Characterization of cytoskeletal and structural effects of INF2 variants causing glomerulopathy and neuropathy

Focal segmental glomerulosclerosis (FSGS) is a common glomerular injury leading to end-stage renal disease. Monogenic FSGS is primarily ascribed to decreased podocyte integrity. Variants between residues 184 and 245 of INF2, an actin assembly factor, produce the monogenic FSGS phenotype. Meanwhile, variants between residues 57 and 184 cause a dual-faceted disease involving peripheral neurons and podocytes (Charcot–Marie–Tooth CMT/FSGS). To understand the molecular basis for INF2 disorders, we compared structural and cytoskeletal effects of INF2 variants classified into two subgroups: One (G73D, V108D) causes the CMT/FSGS phenotype, and the other (T161N, N202S) produces monogenic FSGS. Molecular dynamics analysis revealed that all INF2 variants show distinct flexibility compared to the wild-type INF2 and could affect stability of an intramolecular interaction between their N- and C-terminal segments. Immunocytochemistry of cells expressing INF2 variants showed fewer actin stress fibers, and disorganization of cytoplasmic microtubule arrays. Notably, CMT/FSGS variants caused more prominent changes in mitochondrial distribution and fragmentation than FSGS variants and these changes correlated with the severity of cytoskeletal disruption. Our results indicate that CMT/FSGS variants are associated with more severe global cellular defects caused by disrupted cytoskeleton-organelle interactions than are FSGS variants. Further study is needed to clarify tissue-specific pathways and/or cellular functions implicated in FSGS and CMT phenotypes

www.nature.com/scientificreports/ distribute in the proximal N-terminal half of DID (Fig. 2). The p.T161N and p.L162P mutations occurred in a border region that separates the mutation clusters of CMT/FSGS from the FSGS subgroup 5 .

Structural modeling of INF2 variants.
To evaluate the impacts of the mutations on the structure of human INF2, we first modeled the pathogenic variants in the DID using the crystal structure of the mouse mDia1 DID/DAD complex as a basis 16,17 . The side chains of hydrophobic residues of the DAD domain appeared to interact with the DID-binding pockets, which are formed by the α-helices α7, α10, and α13 of the armadillo repeats(ARM) (Fig. 3, Supplementary Figure S2, S3).This structural model was consistent with that derived from the deep learning-based AlphaFold2 program 18 (Supplementary Figure S4). In this model, The CMT/FSGS variants mapped near the central core of the hydrophobic DAD-binding pocket, whereas the FSGS variants occurred in more external portions. We next evaluated DID/DAD interactions in pathogenic INF2 variants by structure modeling. PDBsum interaction plots revealed that the DID-DAD interface of pathogenic INF2 variants has appreciable changes in both amino acid pairing and the number of hydrogen bonds relative to wild-type (Supplementary Figure S5). However, the interface area and solvent free energy (ΔG) did not significantly differ and does not seem to alter overall binding affinity of the DID/DAD interaction. (Supplementary Table S3) 19 . Modeling analyses suggest the structural effects of INF2 variants may be subtle changes in interacting pattern and the need of the evaluation for their molecular dynamics.

RMSD analysis for INF2 variants.
We next investigated the effects of INF2 missense mutations on dynamic flexibility and stability of the DID-DAD complex. The molecular dynamics of INF2 proteins was compared between the wild-type and five pathogenic variants by the comparative modeling using the Rosetta program and molecular dynamics (MD) simulation using the GROMACS program. We first accessed overall Calcium-dependent cellular actin organization (CaAR) was as described in a previous study 10 . CaAR measured in HeLa INF2 knock-out cells is shown as the percentage of cells with Max(R) values > 14. 9. The values range from 100% (in control and cells expressing benign variants) to 0% (in cells expressing pathogenic variants). CMT, Charcot-Marie-Tooth Disease, FSGS, Focal segmental glomerulosclerosis, ND, not determined, NA, not applicable. www.nature.com/scientificreports/ structural stability of the wild-type DID-DAD complex by analyzing the root-mean square deviation (RMSD) trajectories of their backbone atoms during the 500 ns simulation. The RMSD increased over the initial 50 to 100 ns up to around 0.5 nm and reached a plateau till the end of simulation time period 500 ns (Fig. 4). Five INF2 variants displayed the RMSD trajectories that got equilibrated between the 50 ns to 100 ns period similar to the wild-type, with the exception of T161N showing a gradual increase of RMSD that get stabilized around 200 ns. When the trajectories were compared at the steady state, the variants exhibited a variable degree of changes in overall average RMSD: G73D variant showed the largest average RMSD value (0.71 nm vs wild-type 0.52 nm), while L69P exhibited minimally increased level compared with the wild-type (0.55 nm) (Supplementary Table S4). In contrast, the other three (T161N, R218W, E220K) displayed subtle decrease in RMSD (0.44-0.46 nm) in relative to the wild-type. The variation in global RMSD of the INF2 variants suggest that the conformational changes may take place in the protein complex throughout the simulation.

RMSF analysis for the INF2 variants.
To gain more insights behind the altered stability/flexibility of the DID-DAD complex, we estimated fluctuation of the individual amino acid residues within the complexes by measuring root-mean-square fluctuation (RMSF) for the backbone atoms of wild-type INF2 and variants. RMSF estimates the average deviation of each residue from the reference position within the staring minimized structures over the time of MD simulation. RMSF analysis of the wild-type INF2 DID-DAD complex revealed noticeable fluctuations in the four clusters: two highest fluctuating regions encompassing residue 80-96, and 152-164 (> 0.3 nm deviation from the reference positions), while the other two moderately fluctuating regions at residue 180-194, and 208-220 (around 0.1-0.2 nm) (Fig. 5). These four clusters with higher RMSF correspond to the loop segments connecting α helices of DID domain, the conformation of which is reported to be critical for interaction with the DAD domain (Supplementary Figure S2-S6) 16 .
