Torsin ATPase deficiency leads to defects in nuclear pore biogenesis and sequestration of MLF2

Torsin ATPases (Torsins) are widely conserved proteins in metazoans and have essential, yet poorly understood, roles. While Torsins are phylogenetically related to the well-characterized Clp/HSP100 proteins (Rose et al., 2015), they deviate from these ATPases in several fundamental aspects. Torsins are the sole members of the AAA+ ATPase superfamily to reside in both the lumen of the ER and the nuclear envelope (NE; Laudermilch et al., 2016). Another unusual feature is that Torsins are inactive in isolation and require one of two membrane-spanning cofactors, LAP1 or LULL1, for ATPase activity (Zhao et al., 2013). This activation relies on a classical active site complementation mechanism, in which the luminal domain of LAP1 or LULL1 contributes an arginine finger that is notably absent from Torsins (Brown et al., 2014; Sosa et al., 2014). A steadily increasing number of mutations affecting this delicate assembly have been identified as causal factors in human pathologies, as is the case for the highly debilitating movement disorder DYT1 dystonia (Brown et al., 2014; Demircioglu et al., 2016), where TorsinA was originally identified through a positional cloning approach (Ozelius et al., 1997). More recently, a LAP1 mutation was identified that severely limits the lifespan of affected individuals who suffer from diverse symptoms including dystonia and myopathy (Fichtman et al., 2019).

While the diverse set of Torsins exhibits tissue-specific expression (Jungwirth et al., 2010) and differential abilities to be stimulated by their distinctively localizing cofactors (Zhao et al., 2013), the shared hallmark phenotype that is observed upon their genetic manipulation in nematodes (VanGompel et al., 2015), Drosophila melanogaster (Jokhi et al., 2013), mouse models (Goodchild et al., 2005; Liang et al., 2014; Tanabe et al., 2016), and tissue culture cells (Laudermilch et al., 2016; Naismith et al., 2004; Rose et al., 2014) is NE blebbing (Laudermilch and Schlieker, 2016). Major obstacles toward understanding Torsin function in this phenotypic context are the genetic redundancy between Torsin homologues in human tissue culture cells and mouse models (Kim et al., 2010; Laudermilch et al., 2016), and the essential nature of Torsins (Goodchild et al., 2005).

We previously presented a system that resolves both of these limitations by generating a quadruple Torsin deletion HeLa cell line (designated 4TorKO) in which all four Torsin genes (TOR1A, TOR1B, TOR2A, and TOR3A) have been deleted using CRISPR/Cas9 genome engineering. This 4TorKO cell line abundantly exhibits the hallmark cellular phenotype of NE blebbing in which the inner nuclear membrane (INM) bulges into the perinuclear space to form an omega-shaped herniation (Laudermilch et al., 2016). Ubiquitin (Ub) conjugates of the K48 linkage type are enriched in the lumen of the bleb in 4TorKO cells (Laudermilch et al., 2016) and in mouse models of Torsin dysfunction (Pappas et al., 2018). At the base of a bleb, there is electron density with a uniform diameter and dimensions similar to the nuclear pore complex (NPC). This density can be decorated via immunogold labeling using Mab414 antibodies, which recognize several phenylalanine-glycine (FG)–rich NPC components termed FG-nucleoporins (FG-Nups; Laudermilch et al., 2016). NE blebbing was also observed in Saccharomyces cerevisiae upon genetic manipulation of Nups (Onischenko et al., 2017; Wente and Blobel, 1993) and genetic perturbation of Brr6 and Brl1, which were implicated in NPC biogenesis (Zhang et al., 2018).

Whether a causal relationship exists between NPCs and bleb formation in the context of Torsins is largely unknown. However, the finding that nuclear transport is perturbed in Caenorhabditis elegans upon mutation of the TorsinA homologue OOC-5 (VanGompel et al., 2015), as well as the observation of altered in situ distribution of nuclear transport machinery in brain tissue of mouse models of dystonia (Pappas et al., 2018), further support a functional connection between Torsins and the NPC. Clearly, more insight into the molecular composition of these Nup-containing densities and their provenance is required to distinguish whether they are mature NPCs, products of stalled NPC biogenesis, or a result of NPC instability. One hurdle in testing kinetically resolved roles for Torsins in NPC biogenesis or homeostasis is the absence of bleb-specific live-cell imaging markers. A better functional assignment for Torsins is additionally confounded by a lack of quantitative information about NPC number and assembly state in relation to bleb formation.

In this study, we quantify NPCs and observe a considerable reduction of mature NPCs with a concomitant increase of Nup-containing blebs in 4TorKO cells relative to WT cells. These structures form in a strictly cell cycle–dependent fashion. We find that the protein myeloid leukemia factor 2 (MLF2) is highly enriched in the lumen of newly forming blebs, allowing us to develop MLF2 derivatives as bleb-specific probes with broad utility for live-cell imaging and functional characterization. Notably, bleb formation occurs rapidly and synchronously immediately following NE reformation after mitosis, a timing that is reminiscent of interphase NPC biogenesis. This dynamic buildup can be selectively perturbed by depletion of POM121, a Nup that is essential for interphase NPC biogenesis (Doucet et al., 2010). These observations, as well as the diagnostic underrepresentation of the late NPC biogenesis marker Nup358, establish a role for Torsins during interphase NPC biogenesis.

To further explore a functional relationship between Torsins and NPCs, we first asked whether Torsin deletion leads to a reduction in the number of nuclear pores in asynchronously growing cells. To this end, we exploited our previously reported HeLa-based 4TorKO cell line and processed cells for standard TEM along with isogenic WT cells. As expected, WT cells featured an evenly spaced INM and outer nuclear membrane (ONM), with an average of 15.2 nuclear pores per 30 µm of NE (Fig. 1 A, lower panel; and Fig. 1 B). As reported previously (Laudermilch et al., 2016), the blebbing phenotype was highly penetrant in 4TorKO cells (Fig. 1 A, upper panel). The number of mature nuclear pores was significantly decreased relative to WT cells with an average of 11.2 nuclear pores per 30 µm of NE (Fig. 1 B). This was accompanied by a corresponding increase of fuzzy electron density at the base of many blebs (Fig. 1 A) that we have previously shown to represent FG-Nup containing assemblies (Laudermilch et al., 2016). Thus, the observed reduction of the number of nuclear pores with a concomitant increase of FG-Nup assemblies at the bases of blebs represents a highly robust phenotype.

