Postmitotic nuclear pore assembly proceeds by radial dilation of small ER membrane openings

The nuclear envelope has to be reformed after mitosis to create viable daughter cells with closed nuclei. How membrane sealing of DNA and assembly of nuclear pore complexes (NPCs) are achieved and coordinated is poorly understood. Here, we reconstructed nuclear membrane topology and structure of assembling NPCs in a correlative three dimensional electron microscopy time-course of dividing human cells. Our quantitative ultrastructural analysis shows that nuclear membranes form from highly fenestrated ER sheets, whose shrinking holes are stabilized and then dilated into NPCs during inner ring complex assembly, forming thousands of transport channels within minutes. This mechanism is fundamentally different from interphase NPC assembly and explains how mitotic cells can rapidly establish a closed nuclear compartment while making it transport-competent at the same time.

microscopy time-course of dividing human cells. Our quantitative ultrastructural analysis 23 shows that nuclear membranes form from highly fenestrated ER sheets, whose shrinking holes 24 are stabilized and then dilated into NPCs during inner ring complex assembly, forming 25 thousands of transport channels within minutes. This mechanism is fundamentally different 26 from interphase NPC assembly and explains how mitotic cells can rapidly establish a closed 27 nuclear compartment while making it transport-competent at the same time. Introduction: 30 The nuclear pore complex (NPC) is the largest non-polymeric protein complex in eukaryotic 31 cells and composed of multiple copies of around 30 different proteins termed nucleoporins 32 (Nups) 1 . NPCs are the sole gates of macromolecular transport across the double membrane of 33 the nuclear envelope (NE). In higher eukaryotes, NPCs and the NE disassemble at the beginning 34 of mitosis and their rapid reformation during mitotic exit is essential for establishing a 35 functional nucleus in the daughter cell 2-4 . 36 The process of postmitotic assembly of the NPC and the nuclear membranes from mitotic 37 ER has been studied in vitro using nuclei assembled in Xenopus egg extract and by live cell 38 imaging using fluorescence microscopy. Several molecular players regulating the process have 39 been identified, including inner nuclear membrane proteins, ER shaping proteins such as 40 reticulons, nuclear pore components ELYS and Nup107-160 complex, nuclear transport 41 receptors and Ran 2-4 . In addition, kinetic observations of the bulk NPC formation across the 42 NE has shown that postmitotic assembly proceeds by sequential addition of Nups in a clear 43 temporal progression, that is almost identical between rodent and human cells 5,6 . 44 Despite these important insights, the mechanism of NPC assembly after mitosis has 45 remained unclear and is highly debated 7-10 . Whether postmitotic NPC assembly is initiated in 46 an already sealed NE and the NPC is inserted into this double membrane by a de novo fusion 47 event 11,12 similar to NPC assembly during interphase 13 , or if it starts already on the naked DNA   Supplementary Fig. 1). Segmentation of membranes in close proximity to chromosomes 65 showed that the layer of mitotic ER that contacts chromosomes exhibits a high degree of 66 fenestrations (Fig. 1b,c and Supplementary Movie 1) as reported previously 16 . At early times 67 only about 10% of the chromosome surface was associated with ER, but starting at about 5 min 68 after anaphase onset (AO), the surface of ER-chromosome contacts increased rapidly covering 69 the chromosomes with newly formed NE within 2 min (Fig. 1b, EM data showed that at early times (up to 3.9 min) variably sized holes made up 43% of the 72 surface of the ER sheets contacting chromosomes (Fig. 1c,e) and that 59% of these 73 discontinuities displayed a diameter below 100 nm (Fig. 1f), i.e. on the order of NPCs. The 74 degree of fenestration and hole size in the ER sheets contacting chromosomes decreased rapidly 75 (Fig. 1c), with holes making up only 16% of the surface two minutes later and now 75% of 76 them having a diameter below 100 nm (Fig. e,f; 6.3 min). This data demonstrates that the NE 77 forms from highly fenestrated ER sheets that contain a very large number of discontinuities 78 whose diameter shrinks as the ER-derived NE covers the chromosomes (Fig. 1d).