Comparison of RMSF data between wild-type INF2 and variants revealed variations in residue dynamics in the DID domain for the pathogenic variants. First, in the first cluster residues 80-96, L69P variant showed increased amplitude, while T161N displayed broadening at the peak of fluctuations. The other two variants (G73D and E220K) showed mild decreased amplitude of fluctuation. The remaining one (R218W) exhibited subtle or indistinguishable differences for this peak. Second, in the second cluster residues 152-160, all the variants (G73D, T161N, R218W, E220K) lost the normal fluctuation peak, with exception for the L69P variant preserving the flexibility in this region. Our observations indicate that the pathogenic INF2 mutations could cause variable changes in conformational flexibility of the DID domain.
Flexibility of Residues at the DID-DAD interface. We next evaluated the residue interactions at the DID-DAD interface of INF2 variants. We first compared the flexibility of Glu968 residue at the DID-DAD interface by overlaying two snapshots from the dynamic INF2 structure simulated at 0 ns and 500 ns, the latter of which represented the last frame of trajectories reaching their steady state of dynamics. www.nature.com/scientificreports/ images showed that the Glu968 residues at the interface have a variable mobility among the pathogenic variants, compared with the wild-type (Fig. 6). We then assessed the changes in dynamics of the residue pair at the DID-DAD interface. We first compared the time evolution of distance between sidechain of the backbone Cα atoms in Gly62 and Glu968 at the DID-DAD interface (Supplementary Figure S7). The residue pair distances were traced during the 500 ns simulation, a long enough to attain a stable conformation. The time evolution graph revealed that pathogenic variants fall into three classes with regards to their trajectories: One variant L69P showed the longest average inter-residue distance between Gly62 and Glu968 than the wild-type (1.32 vs 0.67 nm, Supplementary Table S4), while the other two (G73D, E220K) exhibited equal or smaller distance compared to the wild-type (0.65 and 0.52 nm, respectively). The remaining two T161N and R218W showed an intermediate distance level that fluctuates over time (average 0.70, 0.73 nm respectively).
We next compared the variation in residue contacts at the DID-DAD interface in total 500 simulated structures captured every 1 ns during the 500 ns-MD simulation between the wild-type and INF2 variants by measuring the dissociation frequency. Dissociation is defined as the distance between the sidechains of Gly62 and Glu968 deviates more than 3-5 Å from the reference position (at 0 ns) during the 500 ns simulation. L69P variant showed Asn110 N202 R218 E220 Figure 3. Structural modeling of human INF2. A 3D structure of human INF2 modeled on the mDia1 DID-DAD complex crystal structure (PDB 2F31) 16 using the Rosetta program. (A) Locations for ten pathogenic variants in the DID are illustrated as ribbon diagrams; Blue: L69P, G73D (G73V), Orange: V108D, Magenta: G157R, T161N, L162P, Pink: N202S, R218W, E220K. Based on this model, the α helix of the N-terminal DAD (turquois) is predicted to contact the core pocket of DID. Side chains of selected residues involved in the DID/ DAD interactions are shown in stick form: N110 and Y153 in DID, E969, V972, and L976 in the DAD. The central α helices, α7, α10, α13, from the 2nd, third, fourth armadillo repeats respectively, create the DID-binding pockets. (B) Core DID-DAD interaction. The DID is shown as a surface representation rotated horizontally by 180˚. The residues E969, V972, L976 in the N-terminal DAD are predicted to make direct contact with the surface-exposed partners N110, Y153 in the hydrophobic binding pocket of DID. www.nature.com/scientificreports/ the most frequent dissociation (77% vs 13% for wild-type), while G73D had conversely the lowest value (0%). T161N and R218W exhibited a mild to moderate higher dissociation rate (37%, and 19% respectively), and E220K displayed slightly lower frequency (5%) compared with the wild-type (13%) (Supplementary Figure S8, S9). Collectively, our MD simulations of the pathogenic variants reveal variable changes in average inter-residue distance and their fluctuations at the DID-DAD interface of the autoinhibitory complex. The observations suggest that mutations could perturbate native residual contacts, leading to the overall less stable structure.   Residue of Cα   Figure S10). In HeLa cells, wild-type INF2 had a diffuse localization throughout the cytoplasm with a fine reticular, ER-localized pattern and perinuclear Golgi accumulation 8,21 . Actin stress fibers formed a thick parallel array that spanned the cells and was similar to that in un-transfected cells 22,23 .
In contrast, HeLa cells expressing CMT/FSGS variants (G73D, V108D) exhibited an altered INF2 subcellular expression that was unevenly distributed in a coarse granular pattern having fewer perinuclear Golgi clusters (Fig. 7). Notably, HeLa cells expressing CMT/FSGS variants (G73D, V108D) were occasionally elongated along the long axis, reminiscent of cells expressing constitutively active mDia1 20 . In these fusiform-shaped cells, some INF2 puncta accumulated at the peripheral pole (Fig. 7). Quantitative analysis showed that cells expressing CMT/FSGS variants have significantly fewer actin stress fiber filaments relative to cells expressing wild-type  Fig S12). HeLa cells expressing variants associated with FSGS alone (T161N, N202S) retained the diffuse ER pattern with Golgi accumulation similar to wild-type INF2. However, these cells had fewer actin stress fibers than those expressing wild-type INF2, indicating that FSGS variants have a cytoskeletal dysregulation that is between that seen for the wild-type and CMT/FSGS variants (Figs. 7, 8 and 9). The severity of actin disorganization in cells expressing the FSGS INF2 variant was further verified by automated-image classification using the deep learning tool ALEXNET. After training the neural network on a manually-labeled representative dataset (normal vs. fewer cables), the test images were automatically processed and classified into the two categories. Cells expressing T161N were principally classified into the "fewer stress fiber" category, while those expressing wild-type INF2 were overall judged as having a normal pattern (p < 0,001, Supplementary Figure S13).
