that NBD2 misassembly—CANX and VX-445 are potentially both needed to support NBD2 assembly. However, any such perturbations of the native ensemble appear to be more subtle for other misfolded variants including G85E, V232D, and N1303K (Figure 4A, C, E), which suggests this allosteric crosstalk may not be fully generalizable. Indeed, our previous findings suggest the activity of VX-445 may depend upon its interaction with nascent CFTR molecules that have yet to achieve the native structures characterized in these models (McKee et al., 2023). Overall, these results suggest that the loss of CANX generally hampers VX-445-mediated rescue of poorly expressed variants as well as those that compromise the assembly of the C-terminal domains.

Combining VX-445 with the type I corrector VX-661 broadly enhances the rescue of misfolding CF variants (McKee et al., 2023; Bihler et al., 2024). To determine whether combining these correctors can suppress the CANX dependence of specific CF variants, we repeated our DMS in the presence of VX-445 + VX-661. Treatment with correctors generates a much smaller increase in the net PME of CANX KO cells expressing the CF variant library (Figure 1B, C), which clearly shows that the combination of these compounds cannot fully compensate for the loss of CANX. Similar to the variant profiles in the presence of VX-445 alone, variants with low basal PME values tend to exhibit a weaker response to this pair of correctors in CANX KO cells (Figure 5A). However, the gain in corrector sensitivity among variants with residual expression is greater than the degree of sensitization observed in the presence of VX-445 alone (Figures 3A and 5A). While this pair of correctors produces a greater overall change in expression in the CANX KO cells, the poorly expressed variants within MSD1, NBD1, and TMDs 10 and 11 still exhibit weaker response to this pair of correctors (Figure 5B, C). The overall similarity of the trends observed with either individual correctors or with this pair of correctors suggests any variant-specific CANX-dependent changes in variant theratype must be relatively subtle (Figure 5—figure supplement 1). Combining mechanistically distinct correctors does not appear to dramatically change the extent to which CF variant theratypes depend on CANX interactions.

Overall, our DMS measurements suggest correctors are less effective at re-balancing the cellular proteostasis of CF variants in the context of CANX KO cells. To determine how the loss of CANX modifies the effects of correctors on CF variants with divergent theratypes, we employed co-immunoprecipitation coupled to mass spectrometry to compare the interactomes of a series of CF variants that exhibit distinct dependencies on CANX (Figure 6A). Based on a comparison of our collective DMS measurements in the parental and knockout cells, we first compared I1366N, which exhibits an enhanced corrector response in knockout cells, to V232D, which instead has a weaker corrector response in CANX KO cells (Supplementary file 2). Importantly, the baseline expression of I1366N is roughly double that of V232D in parental cells and five times higher in the knockout cells (Supplementary file 2). Based on this consideration, it is perhaps unsurprising that V232D exhibits more pronounced increases in translation, folding, and trafficking interactions in the knockout cells (Figure 6B)—a larger fraction of the V232D is likely to be tied up within polysomes and/or engaged by ER QC. Perhaps the most striking difference involves trafficking interactions. The interactions formed between I1366N and cellular trafficking proteins are quite similar in both cell lines and are indistinguishable from those of WT in the knockout cells (Figure 6B). These observations suggest trafficking interactions potentially play a key role in pharmacological rescue (see Discussion).

To expand our search for interactions that modulate theratype, we also profiled the VX-445-selective variants P5L and T1036N. Both of these variants retain minimal expression and maintain comparable corrector response profiles in each cell line, though P5L achieves higher correction overall (Supplementary file 2). These variants exhibit distinct perturbations in their translation and folding interactions, which suggests they are likely to misfold in distinct ways in the absence of CANX (Figure 6B). Nevertheless, they each display a similar uptick in degradative interactions associated with the proteasome and autophagy pathways (Figure 6B), suggesting the cellular outcome of these misfolding reactions is quite similar in knockout cells. Interestingly, both variants form similar interactions with trafficking proteins, though these perturbations appear to be more similar to the severely misfolded V232D than to the I1366N variant. This may again reflect the comparable severity of the P5L, V232D, and T1036N folding defects, as these variants all fail to accumulate at the plasma membrane and/or late secretory pathway. Taken together, these collective observations suggest enhanced proteasomal interactions are associated with poor corrector response while the persistence of native trafficking interactions may facilitate greater rescue.