80
Coverage of chromosomes by nuclear membranes is closely linked to pre-pore formation 81 As ER fenestrae started to shrink significantly as early as 4.3 min after AO (Fig. 1f), many of 82 the pore sized discontinuities started to contain electron dense material (Fig. 1a- appearance, the number of such pre-pores increased rapidly to 2000 with the local density of 85 15 pores/m 2 within only 3 min (Fig. 1d). Kinetic analysis of chromosome coverage by newly 86 forming NEs and the appearance of pre-pores showed that both processes display sigmoidal 87 kinetics and are closely linked in time with pre-pore appearance reaching its half-maximum 88 within less than one minute after chromosome coverage (Fig. 1d). Knowing when exactly pre-pores start to form during NE formation, enabled us to examine the 92 architecture of assembling NPCs at an even higher resolution. We performed correlative live 93 imaging with electron tomography, in cells captured every 1-2 min after AO ( Fig. 2a and   94 Supplementary Movie 3), starting at 4.8 min when pre-pores first appear (Fig. 1d) until 15 95 min when NE formation is completed (Fig. 1b,d). Since NE sealing is delayed in the so called 96 'core-regions' due to clearance of dense spindle microtubules 4 (Fig. 1b), we focused our 97 analysis on the non-core regions of the NE (Fig. 2a). In a total of 27.8 m 2 reconstructed NE 98 surface area, we identified 360 particles consistent with pre-pores (i.e. displaying a NE 99 discontinuity containing regular electron dense material) captured at different times of 100 postmitotic assembly (Fig. 2b, Supplementary Fig. 2a and Supplementary Table 2). At early 101 times we also found 50 small NE discontinuities which were very similar to holes present in 102 the ER not yet in touch with the chromosome surface (Supplementary Fig. 3). We classified 103 these as "small holes" in NE or ER whose associated density is too low to be detectable as a 104 distinct regular structure, although they might contain smaller amounts of proteins. 105 We first focused our analysis on changes in NE topology. Tracing of the pre-pore 106 membrane profiles in the 3D tomograms and their quantitative analysis revealed that pre-pore 107 diameter increased rapidly from 39 nm (4.8 min after AO) to 63 nm (10.2 min) at which size 108 they stabilized (Fig. 2b,c). The profile analysis also revealed other interesting NE topology 109 changes ( Supplementary Fig. 2b,c). The pre-pore dilation showed sigmoidal kinetics, reaching 110 its half-maximum within 1.2 min after pre-pore appearance (Figs. 1d and 2c), predicting that it 111 represents a maturation step into fully assembled NPCs. Detailed analysis of the distribution of  Table 2). Interestingly, at the beginning of pre-pore appearance (4.8 117 min), the slightly lower than expected density of smaller pre-pores was made up by the presence 118 of similarly sized small NE holes, which disappeared at later times (Fig. 2d). Overall, this data 119 indicates that pre-pores mature by membrane hole dilation into fully-assembled NPCs and that 120 small NE holes are likely to be precursors of pre-pores that have not yet accumulated a 121 significant amount of dense material. 122 123 NPC assembly proceeds by centrifugal formation of a membrane associated ring 124 We next analyzed how the distribution of dense material inside pre-pores changes during their 125 maturation. We first radially averaged all density inside the membrane hole in the NE plane of 126 pre-pores (Fig. 3a). The change in mean radial intensity profiles over time showed that initially 127 (4.8 min), pre-pores contained material in the center of the membrane hole; From 4.8 to 10 min, 128 material progressively accumulated next to the membrane, resulting in a growing intensity peak  Figs. 4 and 6), we 139 ordered them independent of time based on structural similarity using spectral seriation ( Fig.   140 3b,c). Such spectral ordering of pre-and mature pores overall recapitulated their temporal 141 sampling during anaphase, with pores at early (4.8 and 6.1 min), middle (7.7 min), and late (10, 142 15 min and interphase) time points ranked together ( Fig. 3c and Supplementary Fig. 7), 143 showing that postmitotic NPC assembly is indeed a progressive process. Based on their 144 structural similarity, we partitioned pores into five assembly states (Supplementary Fig. 7) and 145 performed subtomogram averaging. The averages revealed a striking progression of structural 146 changes during postmitotic NPC assembly (Fig. 4a). Early and smaller pre-pores (cluster 1), 147 exhibited dense material in the center of a narrow membrane gap, which subsequently shifted  Fig. 8a,b), showing that the pore maturation process is largely synchronous.