A second important criterion supporting a connection to NPC assembly would be a possible cell cycle dependency of bleb formation. This is based on prior work that cumulatively suggests that there are likely two biochemically (Doucet et al., 2010) and morphologically distinct mechanisms of NPC assembly, one occurring during postmitotic NE reformation and the other during interphase (Otsuka and Ellenberg, 2018). In the latter case, there is an emerging consensus that NPC assembly begins from the inside of the nucleus on the INM (Doucet et al., 2010) likely through an inside-out evagination of the INM, which ultimately leads to fusion with the ONM (Otsuka et al., 2016). Interphase assembly might be under the control of cell cycle (Talamas and Hetzer, 2011) and other (McCloskey et al., 2018) regulators with a potential burst of this assembly mechanism in early G1 (Doucet et al., 2010; Otsuka et al., 2016; Weberruss and Antonin, 2016). If the Torsin knockout phenotype does arise from a defect in NPC biogenesis, we would expect to observe the first signs of bleb formation early in G1. To investigate this possibility, we synchronized 4TorKO cells in early S phase using the double thymidine block method (Bostock et al., 1971) and processed the cells for immunofluorescence at various times after release from the thymidine block (Fig. 1, C and D). Anti–K48-Ub antibodies were used to score for bleb formation since K48-linked Ub is strongly enriched in the bleb lumen both in 4TorKO cells (Laudermilch et al., 2016) and in mouse models of DYT1 dystonia (Pappas et al., 2018). While the majority of asynchronously growing 4TorKO cells exhibited K48-Ub foci diagnostic of NE blebs (Fig. 1 E), a striking cell cycle dependency was observed in synchronized cells. At the G1/S boundary (T = 0 h upon release of block), nearly all cells contain K48-Ub foci (Fig. 1, D and E). Additionally, we observed essentially the same abundance for subsequent time points during S phase and early G2. Following mitosis, however, this number drops substantially (Fig. 1, D and E). It should be noted that in our experience, the entire HeLa cell population is not sharply synchronized under conditions of the double thymidine synchronization. Some deviation from complete synchrony certainly exists (we estimate that ∼85% of the population is well synchronized). However, midbodies (that are also labeled with anti–K48-Ub; see middle panel in Fig. 1 D) can be used as convenient diagnostic markers to assign cells to the time of late cytokinesis. Notably, K48-Ub NE foci are completely absent from the nascent NE of these dividing or recently divided cells. As cells proceed through G1, the number of cells with K48-Ub foci steadily increases again and reaches a maximum at ∼20 h after release from the thymidine block (Fig. 1 E), which is about one complete cell cycle. To directly confirm that the reduction of K48-Ub NE foci coincides with a loss of blebs, we also performed an analogous double thymidine block experiment and processed synchronized 4TorKO cells for EM at T = 2 h and T = 12 h after release. These correspond to the time points of highest and lowest abundance of K48-Ub foci, respectively. We observed that the number of blebs per cross section for each time point is in good agreement with the numbers derived from immunofluorescence (Fig. 1, compare E with F). We therefore conclude that the formation of blebs is a cell cycle–dependent process and that the majority of blebs are formed during G1.

While the observed timing of bleb formation is consistent with a possible role for Torsins in NPC biogenesis (D’Angelo et al., 2006; Dultz and Ellenberg, 2010; Otsuka et al., 2016), a live-cell imaging readout for bleb formation would enable higher temporal resolution. Moreover, live-cell observations on a single-cell level allow for a direct visualization of mitotic events and should resolve the issue of incomplete synchronization that limited the accuracy of our former measurements (Fig. 1).

We therefore set out to identify a suitable marker that is specific to NE blebs. To this end, we used a comparative proteomic approach in which we compared the respective NE proteomes of 4TorKO versus WT cells. Briefly, we gently homogenized 4TorKO and WT cells and isolated NEs by a series of consecutive centrifugation steps (see Materials and methods; Fig. 2 A). Since at least a subset of bleb luminal components appear to be conjugated to K48-Ub (Laudermilch et al., 2016), we solubilized NEs with mild detergent and immunoprecipitated the obtained extracts with K48-Ub–specific antibodies. Note that both the bleb structure and its reactivity to K48-Ub antibodies are preserved under these gentle NE isolation procedures (Fig. 2 A, inset) before addition of detergent. The resulting immunoprecipitates (IPs) were then resolved by SDS-PAGE and subsequently analyzed by mass spectrometry (MS).

We set strict selection criteria of (i) candidate proteins having a sequence coverage of >20% and (ii) at least a fourfold enrichment of unique peptide counts in the 4TorKO cells over WT cells. Three proteins, MLF2, emerin, and DNAJB2, satisfied those criteria (see Table 1) and were further analyzed. To this end, we engineered the corresponding cDNAs specifying epitope-tagged variants, which were transiently transfected into 4TorKO cells, followed by immunofluorescence. Notably, the MLF2 signal was primarily confined to the blebs, whereas emerin and DNAJB2 signals appeared to be enriched in blebs but also more diffusely distributed in other subcellular compartments (Fig. 2 B). We therefore focused our subsequent analysis on MLF2. Furthermore, MLF2 is a small soluble protein, making the addition of fluorescent tags less likely to impact its nuclear localization due to size constraints, as would likely be the case for larger and membrane-bound proteins like DNAJB2 and emerin, respectively.

To independently confirm biochemically that MLF2 is enriched in the NEs of 4TorKO cells, we employed CRISPR/Cas9 genome engineering and installed a C-terminal tandem HA tag on MLF2 at the endogenous locus in both WT and 4TorKO genetic backgrounds. Clonal cell lines were isolated from WT and 4TorKO backgrounds with equivalent MLF2-HA expression levels (compare whole cell lysates; Fig. 2 C). Using these cell lines, we conducted subcellular fractionations to analyze the relative amount of MLF2 in nuclear, NE, and nucleoplasmic fractions. After fractionation, the corresponding samples were solubilized in SDS and subjected to SDS-PAGE and immunoblotting. While MLF2-HA levels were approximately the same in whole cell lysates, we observed slightly more MLF2 in the nuclear fraction of 4TorKO cells (Fig. 2 C). A further separation of nuclei into NEs and nucleoplasm revealed a major enrichment of MLF2 in the NEs of 4TorKO cells relative to WT cells, while the levels were comparable in the nucleoplasmic fraction (Fig. 2 C). The INM protein Sun2 was additionally monitored via immunoblotting to confirm successful fractionation (Tsai et al., 2019). Finally, we subjected detergent extracts of both NE fractions to immunoprecipitation with anti–K48-Ub antibodies followed by SDS-PAGE and immunoblotting. In line with our original MS-based experiment, we detected MLF2 in the IPs from 4TorKO NEs but not from WT NEs (Fig. 2 D). These results led us to explore the potential use of MLF2 as a live-cell imaging tool to investigate the potential role of Torsins in NPC biogenesis.

To begin, we engineered an MLF2-GFP fusion protein, and asked whether MLF2 localizes to blebs. 4TorKO cells were transfected with MLF2-GFP, and we then monitored the localization of the fusion protein relative to blebs (direct fluorescence versus anti–K48-Ub signal) by immunofluorescence. We observed a punctate pattern for MLF2-GFP at the nuclear periphery combined with a diffuse nucleoplasmic staining (Fig. S1 A). These observations are in excellent agreement with our biochemical fractionation (Fig. 2 C). More importantly, we observed a correlation in 4TorKO cells displaying both K48-Ub and MLF2 foci (Fig. S2, A–C) and a considerable, though not complete, degree of colocalization between K48-Ub staining and MLF2-GFP, which is also apparent in a line-scan analysis (Fig. S1 B).

The question arises whether the observed MLF2-GFP localization is indeed a specific indicator of NE bleb formation or if the perinuclear foci formation merely results from perturbed nuclear export. We therefore asked if GFP variants with a nuclear export signal (NES-GFP) or nuclear import signal (NLS-GFP) give rise to a similar or distinct localization pattern. Both NES-GFP and NLS-GFP variants showed the expected extranuclear and nucleoplasmic localization, respectively (Fig. S1, C and D, upper panels; and Fig. S1, E and F). To validate the functionality of NES-GFP and explore the nucleo-cytoplasmic transport competency of 4TorKO cells, we treated cells with Leptomycin B, a CRM1-dependent nuclear export inhibitor (Kudo et al., 1998). As expected, we observed a strong nucleoplasmic GFP signal upon inhibition of nucleo-cytoplasmic trafficking (Fig. S1, C and D, lower panels), excluding the formal possibility that NES-GFP never enters the nucleus. Additionally, we did not observe an enrichment of HA-tagged Ran (Fig. S1, G and H), a major player in nuclear transport (Adam et al., 1992; Lui and Huang, 2009). Thus, MLF2 is a highly specific marker for NE aberrations in 4TorKO cells.