We previously evaluated how correctors modify the interactome of ΔF508 in HEK293T cells (McDonald et al., 2022a). To compare how correctors re-wire proteostatic interactions in the absence of CANX, we compared the effects of VX-661, VX-445, and VX-661 + VX-445 on the ΔF508 interactome knockout cells. We first note that ΔF508 exhibits enhanced interactions with several categories of QC proteins in the context of CANX KO cells (red stars, Figure 6C–E). Unlike our previous findings for ΔF508 in unmodified HEK293T cells (McDonald et al., 2022a), treatment with correctors does not suppress the autophagocytic and/or proteasomal interactions that mediate its degradation in the knockout cells (Figure 6D), which potentially explains the weaker pharmacological rescue in this cellular context (Supplementary file 2). However, each corrector treatment significantly suppresses the aberrant translational interactions that form in the absence of CANX (Figure 6E). Together, these observations suggest that the loss of CANX further compromises the cotranslational assembly of ΔF508 in a manner that cannot be fully suppressed by corrector molecules. Indeed, this observation is consistent with our recent identification of CANX as a chaperone involved in the cotranslational QC of this variant (Carmody et al., 2024). Nevertheless, the divergent modifications to the interactomes of the other variants in the knockout cells suggest the effects of CANX on variant theratype is likely to vary widely (Figure 6—figure supplement 1).

Interactome measurements suggest the loss of CANX significantly alters the manner in which the CFTR protein is assembled within the early secretory pathway. To evaluate functional consequences of this change in variant assembly and correction, we compared the activity and pharmacological response of the five divergent CF variants described above in the context of the CANX KO and the parental HEK293T cells. Briefly, we generated a series of recombinant stable CANX KO and HEK293T cell lines that inducibly express each CF variant in conjunction with a ratiometric halide-sensitive YFP (hYFP) sensor protein (Galietta et al., 2001; Vijftigschild et al., 2013) (see Methods). We then tracked the ratiometric quenching of hYFP fluorescence over time following CFTR activation at the single-cell level using flow cytometry, as previously described (Carmody et al., 2024). The observed rate of hYFP quenching is slower in the presence of the CFTR-specific inhibitor CFTR(inh)-172 (Figure 7—figure supplement 1), which confirms that quenching is rate-limited by CFTR conductance. As expected, quenching is faster in cells expressing WT CFTR relative to cells expressing CF variants (Figure 7A, B), though there are prominent differences in the activity of these variants (two-way ANOVA, p = 3.8 × 10^–17). However, the trends in quenching half-lives appear unchanged in CANX KO cells (Figure 7B, two-way ANOVA, p = 0.06). Thus, while DMS measurements suggest these mutations reduce the CFTR PME by an average of ~69% (Figure 1A), any resulting changes in functional activity appear to be quite modest. Furthermore, the CFTR modulators found in Trikafta generate a similar degree of functional rescue for each of these variants in the context of the CANX KO cells (Figure 7C). Overall, these findings reveal the proteostatic effects of CANX on CFTR assembly are decoupled from differences in CF variant activity, which suggests the CFTR molecules that are assembled in the absence of CANX are more active than those that undergo CANX-mediated assembly.