Western blot analysis of lysates from HeLa cells expressing wild-type INF2 revealed a single major band having an apparent molecular mass of 170 kDa. The proteins were less abundant for FSGS-causing variants (T161N, N202S) than for the wild-type. Moreover, in cells expressing the variants associated with CMT/FSGS (G73D, V108D) the INF2 protein levels were barely detectable, suggesting that these variant proteins, particularly of the CMT/FSGS subtype, were more labile and thus more susceptible to degradation than wild-type (Fig. 7B, Supplementary Figure S14). Taken together, cells expressing CMT/FSGS variants displayed more prominent alterations in actin organization, cell shape, and organelle arrangement (i.e., altered ER pattern with Golgi dispersal) than FSGS variants, which likely reflects the more pronounced defects in their subcellular distribution as well as the changes in INF2 protein stability.

INF2 variants alter mitochondria morphology and distribution.
We next examined mitochondria morphology and distribution in fixed HeLa cells expressing c-Myc tagged INF2. Cells expressing FSGS-type variants (N202S) had mitochondria that were mostly distributed in the perinuclear region in a pattern similar to that seen for cells expressing wild-type INF2 (Fig. 9). In contrast, cells expressing CMT/FSGS variants (G73D and V108D) displayed a prominent change in mitochondria distribution, with a tendency toward peripheral localization with longitudinal, parallel alignment of some mitochondria along the disorganized actin filaments (Fig. 9). Cells expressing the T161N variant exhibited an intermediate phenotype in terms of shape and subcellular distribution in which mitochondria had a predominantly perinuclear localization with some aberrant enrichment at the cell periphery (Fig. 9).
We next examined the mitochondria morphology in living cells transfected with INF2 variants using the T161N variant as a representative pathogenic variant and labeling of the mitochondrial network in live cells using MitoTracker. In cells expressing wild-type INF2, mitochondria had a normal tubular pattern with perinuclear accumulation. In contrast, expression of T161N variants altered the mitochondrial morphology from the normal tubular pattern into a globular pattern with a fragmented appearance (Fig. 10, Supplementary Figure S15 INF2 variants alter microtubule and mitochondria organization. We next compared the microtubule arrangement of HeLa cells expressing N-terminal GFP-tagged wild-type INF2 and pathogenic variants. In untransfected cells and those expressing wild-type INF2, microtubules generated a radial array that spread evenly throughout the cytoplasm (class 1, in Fig. 12). This pattern suggests normal nucleation and polymerization from a perinuclear, microtubule organizing center. Meanwhile, cells expressing pathogenic INF2 variants exhibited a disorganized microtubule network. Cells expressing the T161N variant showed an uneven, mixed pattern of microtubule disorganization in which the cortex was filled with longitudinally aligned, thick microtubule bundles and a sparse distribution of normal thin, radiating microtubule arrays (class 2). Cells expressing G73D variant typically exhibited global dysregulation in which most microtubules were disorganized as bundles aligned parallel to the cell long axis and lacking polarity (class 3).
Mitochondria double-labeling showed that, in sharp contrast to the predominant perinuclear localization seen in untransfected and cells expressing wild-type INF2, mitochondria in cells expressing the G73D or T161N INF2 variant had mitochondria that were mis-distributed to the cell periphery, suggesting defective mitochondria trafficking along microtubules.
Quantitative classification of the morphologic patterns into three subgroups revealed that approximately 70% of both wild-type and un-transfected cells exhibited randomly-radiating microtubules (class 1). In contrast, about 60% of cells expressing T161N variant exhibited a partially disorganized microtubule pattern (class 2), whereas about 70% of G73D variant-expressing cells displayed a global microtubule disarrangement (class 3, n = 200 cells for each group) (Fig. 12). The observations indicated that pathogenic INF2 variants promoted a disorganized microtubule architecture and mitochondrial distribution, and the CMT/FSGS variants cause more severe disorganization than the FSGS variants.  Histogram showing a proportional distribution of F-actin phenotypes. Immunofluorescence images (n = 100) were visually inspected and scored by two independent investigators. The proportional distributions of actin phenotype categories were significantly different between wild-type INF2 (WT) and each variant, and between FSGS-and CMT/FSGS-causing variant subgroups. Inter-subgroup differences were analyzed by Fisher exact test with multiple testing correction: not significant (ns) or significant (asterisks). www.nature.com/scientificreports/

Discussion
In the present study, we explored the molecular mechanisms by which INF2-DID mutations between residues 57 and 184 cause two subsets of disorders, CMT and FSGS. Structural modeling and molecular dynamics simulation revealed no consistent differences between the variants associated with the two distinct phenotypes. Our expression studies disclosed that both CMT/FSGS and FSGS variants share a common mode of action that principally affects actin-microtubule organization. Such cytoskeletal disorganization perturbs global cell characteristics such as the shape and position of organelles. Variants associated with CMT/FSGS have a more prominent cytoskeletal effect than variants associated with FSGS alone and thus in turn confer more severe mitochondria defects.  Table S6). Thus, in a majority of the INF2 disorders, renal diseases are a common denominator in the clinical phenotype. However, further genotype-phenotype correlation study is necessary to verify the hypothesis that podocytes might be more susceptible to INF2-induced cytoskeletal aberrations than neurons 9,19 . In our study cohort, patients with the dual CMT/FSGS phenotype manifested the two diseases nearly simultaneously and progressed much faster to ESRD than those with FSGS alone 24 . These clinical observations suggest that patients with dual CMT/FSGS suffer from more severe, global cellular damage, which, like patients with FSGS, likely arises from a common mechanism that primarily affects cytoskeletal organization. Consistent with this possibility, results of expression studies confirmed that cells expressing CMT/ FSGS variants indeed had more prominent cytoskeletal disarrangement and organelle perturbation compared to FSGS INF2 variants, and the disorder associated with the CMT/FSGS variant would exceed the critical threshold to trigger disease in both Schwann cells and podocytes (Fig. 13). However, a considerable phenotypic diversity has seen reported for the INF2 variants. Further study with a larger mutational survey is necessary to understand the genotype-phenotype correlation.

