Subtle Cellular Phenotypes Inform Pathological and Benign Genetic Mutants in the Iduronate-2 Sulfatase Gene

Abstract

Molecular genetic testing is increasingly used in clinical care to identify genetic variants and their impact on disease burden. However, variants of uncertain significance (VUS) hamper the utility of molecular diagnostic testing. In patients presenting with Hunter Syndrome and VUS in the IDS gene, clinical testing for iduronate-2-sulfatase enzyme activity has been the mainstay to determine whether a variant is likely damaging. However, enzyme assays alone fail to predict disease severity. In this study, we developed an image-based cellular assay using genome-engineered cells with IDS variants to determine whether a specific variant causes morphological changes that are associated with disease. Specifically, we generated twelve mutant cell lines and documented both IDS biochemical activity and reproducible phenotypic differences therein. Next, we examined patient-derived cell lines and found the same phenotypic differences compared to parental controls. The morphological changes were complex, but measured on a single scale, which we termed PathScore_LC. To determine whether the observed changes are specific to IDS, we reintroduced a recombinant human IDS enzyme (rhIDS) to rescue both the biochemical and phenotypic changes of these cells. We found a partial rescue in the presence of corrective levels of IDS enzyme. Finally, we examined the differences in gene expression and found that a recombinant enzyme was not sufficient to fully restore transcriptional changes in the mutant lines at the time points studied. This proof-of-concept study establishes preliminary validation of the method and sets the stage for future functional studies and broader IDS variant testing.

Introduction

The classification of variants in the IDS gene identified through molecular DNA testing is hampered by difficulties in interpreting variant pathogenicity accurately, which complicates the differentiation among mucopolysaccharidosis type II (MPS II) phenotypes. To address these challenges, investigators have established standardized criteria, incorporating in silico prediction tools, population frequency data from resources such as gnomAD, and assessments of whether a variant is inherited or arose de novo in the proband.^1,2 However, in silico models could potentially overestimate the functional impact of pathogenic variants, as even a small percent of IDS enzyme activity is corrective.³ In vitro assays which utilize artificial substrates present challenges of their own, including the potential to misclassify pseudodeficiency alleles as pathogenic, resulting in false-positive findings.^4,5 Thus, a significant proportion of variants remain unclassified, often categorized as variants of uncertain significance (VUS). VUS have been reported in around 20% of genetic tests, highlighting their prevalence.⁶ Newborn screening for MPS II, while accepted into the Recommended Uniform Screening Panel (RUSP), is available in eighteen states at the time of this writing.^7,8 These obstacles in classification underscore discrepancies among researchers and clinicians regarding the interpretation and application of established criteria.^9,10 Early detection is crucial, as individuals with MPS II exhibit early onset symptoms, reduced lifespans, and irreversible damage due to the progression of the disease.^11–13 Thus, advancing predictive models holds promise for mitigating these challenges, ultimately enhancing diagnostic accuracy, improving patient care, and advancing our understanding of IDS gene variants.

Efforts to address challenges with variant classification, especially for VUS, are ongoing. Platforms such as GeneMatcher and other voluntary reporting services assist by collating information on genetic variants along with phenotypic data to identify similar cases.^14,15 In silico prediction tools, though imperfect, are capable of rapidly evaluating nucleotide variants to determine their impact on protein function.^16–18 These tools, however, rely heavily on a deep understanding of protein functional domains which is currently incomplete. The use of model organisms for studying genetic variants is promising, but is of low throughput and limited by reliance on orthologues and intronic variants that are different across species.¹⁹ Cell-based assays utilizing patient-derived human cell lines exist but are hindered by a lack of functional assessments that are available to determine the pathogenicity of variants in a single cell. Meanwhile, the scope of the VUS problem is expanding. Clinical whole exome sequencing (WES) or genome sequencing have become cost-effective approaches with high diagnostic yields, leading to broader use.^20–23 The American College of Medical Genetics and Genomics (ACMG) now recommends it as a first- or second-line test for developmental delay or intellectual disability manifesting before age 18 years, or congenital anomalies with onset under age 1 year.²³

We attempt to fulfill this need by using cellular phenotypes to proxy disease pathogenicity and aid variant classification. As initial validation, we explored whether image-based screening of genome-engineered and patient-derived cell lines could reveal reproducible phenotypic differences linked to known IDS variants. In this study, we tested our model’s ability and fitness to classify IDS genetic variants known to cause MPS II, also referred to as Hunter Syndrome. This X-linked disorder that is due to mutations in the IDS gene leads to reduced production of the lysosomal enzyme iduronate-2-sulfatase (I2S; referred to hereafter for simplicity as IDS), which is responsible for degrading the glycosaminoglycans (GAGs) heparan sulfate and dermatan sulfate in the lysosome.²⁴ In the absence of this critical enzyme, GAGs build up in the lysosome and cause systemic damage across many cells, tissues, and organs.²⁵ Although biochemical assays such as IDS enzymatic assays and GAG measurements are available, a significant proportion of VUS necessitates a new methods of classification.²⁶ Here we describe initial attempts to address this by using image-based screening to differentiate between various classifications of variants in the IDS gene, with the goal of improving the categorization of VUS.