158
Analysis of the increase in inner ring complex intensity over time in time-clustered averages 159 furthermore showed that its sigmoidal rise coincides with the process of membrane dilation 160 ( Supplementary Fig. 8c), suggesting that inner ring complex self-assembly could drive pore 161 dilation.

162
The structural progression we observed is consistent with previous observations in live cells 163 that nuclear/cytoplasmic ring components (Nup107 and Nup133) start to accumulate in the NE 164 early, followed by the inner ring component Nup93 5,6 . We revisited these observations, by   (Fig. 4b,c). The fact that ER sheets that form the NE contain a sufficient 195 number of small discontinuities for NPC assembly already at early time points (3.1, 3.9, and 196 4.3 min; Fig. 1f), and that we did not observe holes smaller than 20 nm in the NE although our 197 resolution is sufficient to resolve them ( Fig. 2 and Supplementary Fig. 4), strongly suggests 198 that postmitotic NPC assembly starts in pre-existing small NE openings rather than by de novo 199 fusion into already sealed double membranes. While our EM data cannot exclude that individual 200 nucleoporins, that are not yet assembled into higher order structures, are already present in these 201 small holes, they are morphologically clearly different from pre-pores and we cannot detect 202 reproducibly positioned or ring-like densities in them ( Fig. 2 and Supplementary Fig. 3). Since 203 the first pre-pores containing such densities have a similar small size of around 40 nm diameter 204 (Fig. 2b,c), it is likely that the ER/NE hole shrinkage is stalled and stabilized by protein 205 accumulation in the center of the membrane hole. At later time points, pre-pores then dilate the 206 membrane hole to normal NPC size of about 60 nm during inner ring complex formation, the 207 cytoplasmic ring assembles and the central channel matures (Fig. 4b,c). The fact that an inner 208 ring component Nup205 starts to be incorporated at 6 min after AO in live cells 209 (Supplementary Fig. 10), argues that the density in the center of pre-pores at 4.8 min may not 210 be explained by Nup205 and according to our earlier work also not much Nup93 5 . The most 211 likely candidate protein to explain this early density would therefore by exclusion be Nup155, 212 which forms the innermost layer of the inner ring complex 19 and has indeed been shown to be 213 required for recruiting other inner ring components Nups205, 188 and 93 20,21 . 214 We have previously shown that de novo assembly of NPCs into intact nuclei during 215 interphase proceeds via inside-out extrusion of the INM and fusion with the ONM 13 , which is 216 fundamentally distinct from the dilation mechanism of pre-existing membrane holes during 217 postmitotic assembly reported here. While interphase NPC assembly takes about 45 min and is 218 sporadic and rare 13 , the rapid radial dilation of small membrane holes concomitant with inner 219 ring complex formation supports the assembly of ~2000 postmitotic NPCs in only 3 min during 220 NE sealing after mitosis. One of the reasons for this very high efficiency could be that assembly 221 into the holes of highly fenestrated mitotic ER sheets does not require a new membrane fusion 222 at the assembly site that is needed for interphase NPC assembly 13 and could represent a rate-223 limiting step. In addition, the mitotic cell contains a high concentration of "assembly ready" 224 NPC subcomplexes, that become permissive for assembly synchronously by the reduction in 225 mitotic kinase activity 22 , while an interphase cell has to synthesize nucleoporins for assembly. 226 Combined, this could explain the high efficiency of postmitotic NPC assembly that is essential 227 for the rapid establishment of functional nuclei to exit mitosis. Our finding that the NPC 228 assembles via a fundamentally different mechanism in mitosis than in interphase provides the 229 basis to dissect the key structural and molecular transitions and regulatory steps in the future.       The remaining imprecise labelling was removed manually in MIB, and nuclear pores were 388 added in IMOD. For Fig. 1c, the contrast of the FIB-SEM image was enhanced by projection 389 of a series of 2-5 images for presentation purposes. For the analysis of chromosome coverage by the NE, the 3D models of fully-segmented 393 chromosomes and ER were converted to lists of points every 10 pixels in IMOD. The minimal distance from each point on the chromosome surface to ER was measured in 3D space, and the 395 ER within 100 nm away from chromosome surface was defined as the NE.