Structural basis of INF2 disorders. Structural modeling revealed that the INF2 mutations causing CMT/
FSGS were exclusively located in the concave pocket of the DID that forms the DAD binding interface and is critical for self-autoinhibition 5,8,9 . Our structural studies supported that the INF2-DID mutations could impair DID-DAD interactions. First, the structural modeling of pathogenic INF2 variants showed changes in amino acid pairings and the number of hydrogen bonds, which are two key elements for the DID-DAD interaction. However, our estimation for the free energy showed no observable difference between CMT/FSGS and FSGS variants, in accordance with previous studies 19 . Second, our molecular dynamics simulations, assuming an aqueous solution, revealed that all the pathogenic INF2 mutations could alter the flexibility and stability of the DID-DAD complex. We found that the mutations affect overall structural stability by RMSD analysis and cause variable changes in the residual dynamics by RMSF analysis. In order to estimate the impact of the mutations on the protein dynamics accurately, we performed the MD analysis for an adequate time (500 ns). Such MD simulations are long enough for the protein to explore conformational space and settle the most stable one. This allowed precise estimation of amino acid interactions, e.g., the changes in side-chain interaction over time (Supplementary Figs. S6, 7,8,9). The MD analysis revealed a variation in the interacting residue flexibility among the pathogenic variants. Our RMSD and RMSF analyses disclosed a trend that CMT/FSGS variants may cause more pronounced changes in the flexibility of the DID-DAD complex, compared with the other INF2 variants (Figs. 4, 5 and 6, Supplementary Figure S6-S9). However, we could not find any consistent trends of MD that might explain the difference of phenotypic severity between severe dual CMT/FSGS and mild single FSGS disease. This is also partly because of our limited numbers of variants analyzed and variability in structural effects of mutations. Specifically, some local residue fluctuations may directly affect the conformation of binding pockets, while the others may secondarily influence the size and the curvature of the DAD recognition surface and its neighbors. For example, either increase (destabilized) or decrease (over-stabilized) in the residue flexibility of DID domain, could sterically hinder the local DID-DAD interaction, thereby affecting the binding affinity and/or frequency of the DID-DAD interfaces 25 .
We found a remarkable high RMSF for the top three clusters in the DAD-binding regions of the DID in wildtype INF2 at residues of 80-96, 152-164, and 180-194. Each region corresponds to the flexible loop structures connecting of α helices in the DID-binding pockets, which are assembled with the central helices (α7, α10 and α13) from the second, third, and fourth armadillo repeats, respectively (Supplementary Figure S3, S6) 16,25 . Those fluctuating loci are located in the vicinity of interacting residues, 107-115, 146-152, and 203-206, those of which are predicted to form the DAD-interacting pocket 16,25 . The super-helical structures formed by armadillo repeats determines the curvature of the protein that suits for the peptide binding. Numerous studies show that armadillo repeats formed by DID α helices allow these proteins to have various interaction partners and functions in the cell and are conserved across eukaryotic kingdom 16,25 . Our results together with others indicate that the DID mutations could alter the native interaction pattern of neighboring residues in the vicinity of mutations, thereby compromising the protein stability and flexibility of the DID-DAD autoinhibitory complex. Further MD simulation, e.g. RMSF determination including more pathogenic variants and comparison with benign variants, will help understand which individual residues might play a key role in the protein complex structural fluctuations.  Figure S10). The deleterious effects of the CMT/FSGS variants might be due to their reduced INF2 dosage as was suggested by results of western blotting  www.nature.com/scientificreports/ showing that protein levels were decreased for the CMT/FSGS variants relative to wild-type. Such a trace amount of the CMT/FSGS variants could even trigger disease 10 . Alternatively, CMT/FSGS variants affecting the residue 57 to 184 of DID could impair preexisting molecular interactions or pathways specific to neuronal tissues 10 . For example, CMT/FSGS variants may preferentially interfere with acute cell adaptation in response to stress 10 . Previous studies showed that INF2 is indeed involved in remodeling the actin architecture during tissue repair after injury 10,11 . INF2 mutations causing CMT/FSGS may thus render the affected tissue more susceptible to injury due to delays in reparative processes needed to recover from injury. Previous studies revealed that pathogenic INF2 variants disrupted calcium-induced, transient actin bursts in living cells and could alter long-standing transcriptomes regulated by serum response factor that in turn dysregulate other cell functions beyond the cytoskeleton 11,26 Further study is needed to define the tissue-specific mediators that act in podocytes and/or Schwann cells 9 .