Results

Section 1: IDS Variants Affect both Enzyme Catalytic Ability and Cellular Phenotype

First, we chose a set of ‘anchor’ variants for IDS including known severe (pathogenic), known attenuated (pathogenic), and known benign missense variants (Supplemental Table S1 and Figure 1A). We then engineered these variants into an A549 cell line (Supplemental Table S2). The A549 cell line has more than one X chromosome, so variants were present in combination. To evaluate the cellular impact of IDS mutations, we compared biochemical enzyme activity and pathogenicity scores based on a combination of phenotypes obtained from cellular imaging of these A549 cells, including morphological (shape) features, and lysosome and cell membrane intensity measures (Supplemental Table S3).

Figure 1. Phenotypic Signatures in A549 cells with Point Mutations in the IDS Gene.

(A) Selected variants curated from ClinVar with documented clinical phenotypes and genetic consequence. Red text indicates a severe variant and green a benign variant. *VUS is the variant of uncertain significance tested in this study. (B) IDS enzyme activity measured in the 12 isogenic variant cell lines and controls by fluorometric assay with a 4-methylumbelliferyl substrate. Severe genotypes are shaded in red, attenuated in blue, and benign in green. (C) Fluorescent confocal micrographs of A549 cells, with nuclear stain in blue, cell membrane in red, and lysosome in green. The left panels show a larger field of cells, magnified for more detail in the panels to the right. Four of the mutant lines are shown, one per row. (D) Cumulative histograms of every individual cell from multiple replicates. X-axis shows the lysotracker non-uniformity (NU) texture, and the y-axis the rank percentile of every cell, where red lines are the G224R/G224R and the P228fsX/KO pathogenic lines, green are the block/block and V223I benign lines, and gray are all other variants. (E) Bar chart displays the average lysotracker texture NU with error bars representing standard deviation of 20–24 well replicates from multiple plates. (F,G) Same series as D,E, but reflecting the PathScore from the lysosomal and cell membrane (LC), a composite feature learned from the pathogenic and benign cells. (F) Cumulative histogram and (G) summary bar-chart with mean and standard deviation. Kruskal-Wallis p-value~0, H-stat 17,869 among cells.

Of the twelve IDS-variant A549 isogenic cell lines, the pathogenic mutants exhibited markedly reduced IDS activity, with values ranging from 0 to 0.8 units/mg protein (Figure 1B). As expected, there was no clear separation between severe and attenuated pathogenic variants based on enzymatic activity. In contrast, benign variants demonstrated higher IDS activity, ranging from 3.5 and 7.7 units/mg protein.

We next performed fluorescent confocal imaging of the A549 IDS-mutant cells (Figure 1C). We found that benign variant cell lines and control cells exhibited well-defined nuclei, and high cell density, while attenuated pathogenic variant line P228A/P228A and severe pathogenic line P228fsX/KO showed increased irregularities by severity, greater overlap of the lysosomal and cell membrane markers, and reduced density compared with benign variant and control cell lines. Cells from the severe pathogenic line, G224R/G224R, displayed more disorganization, pronounced colocalization of markers, and the lowest cell density. Quantifying individual features of the cells, such as the intensity of the lysosomal stain or the size of the cell, did not generally discriminate between benign and pathogenic-mutant cells (not shown). The best single feature that could delineate the samples was a lysosomal ‘texture’ feature, that measures the patterns in the lysosomal staining (Figure 1D, E) but was still poor at discriminating individual mutant cells. Next, we used a composite metric derived from multiple features, termed the PathScore_LC (Kruskal-Wallis p-value~0, Figure 1F,G).²⁷ Neither the lysosome, cell membrane, or nuclei alone performed as well as a classifier which took both the lysosomes and cell membrane into account (Supplemental Figure S1). The phenotypic PathScore increased according to severity from benign to pathogenic variants and was approximately inverse of the IDS activity levels. Interestingly, some variants that had intact enzyme activity due to expression from a single normal allele (G224R/block and P228fsX/WT) displayed higher-than-expected PathScores, potentially indicating an effect of haploinsufficiency on the cellular phenotype. Overall, these findings demonstrate that mutations affect both IDS production and cellular phenotypes in a quantifiable and distinct manner (Supplemental Figure S2), allowing us to determine the degree of cellular damage caused by these IDS mutations.

Section 2: Patient-derived Fibroblasts Phenocopy both Enzymatic and Phenotypic Signatures of Genome-Engineered lines

After establishing the cellular phenotype in engineered A549 cells, we examined IDS enzymatic activity in patient-derived fibroblasts with naturally occurring IDS variants. Cell line GM13203 had been obtained from a severely affected three-year-old male who was hemizygous for a 1 bp insertion at nucleotide 208 in exon 2 of the IDS gene (c.208insC) resulting in a frameshift and premature stop codon (H70PfsX29). These cells had IDS activity of 0 units/mg protein, whereas cells from the unaffected carrier mother (GM01392) showed IDS activity of 39.9 units/mg, as expected (Figure 2A). Both cell lines were imaged, and their PathScore was measured exactly as with the A549 cells. The patient-derived cells from the affected child had a much higher PathScore both on average (t-test p=4×10⁻¹⁹, Figure 2B), and when considering each individual cell (Figure 2C), compared with the unaffected parent. This result confirmed the robustness of the image-based score across cell lines.

Figure 2. Cellular Morphology of Hunter Patient Fibroblasts.

(A) IDS activity in patient fibroblasts of MPS II affected 3-year-old boy (GM13203) and the unaffected MPS I mother (GM01392). (B) Bar chart of PathScore_LC obtained from images of each patient fibroblast sample. T-Test p=4.6×10⁻¹⁹ from 2 plates, 8 wells per plate, error bars are standard deviation. (C) Cumulative histogram of all cells (9,181 total) from separate replicate wells shows the distribution of PathScores for the control and affected individuals. (D) Pseudo-color micrographs of fluorescent confocal images of fibroblasts with staining of lysosome (white), perinuclear lysosome (teal), and nucleus (red) between control and affected samples. (E) Staining of membrane (white) and nucleus (teal) in control and affected samples. Yellow arrows indicate a decrease presence of membrane inclusions found in mutant cells compared to control cells. Scalebar 50 μm.