We next asked whether MLF2-GFP distinctively localizes to NE blebs, a question that is best addressed via EM, which also provides information about membrane topology. We processed MLF2-GFP–expressing 4TorKO cells for EM via high pressure freezing and subjected sections to immunogold labeling with anti-GFP antibodies. 4TorKO cells displayed the typical accumulation of NE blebs (Laudermilch et al., 2016). We observed a striking enrichment of immunogold labeling of MLF2-GFP in these blebs and can clearly assign this accumulation to the bleb lumen that is enclosed by the INM (Fig. 3 A). Furthermore, FG-Nups localize at the neck region of blebs containing either endogenous MLF2 or MLF2-GFP, as judged by double immunogold labeling (Fig. S2 D, top and bottom panels, respectively). At a low frequency, we additionally observe cases in which gold particles concentrate in direct juxtaposition of a deformed INM (Fig. 3, B and C). These could represent early bleb intermediates in which evaginations of the INM begin to form. The early addition of MLF2 to bleb intermediates would imply that MLF2 might be added before K48-Ub conjugation occurs. Supporting this idea, we observe more 4TorKO cells with MLF2 foci earlier than K48-Ub in G1 phase (Fig. 3 D). We therefore conclude that MLF2 has the potential to be an effective tool in elucidating the dynamics of bleb formation.

Resolving the dynamics of bleb formation relative to NE reformation during mitosis requires a noninvasive, robust live-cell imaging platform. To this end, we generated a 4TorKO cell line stably expressing MLF2-GFP and mScarlet-Sec61β through retroviral transduction. To mitigate the potential occurrence of any artificial morphological effects on cells resulting from the constitutive activation of either gene, we employed a doxycycline (Dox)-inducible promoter system to control the expression levels of both genes. Furthermore, we used lattice light-sheet microscopy (LLSM) as it provides rapid 3D image acquisition with reduced phototoxicity and photobleaching, thus allowing the acquisition of longer time series data. Since we crudely assigned bleb formation to the early G1 phase (Fig. 1 E), we identified prometaphase/metaphase mitotic cells based on their round appearance and visible metaphase plates (Fig. 4 A) and followed MLF2-GFP and mScarlet-Sec61β into early G1 (Video 1). For standardization, the onset of anaphase was arbitrarily defined as T = 0 s. In agreement with previous observations, NE reformation after open mitosis occurred between 400 and 500 s after anaphase onset (Fig. 4 A; Dechat et al., 2004). At ∼700 s, the first small MLF2-GFP foci appear in the nuclear periphery, while larger ones are observed at later times (Fig. 4 A). Additionally, a small number of foci appear to form at some distance from the nuclear periphery in the nucleoplasm (Fig. 4 A and Video 1). Since we observed this trend repeatedly, we scrutinized this process further in an independent experiment to obtain a deconvoluted image series. We observed that the vast majority of these seemingly nucleoplasmic “outliers” are in fact closely associated with evaginations or wrinkles of the NE as judged by their colocalization with mScarlet-Sec61β (Fig. 4 B and Video 2).

Since LLSM is superior to conventional fluorescence microscopy in terms of phototoxicity and photobleaching, we were able to closely resolve the growth of individual foci over time. Focusing on the formation and maturation of a subset of MLF2-GFP foci in an individual daughter cell, we observed an initial steep growth phase that reached a maximum fluorescence intensity around 1,200 s after anaphase onset (Fig. 4, C–E). After this rapid growth phase, the foci appear to be static (Fig. 4 D and Video 3). Thus, the formation of the blebs is far more rapid than previously inferred from using K48-Ub as a readout in fixed cells (Fig. 1 E). Another unexpected observation is the synchrony with which the bleb formation occurs (Fig. 4, A, D, and E; and Video 1). Based on our time-resolved recordings, we estimate that the vast majority of initiation events can be narrowed down to an ∼100-s time window starting at ∼700 s after anaphase onset, which is in agreement with the appearance of NPC intermediates and subsequent formation of nascent NPCs from previous reports using diverse microscopic methodologies (D’Angelo et al., 2006; Otsuka et al., 2016, 2018).

Ub was the first characterized marker to label NE blebs in a TorsinA-deficient mouse model (Liang et al., 2014). However, it has been unclear if a functional relationship between ubiquitylation and bleb formation exists. To test for a possible requirement, we asked whether we could engineer MLF2 to recruit Ub-modifying enzymatic activities to the bleb lumen. To this end, we engineered a construct consisting of an N-terminal MLF2 moiety fused to a deubiquitinating enzyme (DUB) domain derived from M48, the largest tegument protein of murine cytomegalovirus (Schlieker et al., 2005), followed by a C-terminal Flag tag to create MLF2-M48^WT (Fig. 5, A and B). This DUB domain potently deconjugates K48-linked Ub chains (Schlieker et al., 2007), the linkage type that is present in blebs (Laudermilch et al., 2016; Pappas et al., 2018; Fig. 1 D). As a control, we engineered a catalytically inactive variant in which the active site cysteine is mutated to an alanine, MLF2-M48^C23A.

4TorKO cells were transfected with either variant and processed for immunofluorescence with anti-Flag and anti–K48-Ub antibodies 24 h after transfection. With a transfection efficiency of ∼50%, nontransfected cells serve as a convenient control. While nontransfected cells display the expected K48-Ub foci phenotype, K48-Ub foci are virtually absent from cells expressing MLF2-M48^WT (Fig. 5 C). However, MLF2-M48^WT is still found in perinuclear foci in a manner identical to canonical MLF2 staining (compare Fig. 2 B, Fig. S1 A, and Fig. S2 A), indicating that blebs form independently of K48-Ub enrichment. In the case of catalytically inactive MLF2-M48^C23A, the signals of K48-Ub and MLF2-M48^C23A show the expected degree of colocalization, indicating that it is indeed the DUB activity that is responsible for the lack of K48-Ub signal in MLF2-M48^WT–transfected cells.

These data argue against a critical role for K48-Ub conjugation in bleb formation, while establishing MLF2 as a useful tool to recruit specific enzymatic activities to NE blebs.

Nup358 is a cytosolic-facing Nup that is recruited to a nascent NPC after the assembly of the bulk of the FG-Nups and the fusion of the INM and ONM (Otsuka et al., 2016). Therefore, as we previously proposed, the absence of Nup358 from FG-Nup containing blebs may provide a useful tool to assess whether these blebs are formed at sites of stalled NPC biogenesis (Chase et al., 2017). We therefore imaged WT and 4TorKO cells via 3D structured illumination microscopy (SIM) and compared the localization of Nup358 and other FG-Nups using anti-Nup358 antibodies and the pan anti-FXFG antibody, Mab414, respectively. While we recognize that Mab414 is capable, in principle, of labeling Nup358 (Wu et al., 1995), it was the only antibody tested that provided the necessary specificity and signal-to-noise ratio to confidently assign NPCs using SIM. Moreover, it is established that Mab414 favors labeling Nup62 (Davis and Blobel, 1986) because it has more FXFG repeats and is found at higher copy numbers in the NPC (when fully formed) compared with Nup358 (Ori et al., 2013). Thus, the contribution of any Mab414-specific Nup358 labeling would likely be negligible.