Microtubule disorganization in INF2 disorders. We found that the extent of microtubule network disruption is proportional to the severity of actin disorganization. INF2 is known to colocalize, directly bind, and mediate both bundling and stabilization of microtubules [27][28][29] . Therefore, two mechanisms for the deleterious effects of FSGS/CMT variants are plausible: INF2 mutations could disorganize microtubule networks through direct effects on microtubules and/or have indirect consequences arising from altered actin-microtubule interactions. Here we found that HeLa cells expressing pathogenic INF2 variants tended to have fusiform cell shapes with bipolar elongation (Fig. 7, 9), as well as an unusual parallel arrangement of microtubules. These cell features are reminiscent of cells expressing constitutively active mDia1 variants 14,20 , in which alterations in microtubules primarily drive cell elongation. In contrast, previous studies with INF2 knock-out cells demonstrated that INF2 plays a key role in lamellipodia formation of the cell cortex 30 . Since a range of microtubule crosstalk with actinbased structures in the cell cortex is essential to stabilize microtubules 27,29 , we speculate that in cells expressing INF2 variants, microtubules needed to capture cortical actin may be defective. The mechanisms likely involve capping of faster-growing plus-end microtubules through interaction with different formins and other "stabilisome" proteins comprising plus-end-tracking proteins and scaffolding elements (e.g., IQGAP1) 19,27-31 . Actin disorganization in INF2 disorders. Given the large excess of actin monomers (~ 100 μM) vs. regulators (0.05-4.21 μM), cells must have a tight spatio-temporal regulation of actin assembly to maintain cell polarity and shape, particularly during active proliferation and differentiation 32 . For this reason, INF2 activation must be confined to specific regions, usually at the cell cortex 6 . Here we showed that cells expressing pathogenic INF2 variants had reduced levels of actin stress fibers and cortical actin, consistent with previous reports 8,9 . Notably, the actin disorganization was morphologically more pronounced in cells expressing CMT/FSGS variants than FSGS variants, which is also consistent with the prior study 10 . The degree of cytoskeletal perturbations caused by the pathogenic INF2 variants in our fixed cell immunocytochemistry correlated with the severity of defective Ca 2+ -induced actin reorganization (Ca-AR) reported in live HeLa cells 10,26 . However, we saw no noticeable changes in filopodia in our fixed cells expressing INF2 variants, which is in contrast to prior observations indicating that constitutively active INF2 proteins increase the length of filopodia 10,14 . Our data, together with that of other investigators, indicate that CMT/FSGS variants are associated with more pronounced disorganization of actin-networks than FSGS variants. Live-cell imaging will help to clarify how INF2 mutations affect actinremodeling dynamics.
The diminished actin cable/bundles in HeLa cells expressing INF2 variants is consistent with some prior studies 8,9 , but conflict with other studies that showed increased actin filaments at the ER and/or cytoplasm of osteosarcoma-derived U2OS cells 12,17 . Histologic analysis of biopsied sural nerves from patients with CMT/FSGS associated with INF2 mutations revealed that non-myelinating cells produce filopodia-like supernumerary protrusions with abnormal actin accumulation in the cytoplasm, suggesting the presence of global actinopathy. INF2 has a unique ability to promote both polymerization and depolymerization of actin 7,12,17,21,33 (Supplementary Figure S17). Several in vitro studies with biochemical assays as well as cell transfection showed that pathogenic INF2 variants are functionally constitutively active 7,10,12,14,17,31 . However, the net effect of this constitutive activation on local actin assembly is unclear. We propose several explanations for the differences between our results and previous results. First, regulation of INF2 activity by autoinhibition in vivo may differ from that reported in vitro. Biochemical studies in vitro showed that, unlike other formins, binding of G-actin to the WH2-DAD domain relieves INF2 autoinhibition to activate actin polymerization 17,21 . In contrast, INF2 autoinhibition activity in vivo may be modified by local cellular actin monomer concentrations 29 . At lower concentrations of actin monomers than physiological levels, INF2 DID-DAD interactions inhibit polymerization, but at higher concentrations there is no inhibition 17,29 . G-actin monomer concentrations within the cell could dynamically change (1-100 μM range) in response to cofactors (e.g., actin monomer sequestering proteins like profilin and thymosin β4) capping proteins, and actin-acetylation 7,27,32 . The high affinity of INF2 WH2-DAD for G-actin allows it to serve as a sensor for subtle physiological oscillations of cytosolic G-actin levels, and to fine-tune actin polymerization in response to the cellular milieu 27 . Moreover, cellular INF2 inhibitors like CAP1 and CAP2 7,17 , as well as local demand for morphologic adaptations, may also modify actin nucleation and elongation 34 .
Second, INF2 activity would generate short and transient actin filaments 33 . In vitro, INF2 biochemically rapidly depolymerizes actin filaments, but whether a similar activity occurs in vivo where INF2 is abundant relative to actin, is unclear 12,17 . The severing/depolymerizing activity of INF2 could be engaged in response to the actin burst.
Third, actin disorganization caused by INF2 variants may be modified by aberrant heteromeric interactions. INF2 DID variants affect not only the actin polymerization/depolymerization equilibrium, 33 but also crosstalk with interacting partners including members of the Rho family of small GTPases 9,19 (Supplementary Figure S18). www.nature.com/scientificreports/ Inter-formin interactions are also important: mDia-DAD competes with actin monomers for INF2-DID binding in trans 11,29 , thereby negatively regulating mDia activity. Thus, pathogenic INF2-DID variants could diminish normal down-regulation by Rho-mDia and in turn promote mDia-mediated actin polymerization 11 .