Examining the fibroblasts more closely in terms of lysosomal staining (Figure 2D) and membrane staining (Figure 2E) revealed further insights. Lysosomes appeared similar in the two genotypes, although they were spread subtly thinner throughout a large cytoplasm in control cells. The membrane in the affected cells showed decrease in intensity and a reduction in the number of membrane inclusions, suggesting a possible dysfunction in membrane turnover or recycling processes.²⁸ These results imply poorer endocytosis and autophagic pathways in the mutants, which are critical for maintaining cell survival and homeostasis.

Section 3: Recombinant Iduronate-2 Sulfatase enzyme restores wild-type Morphology in pathogenic Mutant A549 cell-lines

We next asked whether the addition of recombinant human IDS enzyme to IDS-mutant A549 cells would reverse the pathogenic phenotypes. One control (Block) and one benign variant (V223I/V223I) isogenic cell lines were compared to the attenuated (P228fsX/P228fsX) and severe (G224R/G224R) isogenic lines. First, we evaluated the IDS activity and confirmed a dose-dependent restoration of IDS activity by recombinant human IDS at 24 hours (Figure 3A). We next asked whether restoration of IDS enzyme activity by exogenous application of recombinant human IDS enzyme was sufficient to ‘correct’ the cellular phenotype. At the higher applied dose (concentration) of recombinant human IDS in the G224R/G224R variant cell line, a noticeable reduction in PathScore_LC was observed when averaging cells among all time points after enzyme treatment (8 wells per condition, Kruskal-Wallis H-stat=26320, p~0 against cells Figure 3B). As anticipated, the cell-based PathScore_LC of enzyme-treated A549 cells showed no significant differences compared with the pre-treatment PathScore_LC for the benign variant and control cell lines. This finding demonstrates that treatment with enzyme may cause measurable corrections in cellular phenotype and suggests that at least some aspects of the cellular phenotype we observed may be caused by IDS deficiency.

Figure 3: Recombinant human IDS restores partial wild-type phenotypes in A549 cells with severe and attenuated IDS variants.

(A) Bar chart showing IDS enzyme activity of cell pellets for isogenic A549 lines with IDS variants of different severities, treated with a functional recombinant human IDS enzyme, 316.7 ug/mL = 14,471 units/ml (H), 31.67 ug/mL = 1,447 units/ml (L), or an artificial CSF vehicle (V) for 24 hours. One unit of activity = 1 nmol of converted substrate per hour. (B) Bar chart showing the deep cellular phenotype’s PathScore_LC for the same set of variants and conditions, averaging across replicate wells, plates, and time points. Kruskal-Wallis (KW) amongst cells H-stat=26320, p~0 125,838 cells. KW amongst wells H=231, p=3.08×10⁻⁴³, 8 wells/group. Only significant post-test is p.G224R/p.G224R H to V p= 0.00015 (all timepoints). (C, D) Change in %cells pathogenic-like after treatment with recombinant IDS at 72, 96, and 120 hours after treatment. Top panels show change in severe variant (G224R/G224R, C and P228fsX/P228fsX, D) compared to benign variant (V223I/V223I, C and block/block, D) below. Notice that benign and mutant cells are non-overlapping on the y-axis, and that the high dose of IDS enzyme yields 50% normal cells after 120 hours. Brighter colors received the high IDS dose, darkest received vehicle, as shown in the bar graphs (A, B). Each line represents one of three plate replicates. Linear regression on 72 and 120 hours where * < .001, ** < 0.0001. (E) Example micrographs of treated cell lines with nuclei (blue), lysosomes (green), and cell membranes (red). The block-only benign control (top) is compared to the severe KO treated with the vehicle (middle) and a high dose of the IDS enzyme (bottom) at 120 hours. (F) Table listing features that contribute the most to whether high dose treated G224R mutant ‘responded.’ The negative signs indicate negative correlation such that higher values in this feature prevented a cell from responding. Black text is lysosomal features, red are cell membrane, and blue are nuclear features. Lower panel shows the top-ranked responsive feature (interaction term), and how it differs in cells that were vehicle-treated pathogenic (VP) or vehicle-treated benign (VB), and cells that were high dose-treated pathogenic (HP) or high dose-treated benign (HB).

While treatment with recombinant human IDS rapidly restores IDS activity in cells (typically within an hour or less after application), the catalysis of glycosaminoglycan substrate and subsequent restoration of the cellular phenotype may require more time. We next examined changes in the PathScore at varying time points several days after incubation with recombinant human IDS. The severe (G224R/G224R) variant cell line showed maximal correction at 120 hours, with ~50% of the cells displaying pathogenic phenotypes at that timepoint (Figure 3C). The attenuated (P228fsX/P228fsX) variant cell line also showed correction at both the low and high doses (concentrations) of recombinant human IDS, with ~40% of the cells displaying a normal morphology (Figure 3D).