Consistent with the idea that we can detect fully formed NPCs by SIM, we observe a near complete colocalization of the Mab414 and Nup358 signals in WT cells in focal planes that illuminate the nuclear surface (Fig. 6, A and B). Moreover, in mid-planes where NPCs are viewed by cross section, it is apparent that the Nup358 signal is spatially separated from the Mab414 signal, with the latter being more proximal and Nup358 being more distal relative to the nuclear interior (Fig. 6 B, lower panels). This is in agreement with our current understanding of NPC structure (Lin and Hoelz, 2019; Rout et al., 2000; Schwartz, 2016; von Appen et al., 2015). In this view, it is also clear that Mab414 does not detectably label Nup358, confirming our prior assumption with respect to the specificity of Mab414. In contrast, in 4TorKO cells, we observed irregularly shaped focal areas with a diameter of up to 5 µm in which we saw robust staining with Mab414 at a density that is comparable to WT cells, but with a notable absence of the anti-Nup358 label (Fig. 6 B). This suggests that these areas may represent the accumulation of aberrant intermediates during NPC assembly, which is further supported by the observed anti-correlation between Nup358 and MLF2 foci (Fig. 6 C).

Finally, we quantified the total number of Mab414 foci based on seven nuclei each (from three independent experiments) from WT and 4TorKO cells, which we interpret as the sum of mature and immature NPCs. We observe a modest, albeit insignificant, reduction in the density of Mab414 foci in 4TorKO cells, suggesting that there is not a major reduction in NPCs or NPC biogenesis sites in the absence of Torsins (Fig. 6 D). However, when we compared the fraction of colocalizing Nup358 and Mab414 foci as a measure for mature NPCs with the number of Mab414 sites arbitrarily expressed as 100%, we observed an ∼40% reduction in mature NPCs in 4TorKO cells relative to WT cells (Fig. 6 E). This result is in agreement with the observed reduction of NPCs and the concomitant increase of bleb-localized, FG-Nup–containing densities in 4TorKO cells in electron micrographs (compare Fig. 6 E with Fig. 1 B).

In conclusion, the observed underrepresentation of Nup358 from sites containing FG-Nups is consistent with the interpretation that a large proportion of FG-Nup–containing Nup assemblies is devoid of cytoplasmic fibrils, likely because of defective NPC biogenesis at a step before INM/ONM fusion.

Having shown that NE blebs in 4TorKO cells feature NPC-like structures at their bases, a key question that remains is whether a causal relationship of NPC components for NE blebbing exists. An essential requirement of an NPC component for bleb biogenesis would support the idea that NE blebs occur as a consequence of defects in NPC formation and might therefore represent aberrant, stalled intermediates (Chase et al., 2017; Laudermilch and Schlieker, 2016). Previous work has established that the mitotic insertion of NPCs during open mitosis requires the Nup ELYS, while insertion of NPCs after reformation of the NE is independent of ELYS but highly sensitive to the depletion of the transmembrane Nup POM121 (Doucet et al., 2010; Franz et al., 2007). Having shown that NE blebs form after NE reformation (Fig. 4 A), and given that these are topologically identical and morphologically similar to NPC biogenesis intermediates (Laudermilch et al., 2016; Otsuka et al., 2016), we reasoned that this selective dependency could be exploited by directly testing whether ELYS and POM121 are implicated in bleb formation.

As a first step, we depleted either ELYS or POM121 in 4TorKO cells via siRNA-mediated silencing and scored cells for any effects on the formation of MLF2-GFP and K48-Ub foci. While both siRNA treatments potently reduced the RNA levels of their targets, only POM121 led to a stark reduction of both MLF2-GFP and K48-Ub focus formation, while silencing of ELYS had no significant effect (Fig. 7, A–C; and Video 4). Since K48-Ub foci form relatively late during bleb formation (Fig. 3 D), we additionally monitored bleb formation via EM to directly visualize membrane deformation during bleb biogenesis under knockdown conditions (Fig. 7, D and E). These results mirrored our observations using MLF2-GFP and K48-Ub as readouts. Bleb formation was essentially unperturbed in ELYS-silenced 4TorKO cells, whereas depletion of POM121 resulted in a stark, statistically significant decrease in the number of blebs per NE length with a concomitant formation of K48/MLF2-positive foci in the cytosol (Fig. 7, A and D). Together, these data establish an epistatic relationship between Torsins and the NPC component POM121. Considering that our knockdown approach did not completely eliminate POM121 on the transcript level (Fig. 7 B) but nevertheless leads to a material reduction in bleb formation (Fig. 7, C–E), it seems reasonable to deduce that POM121 is strictly required for bleb formation.

NE blebbing has been observed in developmentally regulated processes and upon genetic perturbation of Nups in numerous model organisms (Thaller and Lusk, 2018). Genetic ablation or mutation of specific Nups falls into the latter category, with NUP116 deletions as examples of NE blebs with morphological similarities relative to the ones seen in 4TorKO cells (Onischenko et al., 2017; Wente and Blobel, 1993). The identification of a subset of FG-Nups at the electron-dense base of NE blebs in Torsin-deficient cells, as well as the finding that the diameter of this density is similar to mature NPCs (Laudermilch et al., 2016), previously suggested that Torsins could be implicated in NPC biogenesis (Chase et al., 2017; Laudermilch and Schlieker, 2016). The recent discovery of an inside-out evagination in the context of interphase NPC biogenesis (Otsuka et al., 2016) additionally revealed a phenomenon similar to the effects seen upon Torsin manipulation (Fig. 1 A; Chase et al., 2017; Otsuka et al., 2018; Weberruss and Antonin, 2016). Thus, several similarities exist between the two phenomena that relate Torsins to NPC biogenesis. In this study, we asked whether a causal relationship can be established between Torsins and NPC biogenesis. We observed that the number of mature NPCs is strongly reduced in 4TorKO cells, with 23% of NPC-like structures being located at the base of NE blebs (Fig. 1, A and B). Using K48-Ub as a marker for blebs in the context of fixed cells, the timing of bleb formation falls mostly within the early G1 phase of the cell cycle (Fig. 1, C–F), a window during which a burst of interphase NPC biogenesis has been observed (Dultz and Ellenberg, 2010). The observed penetrance of this Torsin knockout phenotype (Fig. 1, A and B) is remarkable if one considers the estimate that ∼50% of all NPCs are installed through interphase insertion (Doucet et al., 2010).