Role of mitochondria deficits in pathogenesis of INF2 disorders. Mitochondrial abnormalities can
have significant impacts on phenotypic variability as well as severity of INF2-disorders. Here we found exaggerated fragmentation and peripheral misdistribution of mitochondria in cells expressing INF2 variants. Notably, these mitochondria defects were more pronounced in cells expressing CMT/FSGS variants compared to those expressing FSGS variants and the severity of mitochondrial changes correlated with the degree of cytoskeletal disorganization. Our results thus suggest that the mitochondrial defects likely reflect the dysregulated actinmicrotubule network. In cells expressing INF2 variants, mitochondria tended to shift towards the cell periphery and some mitochondria aligned in parallel with disorganized microtubules. The mitochondrial misdistribution induced by INF2 variants may be due to microtubule dysfunction. Microtubules guide transport of mitochondria to specific subcellular areas such as cellular processes of neurons and podocytes 35 . Furthermore, microtubule networks have a global function that positions the Golgi complex near the centrosome and spreads the ER to the edge of cells 36 . Defective mitochondria fusion-fission at ER-mitochondria contact sites may also impair mitochondrial trafficking and tethering. Alternatively, changes in mitochondria distribution may be associated with an intrinsic property of INF2 variants that perturbs its interactions at the ER and mitochondria. Consistent with results of our expression study, expression of the constitutively active INF2-A149D variant in HeLa cells increases mitochondria fission while reducing cell mobility and fusion events, as was seen in a prior study using U2OS cells 12,17 . Mitochondria undergo continuous fusion and fission, the balance of which determines their morphology. INF2 plays a supportive role in mitochondria fission possibly by driving initial mitochondria constriction, which facilitates Drp1-driven secondary constriction 12 . The regulation of mitochondria dynamics also serves diverse cell functions including mitochondrial motility and trafficking, as well as interactions with other organelles such as the ER 37,38 . Up to 20% of the mitochondrial surface is in close apposition with ER membranes. This interface functions as a crucial hub for calcium signaling, apoptosis, autophagy, and lipid biosynthesis, and is thereby closely associated with cell life and death 37 .
Podocytes and peripheral nerves have a higher energy demand relative to other cell types and thus must maintain mitochondria dynamics and distribution along their highly specialized long and/or multiple processes. Importantly, mutations in genes that are involved in mitochondrial fission (e.g., MFN2, OPA1 and GDAP1) are a major cause of CMT 39,40 (Supplementary Table S7). In a manner similar to INF2, MFN2, encoding a mitochondria outer-membrane protein enriched at ER-mitochondria contact sites, causes an axonal neuropathy (CMT2A2) associated with disruption of ER-mitochondria interactions and mitochondrial trafficking 39,41 . Previous histological assessment of biopsied sural nerves in patients with CMT/FSGS induced by INF2 mutations showed no mitochondrial abnormalities, unlike those of CMT arising from MFN2 mutations 42 . A significant fraction of axonal CMT genes are related to mitochondrial mobility, suggesting that the organelle trafficking defect is a central mechanism for CMT 38 (Supplementary table S7).

Conclusion.
Our study with cells expressing INF2 variants suggests that mitochondrial abnormalities may be a critical consequence of actin and microtubule network disorganization. These changes together can affect the viability of both podocytes and peripheral neurons (Supplementary Figure S19). Further study to determine tissue-specific mechanisms underlying INF2 disorders could lead to more rational therapeutic interventions for CMT/FSGS and pave the way for development of targeted and personalized medicines.

Materials and methods
Patients. All procedures were in accordance with the ethical standards of the committee on human study (institutional and national) and with the Helsinki Declaration. Informed consent was obtained from all subjects and/or their legal guardian(s). We enrolled 20 Japanese families and 50 sporadic cases with FSGS 43 . FSGS was defined by the presence of segmental sclerosis in some glomeruli, and the families were ascertained based on at least one individual of biopsy-proven FSGS and at least one additional blood relative affected by proteinuria (defined as proteinuria equal or higher than ++ or a urine albumin/creatinine ratio greater than 300 mg/gCr), progressive renal failure, or biopsy-proven kidney disease [1][2][3][4][5] . We excluded all cases of secondary FSGS because of systemic diseases (obesity, hypertension, HIV infection, and so on) or other heredity kidney diseases (Alport syndrome, thin basement membrane disease, Fabry disease, and so on).