Live-cell fluorescence imaging revealed a reduction in membrane staining and unipolar arrangement of the lysosomes in the severe (G224R/G224R) variant cells, consistent with the elevated PathScore_LC of this line (Figure 3E). We observed a partial restoration of the membrane and lysosome morphology following treatment with the high dose (concentration) of recombinant human IDS. We next examined the cellular features of the individual cells that were ‘rescued’ when treated with recombinant human IDS using a logistic regression model. The interaction terms that had the strongest significance (marking cells that had lower PathScores after treatment with rhIDS) were the cytoplasmic lysosomal intensity variation (Cell Intensity CV, coefficient = 2.07), followed by other lysosomal texture measurements (Cells Skewness,1.94, Cells GLNU, 1.75) (Figure 3F). As expected, when examining the lysosomal intensity variation across the population, it is the lowest in the ‘benign’-like cells (with a low PathScore_LC) and increases in cells treated with the high dose of recombinant human IDS. Interestingly, only membrane stain texture (cells entropy) and nuclei shape (gyration radius) were negative coefficients, in that an increase in these metrics was associated with a more pathogenic phenotype, thereby negatively impacting phenotypic rescue. Together, these findings demonstrate that exogenous recombinant human IDS can rescue a portion of mutant cells’ cellular phenotypes, and that those changes are detectable via cellular imaging.

Section 4 Transcriptional Signature is Distinct amongst IDS variant cell lines, and is not rescued by recombinant I2S enzyme after 120 Hours

Because restoration of IDS activity with exogenously applied recombinant human IDS did not completely rescue the cellular phenotype of IDS variant cell lines, we hypothesized that some aspects of the phenotype may be resistant to correction, and that these “irreversible” disease alterations may manifest in the cell’s gene expression profile. To investigate gene expression differences in Hunter syndrome (MPS II) cells and whether recombinant human IDS restores normal expression profiles, we performed bulk RNA sequencing on a subset of A549 mutant IDS cell lines, with and without recombinant human IDS treatment. Differential expression analysis identified several genes that were upregulated or downregulated in pathogenic genotypes compared to benign (Figure 4A, B). Expression of IDS was downregulated in affected cells compared to controls, as expected. An unsupervised UMAP clustering was performed on the RNA count table to determine, in an unbiased way, the underlying structure between the samples. The UMAP clearly separated the pathogenic and benign mutants, especially along axis 2 (Figure 4C). The mutant line was the strongest determinant of UMAP position, with time-of-treatment having a more modest effect. While rescue with recombinant human IDS led to improvements in biochemical assays and cellular phenotypic PathScore_LC, the UMAP clustering indicated that treatment did not restore the transcriptomic profile to a healthy genetic state. This could indicate cell to cell differences acquired by the lines or indicate a broader impact of IDS beyond its immediate enzymatic activity.

Figure 4. Transcriptional Profiling of IDS mutants and Recombinant Enzyme Treatment.

(A, B) Heatmap-shaded table of differentially expressed genes in pathogenic (P228fsX/P228fsX, G224R/G224X) vs. benign (Block, V223I/V233I) conditions. All treatments and timepoints were pooled for these tables to maximize observations. Genes with both positive (upregulated, A) and negative (downregulated, B) log2 fold changes are listed from DeSeq2 analysis. Mean indicates the mean count of the gene across conditions. (C) UMAP visualization of transcriptomic profiles for selected variants after application of vehicle, low-dose recombinant human IDS, or high-dose recombinant human IDS, separated by time points. Filled markers are treated, while open markers were only vehicle exposed. The size of the marker indicates time after treatment, with the largest markers indicating 120 hours. This experiment had 4 genotypes, 3 doses, 3 time points, and 2 replicates each.

Discussion

In this study, we conducted initial experiments aimed at improving the classification of IDS gene variants, particularly those of uncertain significance (VUS), using cellular phenotyping. Our findings highlight the value of using cellular morphology and phenotypic data to aid in variant classification, thereby providing more nuanced insights into the pathogenicity of IDS mutations associated with MPS II.

The lysosomal accumulation of heparan sulfate and dermatan sulfate glycosaminoglycans occurs in Hunter syndrome and is thought to be the primary driver of the disease process. Abnormal accumulation of heparan sulfate in particular is detrimental to the central nervous system (CNS) and results in a wide range of clinical symptoms.²⁹ Roughly two-thirds of individuals affected with Hunter syndrome have a clinically-evident neuronopathic disease and can experience developmental delays, cognitive decline, hyperactivity, seizures, behavioral challenges, and communicating hydrocephalus.^30,31 Somatic manifestations, which are present in individuals who have non-neuronopathic as well as those with neuronopathic disease, include coarse facial features, growth delay, and pulmonary dysfunction.^30,31 Newborn screening has been developed for Hunter syndrome, and eighteen US states have active screening programs at present.^7,8,32 There is a pressing need to develop methods that will predict which genotypes will result in the severe, neuronopathic form, because therapeutic options might one day be different for those children.

One of the major challenges in classifying IDS gene variants lies in the limitations of the biochemical assays for IDS activity and glycosaminoglycans, which may not reliably distinguish severe, neuronopathic from attenuated, non-neuronopathic phenotypes. In our small selection of IDS variants, we likewise found that the fluorometric IDS enzyme assay effectively differentiated the pathogenic variants from the benign variant and control cell lines. However, enzyme activity did not distinguish between the attenuated (P228fsX/P228fsX) variant line and the severe (G224R/G224R) variant line. By examining cellular phenotypes using our image-based assay, we found that the PathScore_LC, a multi-feature indicator of cellular disruption, can be a reliable method for distinguishing between pathogenic and benign mutations, as well as varying levels of pathogenicity within the affected mutations (Figure 1G). This multi-dimensional approach, which captures cellular damage through the analysis of features such as cell size, lysosomal texture, and membrane integrity, provides a more complete picture of the molecular consequences of IDS mutations. While PathScore_LC is not validated for clinical use, benign variants tend to cluster at lower values compared with intermediate or sever variants (Figures 1G). These observations will require further validation in larger, independent cohorts.