Based on our identification of MLF2 as a bleb-specific marker (Fig. 2), we developed a live-cell imaging platform to show that bleb formation occurs synchronously within a narrow window of time immediately after NE reformation following open mitosis (Fig. 4 A and Video 1). Both the speed and synchrony of blebbing were entirely unexpected since we assumed a much broader, “stochastic” emergence of blebs based on experiments with fixed cells (Fig. 1, D–F). Furthermore, it is noteworthy that a high degree of specificity exists for the luminal content. Model substrates of nuclear transport and ribosomes (some of the major nuclear export cargo) do not accumulate in these blebs (Fig. S1; Laudermilch et al., 2016). This argues against the formal possibility that blebs merely occur upon the packaging of “random” nuclear export cargo. The question arises, however, as to whether K48-Ub or MLF2 plays a role in NPC biogenesis or whether they are merely sequestered in blebs. We did not observe a major role for K48-Ub conjugation (Fig. 5 C), and our preliminary MLF2 silencing approach did not suggest a critical role for MLF2 in bleb formation (Fig. S3, A and B). Whether this is due to a possible genetic redundancy with the MLF2 homologue MLF1 remains to be seen. An alternative possibility is that the sequestration of MLF2 into blebs detrimentally affects the normal cellular activity of this protein, or reflects a presently unknown function of this poorly characterized protein (Banerjee et al., 2017; Kuefer et al., 1996). Interestingly, MLF2 features unusually methionine-rich and arginine-rich regions, as well as proposed unstructured regions (Fig. S3 C). Given that methionine side chains frequently engage in interactions with aromatic side chains (Valley et al., 2012) that are abundantly present in FG-Nups, it is tempting to speculate that MLF2 could serve as an FG-Nup–directed chaperone. Alternatively, the arginine-rich regions of MLF2 might drive its incorporation into FG-Nup–rich regions in the bleb, as arginine side chains can engage in cation-π interactions with aromatic side chains in FG-Nups (Frey et al., 2018). Future work will be required to test and/or discern between these possibilities.

Most importantly, our study firmly links Torsins to the process of interphase nuclear pore biogenesis. Apart from the aforementioned kinetics of bleb formation, this functional assignment is supported by the following observations: (i) a reduction in the number of mature pores (Fig. 1 B), (ii) an underrepresentation of the late NPC assembly marker Nup358 from NEs of 4TorKO cells (Fig. 6), and (iii) the strict requirement of POM121—a transmembrane Nup essential for interphase assembly—for bleb formation (Fig. 7). In our model, NE blebbing during interphase NPC biogenesis is likely a consequence of defects in the proper assembly of NPC intermediates or of an inability of INM/ONM fusion (Fig. 8 B). In this context, it might be useful to directly compare EM tomograms representing NE blebs in 4TorKO cells with those observed during interphase NPC biogenesis in unperturbed cells. The latter are somewhat flatter and dome shaped (Otsuka et al., 2016), while larger membrane herniations of ∼200–250 nm are seen in 4TorKO cells (Laudermilch et al., 2016; for a diagrammatic comparison, see Fig. 8, A and B).

How can we interpret this difference? We propose that in Torsin-deficient cells, the otherwise dynamically linked processes of INM deformation and INM/ONM fusion during NPC biogenesis are stalled before or at the step of INM/ONM fusion. As a first possibility, a specific NPC component might require Torsin for folding or trafficking, and the observed effects are indirect and result from a stalling since a specific required component is absent or misfolded. Based on our observation that POM121 is required for bleb formation (Fig. 7), we deduce that this component would likely have to be recruited downstream of POM121 in the assembly pathway. This interpretation would be consistent with previous observations of Ub conjugation in blebs. If we assume that one or several critical components important for NPC formation are misfolded due to the absence of Torsins and consequently ubiquitylated, its mere deubiquitylation would not be expected to restore protein function. As speculated before (Chase et al., 2017; De Magistris and Antonin, 2018; King and Lusk, 2016; Otsuka and Ellenberg, 2018), a second possibility is that Torsins might act on fusogenic components responsible for INM/ONM fusion. Since the fusogenic machinery remains to be identified, it is currently impossible to test this directly. Finally, Torsins have recently been linked to lipid metabolism (Grillet et al., 2016; Shin et al., 2019). We did not observe major changes in the lipid profile in 4TorKO cells versus WT cells (Laudermilch et al., 2016), but we cannot exclude that local, NE-specific changes in lipid composition exist, which might affect the fusogenic properties of NE membranes. It might be interesting to employ lipid-specific probes to scrutinize lipid composition of NEs and its possible perturbation in Torsin-deficient cells in the future.

It remains to be determined why interphase NPC assembly in higher eukaryotes has evolved a need for Torsins while lower eukaryotes and protozoans lack a recognizable Torsin orthologue. However, it is not without precedence that alternative mechanisms for performing analogous cellular processes evolved, particularly in phylogenetically distant organisms (note that budding yeast and humans diverged over approximately one billion years of evolution). For example, vesicle fission during endocytosis relies on a dynamin GTPases in higher eukaryotes, while lower eukaryotes use an actin dependent mechanism (Ferguson and De Camilli, 2012; Lacy et al., 2018). Moreover, Brr6 and Brl1 are required for NPC assembly in yeast and cause NE blebbing upon genetic perturbation, whereas higher eukaryotes lack clearly identifiable orthologues for these proteins (Zhang et al., 2018).

Regarding the broader physiological implications of NE blebbing, it would be interesting to compare naturally occurring instances where blebbing phenotypes closely mirroring our observations have been documented (Thaller and Lusk, 2018). These include NE blebs in the neuromuscular junction in D. melanogaster (Jokhi et al., 2013; Speese et al., 2012), as well as NE blebbing that appears to be an evolutionary conserved process in zygotes and early embryos (Szöllösi and Szöllösi, 1988). In the latter case, blebs with necks of dimensions similar to NPCs were observed, although they have not been linked to NPC biogenesis (Szöllösi and Szöllösi, 1988). It will be interesting to test whether these can be decorated with Mab414 antibodies, and if these structures contain MLF2. It is tempting to speculate that several of these observations can in fact be connected to NPC biogenesis.

Regardless of these questions, we interpret our results to firmly link Torsin ATPases to the process of interphase NPC biogenesis. Our findings have distinct implications for our understanding of movement disorders caused by Torsin dysfunction in neurons. Neuronal cells display a low mitotic index and are thus expected to be particularly vulnerable since these are more dependent on the interphase assembly pathway than dividing cells, which can use the alternate postmitotic insertion pathway. Additionally, neurons are heavily reliant on TorsinA due to a window in neurogenesis during which TorsinA is the dominantly expressed Torsin relative to its homologues (Kim et al., 2010; Tanabe et al., 2016). Thus, our results, along with these data, suggest that defects in NPC biogenesis add to DYT1 dystonia’s disease etiology.

Torsin-deficient HeLa cells and their parental WT cell line were cultured as previously described (Laudermilch et al., 2016). Briefly, cells were cultured in Dulbecco’s modified Eagle’s medium supplemented with 10% FBS (Thermo Fisher Scientific) and 100 U ml⁻¹ of penicillin-streptomycin (Thermo Fisher Scientific). Cells were routinely checked for mycoplasma and determined to be free of contamination through the absence of extranuclear Hoechst 33342 (Life Technologies) staining.