Mutational analysis. Genomic DNA was isolated from the peripheral blood cells using QIAamp DNA Blood isolation kit (Qiagen). We investigated in a cohort of 50 families with SRNS with or without peripheral neuropathy 43 . Index patients were screened by target panel or whole exome sequencing. Co-segregation of variants was confirmed by sequencing for family members. All rare and non-synonymous coding variants were examined. Rare variants were defined by the minor allele frequency less than 1% in the public databases. Regarding the pathogenicity of the variant, damaging effects were predicted by several algorithms including the PolyPhen2, SIFT, M-CAP, LRT, and the Mutation Taster. A systematic, tiered approach was applied for analysis and interpretation of sequence data. Called variants were interpretated and classified according to recommendation by American College of Medical Genetics and Genomics (ACMG). Variants classified as pathogenic or likely pathogenic (category 1. or 2. according to ACMG) were further analyzed. Structural and molecular dynamics analysis. 3D structure modeling. All the variants we identified lie within the DID, which comprises an autoinhibitory domain in the diaphanous formin. In this context, we utilized the structures of the self-autoinhibited mDia1 DID/DAD complex (PDB:2F31), in which the N-terminal DID interacts with the C-terminal DAD, thereby serving the autoinhibition of mDia functions 16 48,49 . Electrostatic interactions were calculated using the particle mesh Ewald (PME) method with a cutoff radius of 1.0 nm. Van der Waals interactions were also cut off at 1.0 nm. The LINKS algorithm was used for bond length fixation. The Steepest Descent method was used for energy minimization, the V-rescale method was used to control the equilibration temperature to 310 K, and the Parrinello-Rahman method was used to maintain the pressure at 105 Pa. First, the energy of the system was minimized to 1000.0 kJ/mol/nm by energy minimization, followed by 200 ps of NVT equilibrium and 800 ps of NPT equilibrium, and then a 500 ns production run was performed. The trajectory output was extracted every 1 ns, where the time-step of 2 fs was set for the all simulations (500,000 steps). We compared the molecular dynamics between wild-type INF2 and pathogenic variants by capturing and aligning the images with VMD software (http:// www. ks. uiuc. edu/ Resea rch/ vmd/). Tracing for the dynamic changes in positions of residues in the DID and DAD interface during 500 ns (250,000,000 steps). For the RMSF analysis, the results after 100 ns, when the RMSD of the MD calculation stabilized were used 50 .
Cellular phenotype imaging and analysis. Cell lines. HeLa cells (RCB0007, RIKEN BRC) or COS-7 (RCB0539) were grown in DMEM medium (Wako, 043-30085) with 10% fetal bovine serum. The cells were maintained under normal culture conditions with 5% CO2 and 37 °C of temperature and medium was changed every 48 h until reached 80% confluency. At this point the cells were gently washed with PBS (−) and trypsinization using 1 ml TrypLE Express (Gibco, 12605-101).
Immunofluorescence staining for actin filaments. Cells grown on coverslips (1 × 10 5 cells in 12 well plate) were transfected with 1 μg of plasmids constructs, either wild-type INF2 or pathogenic variants, by use of TransIT-LT1 (Mirus, MIR 2300). After 8 h transfection, cells were fixed in 4% PFA in PBS for 20 min at room temperature. After rinsing in PBS, cells were permeabilized with 0.1% Triton X-100 for 10 min, followed by blocking in 5% goat serum (Wako, 143-06561) for 1 h at room temperature. cMyc-tagged INF2 was visualized by primary monoclonal anti-cMyc (1:1,000, Sigma, #4439) antibody and fluorescent-labeled secondary antibody (1:1000, anti-mouse Alexa 488) for 60 min at room temperature. For actin filaments detection, phalloidin-conjugated fluorescent 633 (1:50, Thermofisher, A22284) was added to the secondary antibody solution. Nuclei were counterstained using DAPI and coverslips were mounted with Prolong Gold antifade reagent (Invitrogen, P36930). The immunofluorescence images were obtained by confocal microscope Olympus FV3000 and focus on actin filaments at the peripheral area. The photographs were captured by using a 63X/1.4 Plan Apo Oil matched with the large-format 1024 × 1024 pixel.
Staining pattern of stress fibers by phalloidin were classified into the four subgroups based upon the morphologic features : types A (> 90% of cell area filled with thick cables), type B (at least 2 thick cables running under nucleus and rest of cell area filled with fine cables) staining patterns, type C (no thick cables, but some cables present) and type D (no cables visible in the central area of the cell) 22,23 . Images were analyzed by visual inspection or automated classification by us of machine learning.
Approximately 500 cells were counted per transfection construct for in 3-6 independent experiments. www.nature.com/scientificreports/ Immunofluorescence staining for mitochondria. For mitochondrial labeling in live-imaging, cells were seeded in a glass-bottom dish (35 mm) with the cell density of 3 × 10 5 cells for 24 h to reached 80% confluency and the medium was changed before transfection. The cells were transfected with 2.5 μg of GFP-tagged INF2 plasmids onto the cells by using Lipofectamine P3000 and incubated at 37 °C for 8 h. After that, the cells were labeled with MitoTracker Red CMXRos (Thermofisher, M7512) at 100 nM concentration for 15 min at room temperature in phenol-red-free DMEM. Live cell imaging was done by confocal microscopy Zeiss LSM700. To visualize the dynamic of mitochondria, time-lapse images were obtained with every 5 s interval for 10 min.
To classify the mitochondria pattern, cells were fixed with 4% paraformaldehyde. The morphologic patterns were quantified by classifying the appearance of mitochondria into three subclasses (tubular, mixed, and fragmented) in transfected cells according to the protocol previously described 51,52 . To analyze the 2D mitochondria motility, we took a representative stack of the cell (20 cells/each type) and measured mitochondrial length, maximum intensity projections of z-series with 0.5 μm increments for red channel (MitoTracker) 12 . The flat regions of cells with clearly resolved mitochondria were selected, and 10 mitochondria per cell were measured using the line tool in NIH image software. For intensity profiling, the intensity profiles of mitochondrial signals along the lines were plotted by ImageJ. Images were digitally converted to grayscale 8-bit and analyzed by use of the surface plot command. A 3D graph was reconstituted as Y axis representing the intensity (gray value) of the mitochondria, whereas X-axis and Z-axis demonstrated the cell shape and size, respectively.
Transfer learning on deep convolutional neural network (CNN). We initialized our network parameters to default parameter set provided for the deep CNN, ALEXNET, and then fine-tuned the parameters of the last layer of the networks on our data via back propagation by using two NVIDIA GeForce RTX 2080Ti graphic processing units. The loss function was defined as the cross entropy between predicted probability and the true class labels, and we used SGDM optimization with learning rate of 0.0001 and momentum of 0.9 for training the weights.