In vitro studies with A549 isogenic cell lines revealed that both severe and attenuated mutations exhibit significant morphological alterations, including changes in lysosomal and membrane distribution and cell shape (Figure 1C). These differences suggest possible disruptions in intracellular trafficking and cytoskeletal organization, which may underlie the pathogenic effects of severe mutations and are consistent with the MPS II disease state.³³ These phenotypic changes were quantified using PathScores, revealing a clear gradient from benign to attenuated to pathogenic mutations (Figure 1G). This trend closely mirrored IDS enzyme activity levels (Figure 1B), further validating the utility of combining biochemical and phenotypic data in variant classification. Thus, the simultaneous use of multiple phenotypic features can greatly improve classification accuracy, providing a robust framework for distinguishing IDS variants.

Our analysis showed a phenotype in heterozygous cell lines, including G224R/block and P228fsX/WT, as reflected in the higher PathScores observed for these two variant lines compared to homozygous benign and control lines (Figure 1G). We are uncertain whether the cellular phenotype is caused by haploinsufficiency or a dominant, “gain-of-function” effect where the pathogenic mutations, G224R and P228fsX (Figure 1A), on one X chromosome may interfere with the function of the protein produced by the normal allele. This disruption can impact the cellular phenotype. In Hunter syndrome, skewed X-inactivation may result in affected females with one pathogenic IDS variant. We did not explore X-inactivation in our cell lines. However, another possible explanation for the intermediate phenotype seen in these heterozygous variant lines is that adult-onset neurodegenerative disorders have been increasingly recognized in otherwise normal carriers of autosomal recessive lysosomal storage disorders. For example, heterozygous carriers of Gaucher disease may develop Parkinson’s disease despite normal glucocerebrosidase enzyme activity, which is thought to be due to lysosomal dysfunction and impaired protein homeostasis.^34–36 Similarly, heterozygous carriers of NPC1 mutations may show late-onset neurodegeneration, even in the absence of classical Niemann-Pick type C disease pathology.^37–39 Evidence of haploinsufficiency or dominant-negative effects in lysosomal proteins has been observed in other disorders. Hexb haploinsufficiency in an Alzheimer’s disease mouse model was shown to contribute to neurodegeneration, causing detectable brain changes and reinforcing the role of lysosomal dysfunction in disease pathology.⁴⁰ Heterozygous mutations in Cathepsin F (CTSF) cause Type B Kufs disease, an adult-onset neuronal ceroid lipofuscinosis.⁴¹ Similarly, heterozygous pathogenic variants in sulfamidase (SGSH; the enzyme deficient in MPS IIIA) have been suggested as a risk factor for early-onset neurodegenerative disease, with observed phenotypic changes despite the fact that a substantial amount of enzyme activity remains.³⁶ Together, these findings raise the possibility that even when enzyme activity is partially reduced or unaffected, subtle cellular disruptions due to dominant-negative effects or haploinsufficiency may contribute to disease pathology in detectable ways. While this effect is yet to be reported in Hunter syndrome, its presence in other lysosomal storage disorders and neurodegenerative diseases suggests a similar mechanism may be at work with these variants.

IDS activity and PathScores correlated well in patient fibroblasts, indicating that our model can effectively differentiate between control and affected cells (Figure 2A–C). Furthermore, our model may also help identify novel phenotypes in Hunter syndrome, as observed in the reduced intensity and number of membrane inclusions in affected cells (Figure 2E). While we expected to see an increase in the number, size, or intensity of membrane inclusions characteristic of a lysosomal storage disorder, we instead observed a reduction in these inclusions. This suggests a potential defect in membrane turnover or recycling. The lysosome has a myriad of functions, including degradation of membrane lipids, such as sphingolipids and cholesterol esters.^28,42 Defects in membrane turnover can impact these processes, altering cellular function. Additionally, lysosomal distention can increase the size and number of membrane-contact sites with the endoplasmic reticulum, hindering lysosome movement along microtubules to peripheral locations.^43,44 Other features, such as the reduction in intensity, number, and size of lysosomes and the nucleus, further suggest disruptions in membrane trafficking and degradation pathways (Figure 2D–E). These observations emphasize the potential of this model to drive new discoveries in Hunter syndrome.

In the recombinant human IDS rescue experiment with the severe variant (G224R/G224R) cell line, significant correction in phenotype was achieved when high concentrations of enzyme were added compared to vehicle and low concentrations (Figure 3B). The rescue effect was visible in all variants at 96 and 120 hours, with the most pronounced improvements seen in the high-dose treatment of the attenuated (228fsX/228fsX) and severe (G224R/G224R) variants (Figure 3C–D). No significant correction was observed in the attenuated variant cell line, which had lesser pathogenic-like cells to begin with compared to severe (G224R/G224R) cells. This supports the idea that sufficient enzyme supplementation over time can partially restore enzymatic function. Imaging and violin plots further illustrated this, as the severe (G224R/G224R) variant, after high-dose rhIDS treatment, regained some cellular phenotypes resembling the benign Block 2D03 vehicle condition, with increased cell number, intensity, and size (Figure 3E–F). These results indicate that treatment with recombinant IDS may produce dose-dependent effects, and that the effects were most dramatic in the severe variant line. Significant phenotypic improvements were primarily observed in the severe variant, suggesting that enzyme activity alone, or the lack thereof, may not fully account for cellular pathology. Our differential expression results from the RNA sequencing analysis confirm that even though loss of IDS disrupts gene regulation, recombinant enzyme treatment does not fully normalize gene expression (Figure 4A). This is further supported by UMAP clustering, which shows that IDS-rescued cells remain distinct from healthy controls, indicating that transcriptional recovery is incomplete. This misalignment between biochemical correction and transcriptomic recovery raises questions about the extent of cellular rescue and suggests that additional therapeutic strategies may be needed to fully restore cellular homeostasis in MPS II patients. Future studies could explore whether prolonged treatment or combination therapies can further normalize the transcriptome.