To determine whether CANX modulates CFTR activity in a more relevant cellular context, we measured short-circuit currents across Fischer rat thyroid (FRT) cell monolayers transiently expressing WT or ΔF508 CFTR. As expected, monolayers expressing WT CFTR exhibit robust forskolin-stimulated currents that are suppressed by CFTR(inh)-172 (Figure 7D). Moreover, the corresponding currents across monolayers expressing ΔF508 only reach 2.4% of the current achieved by cells expressing WT CFTR (Figure 7E, F). Knocking down CANX expression reduces the current in monolayers expressing WT CFTR by 27 ± 12% (Figure 7D, p = 0.003). Notably, this decreased CFTR activity cannot be reversed by CFTR modulators (Figure 7D). Due to its poor basal activity, knocking down CANX appears to have little effect on ΔF508 CFTR (Figure 7E, F). However, knocking down CANX significantly reduces the functional rescue of ΔF508 CFTR by CFTR modulators (Figure 7E, F). Together, these data demonstrate that CANX does support the functional maturation and pharmacological properties of CFTR in the context of cellular monolayers. While our observations potentially suggest expression and function are more tightly coupled in FRT monolayers, we note that the average quenching half-lives in HEK293T also suggest WT CFTR is less active. The pharmacological rescue of ΔF508 CFTR is also less pronounced in the CANX KO HE293T cells, though any differences appear relatively small and are not statistically significant (Figure 7B, C). Thus, functional measurements appear to be qualitatively consistent across the two cell lines. We suspect the statistical differences are likely rooted in differences in methodological precision rather than meaningful functional differences. Overall, our findings suggest that the impact of CANX on CF variant function appears modest in relation to its effect on the PME of CF variants.

CFTR agonists and antagonists were purchased from commercial providers. Compounds included tezacaftor (VX-661; MedChemExpress, #HY-15448), elexacaftor (VX-445; Selleck Chemicals, #S885), ivacaftor (VX-770; MedChemExpress, #HY-13017), amiloride (Sigma-Aldrich, #A7410), forskolin (Selleck Chemicals, #S2449), and CFTR Inhibitor172 (Inh172; Selleck Chemicals, #S7139). For gene-specific knockdown, siRNAs were used to target calnexin (Canx; QIAGEN, #SI02666335) or a non-specific (NS) negative control (QIAGEN, #1027310). The following target sequences were employed (5′ → 3′): Canx, ACCAGGGAATCTGGAAACCAA; and NS, AATTCTCCGAACGTGTCACGT.

A pcDNA5 vector, in which we inserted CFTR, a triple HA tag in the fourth extracellular loop, an internal ribosome entry site-eGFP cassette, and a Bxb1 recombination site was used to generate a molecular library of barcoded CF variants. We first installed a randomized 10-nucleotide UMI upstream of the Bxb1 recombination site using nicking mutagenesis in place of the CMV promoter. A plasmid preparation containing WT plasmids with scrambled UMIs was used as a template for 106 individual site-directed mutagenesis reactions to generate a library of pathogenic CF variants that were found in the CFTR2 Database (https://www.cftr2.org/). Individual clones from each reaction were generated using the ZymoPURE Plasmid Miniprep Kit (Zymo Research, Irvine, CA) and deep sequenced by shotgun Illumina sequencing to confirm the sequence of each mutated open reading frame and to determine its corresponding 10 base UMI. Plasmids encoding individual variants were pooled with 129 CF variant-encoding plasmids previously described by McKee et al., 2023 The plasmid pool was then electroporated into electro-competent NEB10b cells (New England Biolabs, Ipswitch, MA), which were then grown in liquid culture overnight and purified using the ZymoPure endotoxin-free midiprep kit (Zymo Research, Irvine, CA). Based on subsequent deep sequencing, the resulting plasmid pool contains 232 variants. The barcode sequence associated with L1065P could not be detected in the final pool. The Bxb1 recombinase expression vector (pCAG-NLS-HA Bxb1) was kindly provided by Douglas Fowler.