For the mitochondria morphology classification, we obtained 1,024 × 1,024 pixel of 34 images for INF2 wildtype and p.T161N variant, respectively. After clipping the transfected HeLa cell images according to the GFP signals, we randomly extracted 227 × 227 pixel of 6,800 (100 × 34 × 2) images from the original images as the ALEXNET input and used 7% (400/6,800) of those for training.
For the actin morphology classification, we obtained 56 wild-type and 79 mutant expressing cell images and used 3% (400) of the extracted 227 × 227 pixel of 13,500 (100 × (56 + 79)) images for training. Statistical differences in morphology scores between the wild-type and p.T161N variants were evaluated with Wilcoxon rank-sum test.
Immunofluorescence staining for microtubule and phenotyping classification. To stain the microtubule, cells were firstly labeled with anti-tubulin antibody (1:300, Abcam, ab7291) and further incubated with the fluorescence labeled secondary antibody (1:1000, Abcam, ab150115). By visual inspection of immunofluorescence staining of β-tubulin, subcellular patterns of were classified into 3 subgroups by use of the following criteria: (i) oriented in a radiating array extended from centrosome area near the nuclei (the microtubule organizing center) to the periphery (normal subtype), (ii) intermediate, heterogeneous mixed configuration consists of normal radiating and thick bundles disarranged in parallel (intermediate subtype), and (iii) most aligned along the long axis of the cells (disorganized subtype) 53 . All images were scored by at least three independent scorers.
Statistics. Data were expressed as mean ± SEM. Statistical analysis was performed using 2-tailed Student's t test or one-way ANOVA (SigmaStat, version 3.1.1). Survival analysis was carried out using the log-rank test (Prism 8; GraphPad). The pairwise differences in a proportional distribution of four actin-phenotypic categories between wild-type INF2 and variants were analyzed by use of Fisher exact test on the R program with multiple testing corrections. P < 0.05 was considered significant.
Clinical details of the five index cases with CMT/FSGS. Family ID 7-OK. Patient (II-1) first noticed proteinuria at age 11 on regular check-up and exhibited SRNS at age 11 year 54 . Renal biopsy revealed FSGS with the tubule-interstitial injuries. Despite the immunosuppressive therapy, she eventually progressed into ESRD at age 14. She first noticed difficulty in walking around at age 14. Neurological assessment at age of 15 revealed that she had steppage gait with pes cavus deformity and distal muscle weakness. Muscle atrophy was noted in the upper and lower extremities, and deep tendon reflexes was decreased in the lower extremities. No sensory disturbance was noted.
Patient (II-3), a younger brother of case II-1. No parental consanguinity was noted. He first manifested proteinuria at age 6. The proteinuria gradually increased to nephrotic level at age 11. He underwent renal biopsy www.nature.com/scientificreports/ at age 11 with a diagnosis of FSGS and eventually progressed to ESRD at age 15. Neurological assessment at age 12 demonstrated distal muscle weakness with steppage gait, and muscle atrophy in the upper and lower extremities. Deep tendon reflexes were decreased in the lower extremities. There were no sensory disturbances in peripheral neurons.
Family ID 8 NI. Patient (II-1) first noticed proteinuria at age 11 and developed nephrotic syndrome at age 14 55 . Renal biopsy at age 14 revealed FSGS. The renal disease was resistant to steroid therapy and progressed into ESRD at age 16. She had difficulty in walking since childhood. Pes cavus deformity was corrected by orthopedic surgery at age 11. Neurological examination at age 14 demonstrated the steppage gait with asymmetric atrophy and weakness in the peroneal and tibial muscles. An audiogram revealed a bilateral sensorineural deafness with hearing loss of 50 dB. Various auditory testings suggested that the hearing inability is ascribed to the auditory periphery, cochlea, or auditory nerves.
Family ID 9 OS. Patient (II-1) first manifested proteinuria on regular check-up at age 11. He developed nephrotic proteinuria at age 13 with deteriorating renal functions. Renal biopsy at age 14 revealed FSGS in the absence of immune-depositions. He had been treated with steroids and angiotensin converting enzyme inhibitors and angiotensin receptor blockers in combination. The renal disease was resistant to the steroids with progressive functional deterioration and eventually reached ESRD at age 14. He noticed difficulty in walking at age 15. On neurological examination at age 15, he exhibited muscle weakness and atrophy predominantly in the lower extremities and had steppage gait with drop foot, and pes cavus deformity. INF2 variants was reported in mutational survey of SRNS 56 .
Family ID 10 YA. Patient (II-1) first presented proteinuria at regular check-up at age 14 54 . The first renal biopsy was diagnosed as MCNS. Because of deteriorating renal function and increased proteinuria, he underwent the second renal biopsy at age 16, revealing a FSGS histology. The renal disease was resistant to steroid therapy and progressed into ESRD at age 17. He first noticed difficulty in walking at age 10 and was diagnosed as CMT. On neurological examination at age 17, muscle atrophy was noted in the distal muscles of the forearm and lower limb with absent distal deep tendon reflexes. He had progressive gait disturbance with steppage gait. Neither intellectual nor hearing disability were noted. Biopsy of peroneal nerve revealed degeneration in large myelinated fibers with onion bulb formation, which is typically seen for demyelinating CMT1 subtype.
Ethical approval. The research related to human subject has been complied with all the relevant national regulations, institutional policies, and in accordance with the Helsinki Declaration, and has been approved by the institutional review board in the Kansai Medical University (2008902) and each affiliation.