This study highlights the utility of integrating cellular phenotyping with biochemical assays to improve IDS variant classification in Hunter Syndrome. While enzyme activity assays distinguish pathogenic from benign variants, cellular phenotyping provides additional resolution, particularly for variants of uncertain significance. Additionally, rescue experiments demonstrated dose- and variant-dependent effects, revealing the potential for rescue treatments that can correct enzyme activity and some cellular features, especially in severe variants. These changes were detectable through imaging and quantified as PathScores using our model. However, our results show that despite improvements in IDS activity, the rescue effects remained limited, as the overall cellular phenotype remained disrupted.

The current findings are based on a targeted set of isogenic engineered A549 cell lines and two patient-derived fibroblast lines, which were sufficient to capture key cellular and biochemical features of select IDS variants. Further studies with additional variants will be needed to fully explore the diversity of cellular responses. Future applications of these techniques could proceed in several directions. First, additional functional cellular studies could expand the analysis to a larger panel of IDS variants, leading to the systematic screening of variants for entry into variant classification catalogues, like ClinVar. Extending these analyses to neurons or other disease-relevant cell types would further the association with neuronopathic forms of disease. It would also incorporate hierarchical or mixed-effects modeling to account for the non-independence of cells nested within plates, wells, and fields of view. Another more clinical approach could be to use non-cultivated cells, such as peripheral blood mononuclear cells (PBMCs), obtained directly from patient blood and used to rapidly study IDS variants. While the present work establishes proof-of-concept for these methods and could inform the development of a cytogenetic clinical assay, translation into a clinical setting would require assay standardization, rigorous quality control, and careful correlation with patient outcomes. Establishing clinical thresholds, reproducibility across laboratories, and robust validation with patient-derived samples would be necessary steps before clinical implementation.

Deeper mechanistic investigations could help us understand why certain variants produce intermediate phenotypes and why transcriptomic recovery remains incomplete, potentially guiding therapeutic development. We also hope to use lysosomal contents and transcriptional regulation to answer these mechanistic questions. Taken together, these observations define the boundaries of the current study while laying the groundwork for subsequent investigations into variant-specific effects and clinical applicability. This work underscores the potential of image-based screening to improve diagnostic accuracy and inform treatment strategies for Hunter syndrome.

Materials & Methods

Variant Selection

Variants of the IDS gene were selected from ClinVar by a medical geneticist (P.I.D.) and sourced to publications in which the patient phenotype was comprehensively described. A total of twelve variants, ranging from benign to severe, were selected (Supplemental Table 1). These include the severe c.670G>C (p.G224R)⁴⁵, attenuated c.673_674insC (p.P228fsX)^46,47, and benign c.667G>A (p.V223I) variants. Additionally, a three-base pair insertion, resulting in the one codon addition of an alanine residue at position 228, was categorized as a VUS and examined in this study (Figure 1A). The Genome Engineering and Stem Cell Core at Washington University School of Medicine (GESC) generated gRNAs XCC415h.h.IDS.sp3 (GCTTATGATACCCAACGGCCNGG) and XCC415h.h.IDS.sp2 (GTGTGGCTTATGATACCCAANGG), along with a control single-stranded donor DNA (ssODN) incorporating only block modifications to prevent recutting for each variant. The twelve selected variants were generated in isogenic A549 cell lines. These cell lines are diploid-triploid and require mutations to be introduced into the IDS gene on one or both X chromosomes. The mutations were engineered using CRISPR-mediated homology-directed repair (HDR). The arrangement of the twelve variants, along with their genotypic details are listed in Supplemental Table 1.

Cell Lines

Hunter Syndrome human fibroblasts (GM01392 and GM13203) were transfected with this construct as well. These two patient fibroblast samples were procured from the Coriell Institute for Medical Research. The control sample is from a 24-year-old mother (GM01392) who is a carrier of MPS I, also known as Hurler Syndrome, and has a normal genotype for MPS II. The affected sample is from a 3-year-old male (GM13203) who is hemizygous with a frameshift mutation in the IDS gene caused by a 1 bp insertion at nucleotide 208 in exon 2 (208insC). This mutation created a premature stop codon, consistent with the severe phenotype (H70PfsX29). Both the A549 and fibroblast lines were kept in an incubator at 37°C with 5% CO2 and monitored daily to assess growth and health.

Ethical approvals and patient consent

This study used established, commercially available cell lines (sourced from ATCC and/or the Coriell Institute) and did not involve human participants, primary human tissue, identifiable human data, or animals. Therefore, ethics committee/IRB approval and informed consent were not required as no human subjects research was performed per the U.S. Department of Health and Human Services definition and Washington University Human Research Protection Office guidance.