The sequences encoding MLF2 and DNAJB2 were amplified by standard PCR procedures from cDNA (Dharmacon; MLF2 GenBank accession no. BC000898 and DNAJB2 GenBank accession no. BC011609) and subcloned into either a pcDNA3.1⁺ vector (MLF2-HA, MLF2-Flag, and DNAJB2-HA) or a pEGFP-N1 vector (MLF2-GFP). Emerin and VCP (Valosin-containing protein) were amplified by standard PCR procedures using a cDNA library prepared from total HeLa polyA⁺ mRNA. MLF2-M48 fusion variants were constructed through standard Gibson assembly procedures. Plasmids containing the WT and catalytic mutant variant of M48 were a gift from Hidde L. Ploegh (Whitehead Institute for Biomedical Research, Cambridge, MA; Schlieker et al., 2007). The plasmid containing Sec61β was gifted from Shirin Bahmanyar (Yale University, New Haven, CT). Sec61β was subcloned into a modified pEGFP-C1 vector in which EGFP was replaced with mScarlet (pmScarlet-C1). Both MLF2-GFP and mScarlet-Sec61β were subcloned into the pRetroX-Tight-Pur-GOI vector (Takara Bio). NES-GFP was custom synthesized as a 789-bp gBlock gene fragment (Integrated DNA Technologies) with an N-terminal nuclear export sequence (Cardarelli et al., 2012). NLS-GFP was gifted from Anton Bennett (Yale University School of Medicine). WT Ran was derived from pmCherry-C1-RanQ69L (Addgene, 30309) by reverting the point mutation through site directed mutagenesis. WT Ran was subcloned with an N-terminal HA tag into pcDNA3.1⁺.

To generate a 4TorKO cell line stably expressing MLF2-GFP and mScarlet-Sec61β, we employed the Retro-X Tet-On advanced inducible expression system (Takara Bio) following the manufacturer’s protocol. For the production of retrovirus, low-passage 293T cells were transfected with 2 µg MMLV gag/pol, 1 µg viral envelope protein VSV-G, and 6 µg of pRetroX-Tight-Pur-MLF2-GFP, pRetroX-Tight-Pur-mScarlet-Sec61β, or pRetroX-Tet-On using X-tremeGENE 9 (Roche).

Supernatants containing retroviruses were collected 72 h after transfection, filtered via a 0.45-µm filter unit, and stored at −80°C. 4TorKO cells were seeded in six-well plates 24 h before transduction. The next day, medium was replaced with complete growth media supplemented with 4 µg ml⁻¹ polybrene (Sigma-Aldrich), and 100 µl of the respective retroviruses were added dropwise to the wells. Medium was replaced 24 h after transfection to fresh complete media containing 1 µg ml⁻¹ puromycin (Sigma-Aldrich) and 800 µg ml⁻¹ Geneticin (Thermo Fisher Scientific). Antibiotic selection was performed for 7 d. Cells positive for both GFP and mScarlet signal under the Dox-inducible promoter were isolated through FACS. FACS was performed at the Yale University Flow Cytometry Facility using a FACS Aria III sorter (BD Biosciences).

To establish WT and 4TorKO cells stably expressing MLF2 with an endogenous C-terminal 3xHA tag, we used a CRISPR/Cas12a system for PCR tagging genes (Fueller et al., 2018,Preprint). Oligo sequences targeting MLF2 at its endogenous locus were generated from an open access tool (www.pcr-tagging.com; Fueller et al., 2020). Sequences for the PCR tagging oligo primers are as follows: M1, 5′-GCT​GGG​GGA​CGA​AGG​GCG​GAG​GGG​CCT​CCC​CGC​CTG​GCC​ATC​CAG​GGA​CCT​GAG​GAC​TCC​CCT​TCC​CGA​CAG​TCC​CGC​CGC​TAT​GAC​TGG​TCA​GGT​GGA​GGA​GGT​AGT​G-3′; and M2, 5′-CAC​CCC​ACC​CTC​CTT​ACT​CCT​GAT​ACT​TAC​AAG​AGA​GGC​TGA​GGG​CCC​GGG​GCC​CAA​AAA​AGG​CCC​GGG​GCC​CTC​ACC​AGT​ATC​TAC​AAG​AGT​AGA​AAT​TAG​CTA​GCT​GCA​TCG​GTA​CC-3′ (Integrated DNA Technologies). pMaCTag-P28 (Addgene, 120039) was used as PCR template. Cells were transfected with 1 μg total DNA (0.5 μg PCR product and 0.5 μg AsCpf1_TATV Cas12; Addgene, 89354) using Xtreme-GENE 9 following the manufacturer’s instructions. Medium was replaced 24 h after transfection to fresh complete media containing 1 µg/ml puromycin (Sigma-Aldrich). Antibiotic selection was performed for 7 d. Following selection, cells derived from individual colonies were screened for MLF2-3xHA fusion protein by immunoblot, and colonies with HA signal were propagated and saved.

Cells were synchronized with a double thymidine block (Bostock et al., 1971). Cells were incubated in complete growth media supplemented with 2.5 mM thymidine (Sigma-Aldrich) for 18 h. Cells were released from the thymidine block by washing with Dulbecco’s PBS (Thermo Fisher Scientific), and replacing the media with fresh complete growth media. Cells were incubated for 9 h, at which point a second round of 2.5 mM thymidine treatment was administered. Following a 16-h incubation, cells were again washed and incubated in complete growth media without thymidine. This final media replacement was designated as T = 0, and time points were then collected afterwards as indicated in the text.

siRNA knockdown was done using Lipofectamine RNAimax (Life Technologies). The Lipofectamine reagent was diluted in Opti-MEM reduced serum medium (Thermo Fisher Scientific) for 15 min, followed by the addition of the appropriate concentration of siRNA, and incubated further for 15 min. This solution was then added dropwise to cells and allowed to incubate overnight. Media were then replaced with fresh antibiotic-free media. Cells were harvested for qPCR or immunofluorescence 48 h (nontargeting control [siNT], siELYS, and siMLF2) or 72 h (siPom121) after transfection. Forward siRNA sequences 5′-CAGUGGCAGUGGACAAUUCA[dT][dT]-3′ (Sigma-Aldrich) and 5′-UCGUGGAAAGUUUGCUGCAGGGAAA[dT][dT]-3′ (Sigma-Aldrich; Doucet et al., 2010) were used for POM121 and ELYS, respectively, while SMARTpool siRNA was used to target MLF2 (Dharmacon). Following treatment, cells were fixed for immunofluorescence analysis or total RNA was extracted for qPCR following previously described methods (Tsai et al., 2016). In short, 100 ng of RNA was transcribed into cDNA using SuperScript II reverse transcription (Thermo Fisher Scientific) with random hexamer primers (Invitrogen). qPCR was performed using iQ SYBR Green mix and executed on a CFX Real-Time PCR 639 Detection System (Bio-Rad). The ΔΔCt values for each sample were calculated from the subtraction of an internal control value (GAPDH), and results were normalized to the siNT control. Primer sequences (5′-3′) used for qPCR were as follows: GAPDH (forward, CGA​CCG​GAG​TCA​ACG​GAT​TTG​GTC​G; reverse, GGC​AAC​AAT​ATC​CAC​TTT​ACC​AGA), ELYS (forward, CCA​ATT​TCT​GAC​AGC​CCT​CCT​GA; reverse, AGA​TTC​CTA​GCC​TCT​TCT​CCT​GAA), POM121 (forward, CCT​TCA​GCC​AGT​CCC​TGC​AC; reverse, GAG​GGT​GCT​GCC​AAA​ACC​AC), and MLF2 (forward, GGA​CTC​CCC​TTC​CCG​ACA​GT; reverse, GCC​TCT​CAG​CCT​GTA​CAA​GAG; Integrated DNA Technologies).