HEK293T cells were grown in Dulbecco’s modified Eagle’s medium (Gibco, Grand Island, NY) containing 10% fetal bovine serum (Corning, Corning, NY) and a penicillin (100 U/ml)/streptomycin (100 µg/ml) antibiotic supplement (Gibco, Grand Island, NY) in a humidified incubator containing 5% CO₂ at 37°C. HEK293T cells used for interactome measurements were grown under similar conditions but with 1% l-glutamine (200 mM) supplementation. CFTR variant DNA constructs were stably recombined in modified HEK293T cells containing a genomic Tet-Bxb1-BFP ‘landing pad’ (Matreyek et al., 2017) by co-transfecting with Fugene 6 (Promega, Madison, WI) and Bxb1 recombinase expression vectors. The cells were then incubated at 33°C for 3 days and induced with doxycycline (2 µg/ml) 1 day after transfection. The cells were then incubated at 37°C for 24 hr prior to the isolation of GFP-positive/BFP-negative cells that had undergone recombination using a BD FACS Aria II (BD Biosciences, Franklin Lakes, NJ). These cells were grown in 10 cm dishes with complete media supplemented with doxycycline (2 µg/ml) until confluency when they were divided into 15 cm dishes for the three-drug trial replicates. Cells were dosed with 3 µM VX-661 and/or VX-445 16 hr prior to harvesting when applicable. Seven to twelve days post-transfection, cells were washed with 1X phosphate buffered saline (AthenaES, Baltimore, MD) and harvested with 1X0.25% Trypsin-EDTA (Gibco, Grand Island, NY). CFTR expressed at the plasma membrane of recombinant cells was labeled with a DyLight 550-conjugated anti-HA antibody (Thermo Fisher, Waltham, MA). Labeled cells were then fractionated into quartiles according to surface immunostaining intensity (Figure 1—figure supplement 4) using an FACS Aria IIu fluorescence-activated cell sorter (BD Biosciences, Franklin Lakes, NJ). At least 1 million cells from each fraction were isolated to ensure exhaustive sampling. Sorting gates for surface immunostaining were independently set for each biological replicate and in each condition to ensure that the population was evenly divided into four equal subpopulations. Fractionated subpopulations were expanded in 10 cm culture dishes prior to harvesting and freezing 15–25 million cells per quartile fraction for the downstream genetic analysis.

Parental FRT cells were a gift of Michael Welsh (University of Iowa, USA). As detailed previously (Oliver et al., 2019; Sabusap et al., 2016; Rauscher et al., 2021), cells were cultured in F12 Ham Coon’s modified nutrient mixture (Sigma-Aldrich, St. Louis, MO) supplemented with 2.68 g sodium bicarbonate, 850 μl 2 N hydrochloric acid, and 5% fetal bovine serum (pH 7.3). Incubator conditions were set to 5% CO₂, 95% O₂, and 37°C. Parental FRT were co-transfected with plasmid DNA and siRNA based on established protocols (Oliver et al., 2019; Rauscher et al., 2021; Ramachandran et al., 2013; Sabusap et al., 2021) In short, cells were seeded on 0.33 cm² permeable transwells at 1.5 × 10⁵ cells/well and allowed to adhere overnight. The following day, human F508del- or WT-CFTR cDNA cloned into the pcDNA5 vector (Invitrogen, Waltham, MA) was transfected (0.5 μg) onto the apical surface of FRT monolayers with 0.5 μl Lipofectamine 3000 and 0.75 μl P3000 (Invitrogen, Waltham, MA) suspended in 20 μl 1X Opti-MEM Reduced Serum (Gibco, Grand Island, NY). Plasmids incubated 24 hr, then were aspirated to facilitate apical transfection of siRNA duplexes. CANX or NS siRNA (100 nM) were delivered to polarized FRT cells by forward transfection using 0.5 μl Lipofectamine RNAiMAX (Invitrogen, Waltham, MA). Duplexes of siRNAs were incubated 24 hr, then aspirated to induce cell culture at air–liquid interface. Transfections with siRNA generated a ~50% knockdown efficiency (Figure 7—figure supplement 2).

A modified HEK 293T cell line containing a bxb1 genomic landing pad (Matreyek et al., 2017) was transfected with CANX using CRISPRMAX (Thermo Fisher, Pittsburgh, PA) and verified via western blot. Alt-R CRISPR–Cas9 crRNA complement to exon 8 was identified and ordered via IDT Genome editing. 1 µM of this crRNA was incubated with 1 µM Alt-R CRISPR–Cas9 tracrRNA at 95°C for 5 min. 6 µl of this mixture was incubated with 6 µl 1 uM Alt-R S.p. Cas9 Nuclease V3, 2.4 µl Cas9 Plus, and 85.6 µl OPTI-MEM at room temperature for 5 min. 100 µl of this RNA complex was incubated with 4.8 µl CRISPRMAX and 85.2 µl OPTI-MEM at room temperature for 20 min. While incubating, the cells were washed and resuspended using 1 × 0.25% Trypsin-EDTA, counted, and diluted to 400,000 cells/ml in DMEM complete media. 50 µl of the RNA complex and 100 µl of cell were added to each of 48 wells, incubated for 48 hr at 37°C. The cells were then resuspended and sorted for single cells on ATTO550 using an FACS Aria IIu fluorescence-activated cell sorter (BD Biosciences, Franklin Lakes, NJ) into a single well of a 96-well plate. The cells were grown up, harvested, and lysed to be analyzed by western blot and sanger sequencing.