A549 and Fibroblast Cell Plating

Conditioned A549 cell lines were transferred to 12-well plates in F-12K media with 10% heat-inactivated FBS, 1% penicillin-streptomycin 1X, and 1% GlutaMAX 100X at 80,000 cells/well. They were allowed to adhere for 24 hours. On the second day, they were treated with 31.67 ug/mL and 316 ug/mL of the synthetic I2S enzyme (in rescue experiments). Cells were incubated for 24 hours with enzyme treatment. Cells from the 12 well plates were then randomly arrayed (to reduce batch variation by averaging out potential plate-related effects) across three 96-well tissue culture treated plates using custom software and plated using a Biomek i5 liquid handling robot. 500 cells were seeded to each well, and 6–8 well replicates were used. After 72 hours, imaging protocols began. The low plating density was to allow for the extensive expansion to the 120 hour rescue timepoints.

Patient fibroblast lines were thawed directly into 12-well plates at 200k cells/well in DMEM media with 10% regular FBS, 1% penicillin-streptomycin 1X, and 1% GlutaMax 100X. After several days of proliferation, cells were passaged to new 12-well plates with the same density and grown for 48 more hours. As with the A549, cells were randomly arrayed across two 96-well tissue culture treated plates. 2,000 cells were seeded to each well, and 6–8 well replicates were used. 48 hours later, staining and imaging began.

Staining and Imaging

Plates with the randomized fibroblast lines were first stained with SPY650 Nuclear dye (Cytoskeleton Inc., CY-SC501) at 1:1000, for one hour. Subsequently, the plates were stained with a master mix of LysoTracker Blue DND-22 (ThermoFisher Scientific, L7525), 50 nM, and Cell Mask Green (ThermoFisher Scientific, C37608), 1:2000, for 30 minutes. Plates with the randomized A549 lines were stained with a master mix of Hoechst 33342 (ThermoFisher Scientific, H1399) at 1:2500, Cell Mask Orange (ThermoFisher Scientific, C10045) at 1:2000, and LysoTracker Deep Red (ThermoFisher Scientific, L12492) at 1:20,000. Plates were rinsed twice with respective media and then immediately imaged on a high content confocal microscope at 20× 0.45 NA utilizing a live cell chamber (GE IN Cell Analyzer 2000). To optimize focus during imaging, each field of view overlapped by 8% of their area. Imaging settings were held constant throughout each experiment. Images were manually inspected to exclude any out-of-focus field of views.

Cellular Phenotype Analysis

Analytical pipeline has been described previously in detail in.^27,28,48 This image-based screening approach has also been utilized by other groups.^49,50 Image files were processed in Molecular Devices InCarta for cell-based segmentation and quantification.²⁷ Segmentation masks were generated on the nuclear channel, and single-cell features including mean and integrated intensity, texture, and morphology were extracted for each fluorescent channel. The resulting cell-level data were compiled by batch using FIVTools (https://gitlab.com/buchserlab/fivetools) and imported into TIBCO Spotfire Cloud for curation and visualization. Within Spotfire, data were gated to include only live nuclei and to exclude segmentation or tracing errors. This included bounds on nuclei area, nuclear intensity, and nuclei shape. Experimental metadata were joined to the per-cell measurements to produce a complete annotated dataset, which was exported as a master file for downstream statistical and computational analyses.

After quantitative features were extracted from the images, binary classification machine-learning models were developed to distinguish wild-type from pathogenic mutant cells. Quantitative phenotypic measurements from individual cells served as the input variables for training. Model training was performed using a subset of the twelve selected variants detailed in Supplemental Table 1. Artificial neural networks (aNNs) were implemented in TensorFlow, which served as the computational framework for model generation. Each aNN was trained to recognize perturbed phenotypes solely based on the cellular features present in the training data. Model training parameters were as follows: dropout = 40%, L1 regularization = 0.00001, batch size = 1024, epochs = 2500, patience = epochs/10, and layer sizes = 51, 72, 61 unless otherwise stated.

To minimize overfitting and prevent the models from memorizing experimental artifacts, several precautions were implemented. Fields of view (FOVs) were restricted to a common range so that differences in cell density could not artificially influence training. Cells were mixed from within the vertical hierarchy of plates, wells, and FOVs. A biologically motivated hold-out strategy was used to assess generalization instead of genotype-wise cross-validation where several clones were withheld from training to evaluate the model’s ability to generalize to unseen examples. The benign-like control training samples included block /block, p.228fsX/WT, p.G224R/block, p.V223I/p.V223I, and the ML II and MPS I fibroblasts. The pathogenic-like training samples included p.G224R/p.G224R, p.228fsX/KO, p.228fsX/p.228fsX, and the MPS II fibroblasts. Additional variants were withheld from training, including p.228A/p.228A, p.G224R/p.228fsX, p.228A/p.228fsX, and p.228fsX/3bp ins.

The overall dataset comprised approximately 91,000 individual cells for training and 39,000 for testing (a 70/30 split). Each model was trained on the same set of samples, balanced by class across experimental batches to minimize batch-related confounding. Plate layout and sample assignment were randomized to prevent potential positional or plate-specific biases. To further limit the risk of overfitting to specific features, feature subsampling was employed: each trained model received a randomly selected subset of features drawn from a curated list of “safe” variables known to not leak condition information. Supplemental Table 3 lists the morphological features used in PathScore_LC, which is referenced in Figures 1–3. Although PathScore_LC contained 87 features, models trained with smaller feature subsets retained the ability to discriminate between pathogenic and benign clones (Figure S3).