The isolation of NE membranes was modified from previously described methods (Emig et al., 1995; Tsai et al., 2019). Briefly, WT and 4TorKO cells were collected from five 15-cm plates and centrifuged at 500× g for 5 min at 4°C. Cells were resuspended in cold PBS, and 100 µl of cells were set aside for whole cell lysate input controls. Cells were again centrifuged and resuspended in 5 ml of cold buffer A (10 mM Hepes, pH 7.4, 250 mM sucrose, and 2 mM MgCl₂) supplemented with 1 mM PMSF and incubated on ice for 10 min. Cells were then homogenized by passing through a 25-G needle five times. Homogenates were transferred to the top of 10 ml STM 0.9 buffer (50 mM Tris, pH 7.4, 0.9 M sucrose, and 5 mM MgCl₂) and sedimented at 1,000 g for 10 min. Pellets containing crude nuclear fractions were resuspend in 5 ml STM 1.6 buffer (50 mM Tris, pH 7.4, 1.6 M sucrose, 5 mM MgCl₂, and 1 mM PMSF). Suspensions were underlaid with 1 ml STM 2.1 buffer (50 mM Tris, pH 7.4, 2.1 M sucrose, and 5 mM MgCl₂), and 4 ml STM 0.8 buffer (50 mM Tris, pH 7.4, 0.8 M sucrose, and 5 mM MgCl₂) was added as the top layer. Pure nuclear fractions were sedimented by ultracentrifugation at 28,500 rpm (rotor SW41) for 65 min. Nuclear pellets were washed once in 1 ml TP buffer (10 mM Tris, pH 8.0, 10 mM Na₂HPO₄, and 5 mM MgCl₂) and sedimented at 1,000× g for 10 min at 4°C. Nuclear pellets were resuspended in 0.5 ml TP buffer supplemented with heparin (7.2 mg/24 ml buffer), 1 µl benzonase, and 2 mM N-ethylmaleimide (NEM) and rocked at 4°C for 2 h. Samples were centrifuged at 15,000× g for 10 min at 4°C. Supernatants containing NP fractions were saved, and pellets containing NE fractions were solubilized in 1 ml solubilization buffer (50 mM Tris, pH 7.5, 5 mM MgCl₂, 150 mM NaCl, 2% digitonin, 1 mM PMSF, and 2 mM NEM) on ice for 30 min. Samples were centrifuged at 15,000 g for 10 min at 4°C, and the supernatant was transferred to a clean microcentrifuge tube. A 15-µl aliquot was set aside to assess the quality of the fractionation.

Equal protein concentrations from WT and Torsin-deficient cells were immunoprecipitated. Immunoprecipitation was performed with 5 µl anti–K48-Ub (AB_11213655, Millipore) conjugated to protein A Dynabeads for 3 h at 4°C. Beads were washed three times with wash buffer (50 mM Tris, pH 7.5, 150 mM NaCl, 5 mM MgCl₂, and 0.1% digitonin), and proteins were eluted by heating to 65°C for 5 min in 30 µl SDS loading buffer. Eluates were subjected to SDS-PAGE on a Mini-PROTEAN precast gel (Bio-Rad). Immunoblotting was performed to assess the quality of the fractionation protocol with anti-SUN2 antibody (AB_1977547, Millipore), and anti-HA antibody (AB_390919, Roche) at a 1:5,000 and a 1:4,000 dilution, respectively. Gels for MS analysis were stained with SimplyBlue Safe Stain (Thermo Fisher Scientific), and gel samples were sent to the Yale National Heart, Lung, and Blood Institute Proteomics Center for liquid chromatography tandem MS. MS proteomic data were analyzed with Scaffold (Proteome Software Inc.).

EM was performed at the Center for Cellular and Molecular Imaging, Yale School of Medicine, with a previously described workflow (Laudermilch et al., 2016). Briefly, cells were fixed for 1 h in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer, pH 7.4. Following a brief rinse, cells were scraped in 1% gelatin and centrifuged in a 2% agar solution. Chilled cell blocks were processed with osmium and thiocarbohydrazide-osmium liganding as previously described (West et al., 2010), and samples were embedded in Durcupan ACM resin (Electron Microscopy Science). Polymerization was performed by incubating samples at 60°C overnight. These blocks were cut into 60-nm sections with a Leica UltraCut UC7, and stained with 2% uranyl acetate and lead citrate on Formvar/carbon-coated grids.

Samples for immunoelectron microscopy were processed as described above with some modifications. Samples were fixed through high-pressure freezing (Leica EM HPM100) and freeze substitution (Leica AFS) at 2,000 pounds per square inch. Freeze substitution was performed by incubating samples in 0.1% uranyl acetate/acetone solution for 50 h at −90°C. Following treatment, samples were washed in acetone, infiltrated in Lowicryl HM20 resin (Electron Microscopy Science) for 10 h at −45°C, transferred to gelatin capsules, and hardened via ultraviolet light exposure at −45°C. Blocks were subsequently sectioned and placed on Formvar/carbon-coated nickel grids for immunolabeling. Untreated aldehyde groups were quenched by incubating grids in 0.1 M ammonium chloride before the immunolabeling procedure. Samples were subsequently blocked in 1% fish skin gelatin in PBS, and grids were incubated in a 1:50 dilution of rabbit anti-GFP antibody (AB_390913, Roche) for single labeling, or a 1:100 dilution of mouse anti-Mab414 (AB_448181, Abcam), which required a rabbit anti-mouse bridge (JacksonImmuno), and either 1:50 dilution of rabbit anti-GFP antibody or 1:50 dilution of rabbit anti-MLF2 antibodies (raised against the full-length human MLF2 proteins with an N-terminal His6 tag; Covance) for double-labeling experiments. Protein A–gold (PAG) particles of 10 nm were used for single labeling experiments, while 5 nm and 15 nm PAG were used for double labeling experiments (Utrecht Medical Center). Grids were fixed using 0.5% glutaraldehyde, washed with PBS, dried, and stained using 2% uranyl acetate and lead citrate.

Both conventional EM and immunogold labeled samples were visualized with an FEI Tecnai Biotwin TEM at 80 Kv, and pictures were taken with Morada CCD and iTEM (Olympus) software.

Immunofluorescence staining was done as previously described (Laudermilch et al., 2016; Rose et al., 2014; Tsai et al., 2016). Briefly, cells were cultured on coverslips (VWR), fixed with 4% PFA in PBS, and permeabilized for 10 min with 0.1% Triton X-100 (Sigma-Aldrich) in PBS. Cells were blocked with 4% BSA (Sigma-Aldrich) in PBS for 10 min and incubated with the appropriate antibodies for 45 min. The following antibodies were used at a 1:500 concentration: anti–K48-Ub (AB_11213655, Millipore), anti-HA (AB_390919, Roche), anti-Flag (AB_259529, Sigma-Aldrich), and anti-VCP (ab11433, Abcam). After five washes, cells were blocked with 4% BSA and incubated with secondary antibodies conjugated to Alexa Fluor 488 (Life Technologies) or Alexa Fluor 568 (Life Technologies) for 45 min. Cells were washed with PBS, incubated in Hoechst 33342 (Life Technologies) for 5 min, and washed in PBS wash before being mounted onto slides using Fluoromount-G (Southern Biotech).

Standard wide field images were obtained on a Zeiss Axio Observer D1 microscope with a 63× oil immersion objective and an AxioCam MRm microscope camera using AxioVision 4.8 software (Zeiss). Samples were also imaged on a LSM 880 laser scanning confocal microscope (Zeiss) with a C Plan-Apochromat 63×/1.40 oil DIC M27 objective using ZEN 2.1 software (Zeiss).