FRTs were grown on permeable filters for a total of 5 days to facilitate monolayer establishment and polarity. Forty-eight hours prior to functional and biochemical assessments, cells were treated basolaterally with CFTR correctors and/or potentiators (5 μM per drug). Cells were mounted in P2300 Ussing chambers (Physiologic Instruments, MC8 Apparatus) and analyzed for short-circuit currents as previously described (Oliver et al., 2019; Rauscher et al., 2021; Sabusap et al., 2021). Briefly, cells were equilibrated 5–10 min with basolateral regular Ringer buffer and apical low-chloride Ringer buffer. Amiloride (100 μM) was applied apically as a sodium channel inhibitor, after which the following CFTR agonists were added apically and/or basolaterally: forskolin (10 μM; activator of PKA) and VX-770 (5 μM; potentiator of CFTR gating). At the conclusion of each experiment, Inh172 (10 μM) was applied to the apical surface. CFTR-dependent transepithelial ion transport was measured as the difference (Δ) from baseline current to the highest or lowest value of a stable plateau achieved for 5–10 min.

Steady-state mRNA levels were measured from FRT cells collected after short-circuit current analysis. Total RNA was isolated using a Direct-zol RNA Miniprep Plus Kit with DNase I (Zymo Research, Irvine, CA). RNA was then reverse transcribed using oligo-(dT)18 primers and Luna Probe One-Step RT-qPCR Mix (New England Biolabs, Ipswitch, MA) according to the manufacturer’s instructions. Primers were purchased from IDT Technologies for human CFTR (Hs.PT.58.3365414), rat CANX (Rn.PT.58.34943747), and rat β-actin (Actb; Rn.PT.39a.22214838.g). These targets were amplified with the following primer sequences (5′ → 3′):

PCR was performed in sealed, clear, 96-well plates (Applied Biosystems, Waltham, MA) placed in a thermocycler (Applied Biosystems, 7500 Real Time PCR System). The ΔΔCT method was employed to calculate product levels of CFTR and CANX relative to Actb. Non-template and non-reverse transcribed samples served as negative controls.

CFTR mutants were modeled using Rosetta comparative modeling as previously described (McDonald et al., 2025; McDonald et al., 2022b) Briefly, the five lowest-energy cryo-EM refined models (PDB ID 6MSM) were selected for VX-445 docking. Ligand conformations were generated with BCL ConformerGenerator, and VX-445 was aligned to published structures in Chimera before being docked 1000 times using RosettaLigand, with 24 cycles of full-atom docking and repacking. The final pose was chosen based on interface energy scores and root mean squared deviation (RMSD) from the published VX-445-bound structure (PDB ID 8EIQ). Mutations were introduced using Rosetta’s MutateResidue mover.

Thermodynamic stability was assessed using Rosetta scores and Cα RMSD. Energy changes were calculated by comparing the Rosetta scores of apo and VX-445-bound ensembles. RMSD differences between mutant and wild-type (WT) CFTR were mapped onto CFTR structures. Mutation-induced instability was determined by subtracting the variant RMSD from WT RMSD, while VX-445 stabilization was assessed by comparing apo and VX-445-bound RMSD. These summed ΔRMSD values were quantified for NBD2 (residues 1201–1436). Finally, ΔRMSD was mapped onto the CFTR structure (PDB ID 6MSM) to investigate changes in local ensemble diversity interpreted as structural fluctuations.