IDS Protein Purification

In order to express and purify IDS from cell culture media, nucleotides encoding the anti-Protein C epitope (HPC4) were appended to the 3’ end of the cDNA, and in the process removing the nucleotides for the myc epitope. IDS-HPC4 and GNPT-S1S3 cDNA were co-transfected into Expi293 (Thermo Fisher) cells growing in suspension as per the manufacturer’s protocol S1S3 phosphotransferase is a manufactured version of GlcNAc-1-phosphotransferase, a Golgi-complex enzyme that modifies lysosomal proteins with mannose 6-phosphate glycans for lysosomal targeting.⁵¹ We found that when expressed recombinantly, most of the secreted IDS fails to acquire the mannose 6-phosphate lysosomal targeting signal on their glycan chains, and that this was corrected with co-transfection of S1S3. The serum-free media containing the secreted IDS was collected 5 days post-transfection and purified on a HPC4-agarose matrix (Sigma-Millipore) according to the manufacturer’s protocol. Briefly, the equilibration and binding steps were performed using buffer containing 20 mM Tris-HCl pH 7.5, 100 mM NaCl, and 1mM CaCl₂. The same buffer was used for the wash step except that the NaCl concentration was increased to 500 mM. The proteins were eluted in 1 ml fractions in buffer containing 20 mM Tris-HCl pH 7.5, 100 mM NaCl, and 5 mM EDTA. Protein purity was determined by SDS-PAGE and Coomassie staining of the gels.

IDS Enzyme Assay

Enzyme assays were conducted on isogenic cultures grown in 12-well plates. After trypsinization, between 100,000 and 300,000 cells were collected, pelleted, and lysed for assay. A 1.25 mmol/L solution of 4-methylumbelliferyl-α-l-iduronide 2-sulfate (4-MU-α-IdoA 2-sulfate, BIOSYNTH, Louisville, KY Cat# EM0301) was prepared by dissolving 2 mg of 4-MU-α-IdoA 2-sulfate in 3.36 mL buffer consisting of 0.1 M sodium acetate buffer pH 5.0, 10 mM lead acetate, and 0.02% sodium azide. The reaction was initiated by mixing 10 μL of each sample with 20 μL of the 1.25 mM 4-MU-α-IdoA 2-sulfate solution. The reaction mixture was then left to incubate for 4 hours at 37 °C in a water bath. This allowed any enzyme present in the samples to cleave the sulfate group from iduronide 2-sulfate from the 4-MU-α-IdoA 2-sulfate substrate. After the incubation period, the reaction was terminated using 40 μL of McIlvain buffer (0.4 M Na-phosphate/0.2 M citrate pH 4.5). The samples were then vortexed and centrifuged to form pellets. A second enzymatic reaction was initiated by adding 10 μL of partially purified recombinant alpha-l-iduronidase (IDUA from R&D Systems) to each sample, followed by incubation at 37 °C in a water bath for approximately 24 hours. During this step, IDU cleaved the iduronide moiety, releasing 4-MU, which generated fluorescence detectable by a plate reader. The reaction was then stopped with 200 μL of glycine carbonate at pH 10.7.

Fluorescence was measured using a BioTek Synergy plate reader, with excitation and emission wavelengths set to 360 nm and 455 nm, respectively. Fluorescence values were normalized against a standard curve generated from 4-MU solutions with known concentrations to quantify the enzymatic activity of iduronate-2-sulfatase (IDS) in each sample. Enzymatic activity was expressed as units, with one unit defined as the conversion of 1 nmol of substrate per hour.

The protocol for IDS enzyme assays was adapted from the method described by Voznyi et al. in their 2001 publication, “A fluorometric enzyme assay for the diagnosis of MPS II (Hunter disease).”⁴ The described methodology was applied to measure IDS activity in the 12 A549 cell line variants, patient-derived fibroblasts, and recombinant human IDS rescue experiments.

IDS Uptake for Rescue Experiments

Lysosomal enzymes, such as iduronate-2-sulfatase (IDS), are internalized by cells after binding to the mannose 6-phosphate receptor (M6PR) located on the plasma membrane.^52–54 Purified recombinant human IDS, with a specific activity 45.8k Units/mg protein, was added to the cell culture media at two concentrations: 1,447 units/ml (“Low”) and 10x higher at 14,471 units/ml (“Hi”). Cells were incubated for 24 hours to allow for glycosaminoglycan (GAG) storage accumulation, enzyme uptake, and phenotype manifestation. Cells were then removed at varying time points, 72, 96, and 120 hours, and IDS enzymatic activity was assessed using the IDS activity assay described above. To eliminate any interference from lysosomal enzymes present in the serum, all cells were cultured with heat-inactivated serum to inactivate any external lysosomal enzymes. Cells were then imaged and analyzed following methods outlined above.

RNA-Sequencing (BRB-seq)

Bulk RNA barcoding and sequencing (BRB-seq) uses early-stage multiplexing and sample barcoding to produce 3’ cDNA libraries.⁵⁵ BRB-seq multiplexed libraries were prepared using the Mercurius BRB-seq kit (Alithea Genomics). For each sample, a minimum of 100,000 cells was collected and lysed, and total RNA was extracted using the MagMAX mirVana Total RNA Isolation Kit (Thermo Scientific, Waltham, MA) according to the standard protocol. RNA concentration was measured using the Quant-it RNA HS Kit (Thermo Scientific) and then normalized to a total input of 500 ng. Libraries were sequenced by GTAC@MGI and demultiplexed counts were returned.