Cells grown on coverslips were fixed in 4% PFA for 10 min and washed with PBS. Blocking and antibody dilutions were performed in 3% BSA in PBS with 0.1% Triton X-100. After 1 h of blocking, cells were incubated with an anti-Nup358 antibody (kind gift from Günter Blobel/Elias Coutavas, Rockefeller University, New York, NY) for 1 h at RT. Cells were washed three times with PBS for 5 min each and incubated with Alexa Fluor 568 goat anti-rabbit (Thermo Fisher Scientific) for 1 h at RT. Cells were washed as described above and then incubated with the Mab414 antibody (1:500, Abcam) for 1 h. Cells were washed as described above and then incubated with Alexa Fluor 488 goat anti-mouse (Thermo Fisher Scientific) for 1 h at RT. Cells were washed and mounted using Fluoromount-G (Electron Microscopy Science) before imaging.

3D-SIM imaging was performed on a DeltaVision OMX V3 system (GE Healthcare Life Sciences) equipped with a U-PLANAPO 60×/1.42 NA oil immersion objective lens (Olympus), CoolSNAP HQ2 CCD camera with a pixel size of 0.080 µm (Photometrics), and 488-nm, 561-nm, and 642-nm solid-state lasers (Coherent and MPB communications). Image stacks were acquired in 0.125-µm increments in the z axis in sequential imaging mode. Samples were illuminated by a coherent scrambled laser light source first passed through a diffraction grating to generate the structured illumination by interference of light orders in the image plane to create a 3D sinusoidal pattern, with lateral stripes ∼0.270 µm apart. The pattern was shifted laterally through five phases and through three angular rotations of 60° for each Z section, separated by 0.125 µm. Exposure times were typically between 25 and 150 ms, and the power of each laser was adjusted to achieve optimal fluorescence intensities between 2,000 and 4,000 in a raw image of 16-bit dynamic range, at the lowest possible laser power to minimize photo bleaching. Color channels were carefully aligned using alignment parameters from control measurements with 0.5-µm-diameter multi-spectral fluorescent beads (Thermo Fisher Scientific). Image acquisition was performed at RT.

The 3D-SIM images were subjected to SIM reconstruction and image processing using the SoftWoRx 3.7 imaging software package (GE Healthcare Life Sciences). The channels were then aligned in x, y, and rotationally using predetermined shifts as measured using a target lens and the SoftWoRx alignment tool (GE Healthcare Life Sciences). Subsequent image analysis was performed using Fiji.

The lattice light-sheet microscope was custom built based on the design shared by its original inventors (Chen et al., 2014), which also includes a LabVIEW-based acquisition software. The detection objective is a Nikon CFI75 water-dipping Apochromat with 25× magnification and 1.1 NA. The excitation objective is custom-made by Special Optics with 10× magnification and 0.65 NA. Imaging data were acquired with a sCMOS camera (Hamamatsu Orca Flash 4.0 v3). 4TorKO cells expressing mScarlet-Sec61β and MLF2-GFP under a Dox-inducible promoter were cultured on 5-mm-diameter coverslips (Warner Instruments) in complete media 24 h before cell cycle synchronization. Media was supplemented with 0.5 µg ml⁻¹ Dox 24 h before imaging. Coverslips were securely mounted into a temperature-controlled imaging chamber containing Leibovitz’s L-15 medium (11415064, Life Technologies) supplemented with 10% FBS equilibrated to 37°C. MLF2-GFP and mScarlet-Sec61β were excited using a 488-nm laser and a 560-nm laser, respectively, with inner and outer excitation NA of 0.325 and 0.4 (Chen et al., 2014), respectively, and 5-ms exposure times. Metaphase cells were identified, and 3D datasets were recorded with 20-s intervals between time points. All LLSM data were deconvolved with LLSpy (https://github.com/tlambert03/LLSpy), which is based on the Janelia open source software cudaDeconv (Janelia; https://www.janelia.org/open-science/lattice-light-deconvolution-software-cudadeconv).

All indirect immunofluorescence, confocal, and 3D-SIM images were processed for figures and analyzed with Fiji software (Schindelin et al., 2012). In addition to Fiji software, LLSM images were also processed with FluoRender 2.19 (Scientific Computing and Imaging Institute).

Quantifications of blebs were performed as previously described (Rose et al., 2014). Briefly, cells were randomly selected using the Hoechst channel and were imaged for K48-Ub staining. Nuclei were outlined manually, and the number of foci per nucleus was determined by the “Find Maxima” function in Fiji with a noise tolerance of 10. A threshold was set by comparison to WT nuclei, and cells above the threshold were determined to contain K48-Ub foci.

Line-scan analyses were performed in Fiji. The nuclear periphery of a given region was traced with the segmented line selection tool, and the same trace was superimposed on subsequent channels for consistency through the “Restore Selection” function. A plot of the intensity profile for the selected region was generated through the “Plot Profile” function, and all data were exported and graphed using GraphPad Prism.

Quantifications for the increase in fluorescence intensities of MLF2-GFP foci were performed in Fiji. Individual foci were traced with the freehand selection tool, and the average pixel intensities per foci area were determined through the measure function. Average fluorescence intensities are represented as relative intensity units(RIU), and the RIU values of each foci for a given time point are normalized by subtracting the background intensity of the same area of a respective focus. Data were exported and graphed using GraphPad Prism.

All statistical analyses were performed with GraphPad Prism. For parametric statistical tests, data distribution was assumed to be normal and therefore not formally tested.

Fig. S1 shows that nuclear herniations exhibit specificity in their molecular inventory. Fig. S2 MLF2 localization correlates with K48-Ub enrichment at the nuclear periphery. Fig. S3 shows that bleb formation does not require MLF2. Video 1 shows that bleb formation occurs rapidly and synchronously immediately after NE reformation during mitosis. Video 2 shows that MLF2-GFP foci localize to NE membranes. Video 3 shows that NE blebs remain stable during interphase. Video 4 shows that depleting POM121 abrogates bleb formation in Torsin-deficient cells.

We thank Joerg Bewersdorf and members of his laboratory for continued support, Morven Graham for skilled help with EM, the staff of the Yale/Keck MS and Proteomics resource and Felix E. Rivera-Molina for help with 3D-SIM, and members of the Schlieker laboratory for critical reading of the manuscript.

This work is supported by National Institutes of Health grant R01GM114401 (to C. Schlieker), National Institutes of Health grant 5T32GM007223-44 (to S.M. Prophet), National Institutes of Health grant GM105672 (to C.P. Lusk), and the Dystonia Medical Research Foundation (C. Schlieker, A.J. Rampello, and C.P. Lusk).

The authors declare no competing financial interests.

Author contributions: A.J. Rampello, E. Laudermilch, C. Zhao, N. Vishnoi, L. Shao, S.M. Prophet, C.P. Lusk, and C. Schlieker conceptualized and designed experiments in the text. A.J. Rampello, E. Laudermilch, C. Zhao, N. Vishnoi, L. Shao, and S.M. Prophet performed experiments. A.J. Rampello, E. Laudermilch, C. Zhao, N. Vishnoi, L. Shao, S.M. Prophet, C.P. Lusk, and C. Schlieker analyzed and interpreted data. A.J. Rampello, S.M. Prophet, and C. Schlieker wrote the original manuscript. A.J. Rampello, E. Laudermilch, C. Zhao, N. Vishnoi, L. Shao, S.M. Prophet, C.P. Lusk, and C. Schlieker revised and edited the manuscript.