396
For measuring the size of membrane holes, segmented ER regions were connected in each 397 Z-slice based on connected component analysis and location information. Each of the connected 398 components in a slice was labeled first and its centroid was detected. A 2D matrix was generated 399 that represents pairwise inter-points Euclidean distances of all centroids of ER regions. ER 400 regions were then connected sequentially by utilizing this matrix in a customized breadth first 401 search manner. Specifically, two ER regions having the shortest pairwise distance were selected 402 in the first step. These two regions were connected by a straight line that bridges their nearest 403 boundary pixels. In the second step, previous two regions were used as references to identify a 404 third ER region that has the shortest distance from one of the two references. This region was 405 connected to the closest reference region and taken as the second reference region for next step 406 replacing the reference region it connects with. The connected network was grown gradually in 407 similar fashion by selecting a new region with respect to the two references. One of the 408 references was updated in each step and no region was connected more than twice. Thus, for n 409 number of regions n-1 lines were generated that connects 2 × (n-1) boundary points of the ER   Supplementary Fig. 8, and 32 slices (24 nm) for top views ii in Fig. 4a. The surface area of 428 the NE analyzed by EM tomography was measured, and the nuclear pore density was calculated 429 as described previously 13 . The overall structural similarity of the nuclear pores in freeze-430 substituted and heavy metal-stained cells to the respective cryo structures 13 indicates a good 431 structure preservation and representation of actual molecular density of nuclear pores. Also 432 negative staining analysis has been often done together with averaging and even molecular 433 interpretation. Our interpretation relies on rather low resolution features such as the presence 434 or absence of an entire ring, all of this is clearly apparent also prior to averaging (Figs. 2 and 4   435 and Supplementary Fig. 8).

437
Membrane profile analysis 438 For Fig. 2b and Supplementary Fig. 3, the ER and nuclear membranes were manually traced 439 and aligned as described in a previous report 13 . The tip-to-tip distance, the ONM/INM distance,  Fig. 3a and Supplementary Fig. 8c, and on 15 nm-Z-projected images for Supplementary 457 Fig. 3b. For measuring inner ring intensity in Supplementary Fig. 8c, the region comprising 458 80% of the pore radius was regarded as inner ring, and the average intensity of the local 5 nm 459 thick volume was quantified.  in each mode. U 4 in particular can be seen as containing the coordinates of each nuclear pore 483 on latent axes. From this, we computed the similarity between nuclear pores as Si,j = 1/(1+Di,j), 484 where Di,j is the Euclidean distance between pore i and pore j in the space defined by the 5 485 components of U 4 .

486
Seriation aims at finding a good ordering of items such that similar items are close to each 487 other in the sequence and dissimilar items are further apart or, equivalently, dissimilarities 488 between items increases with their rank difference. This can be formalized using the 2-sum 489 criterion which multiplies the similarity between items by the square of their rank difference.

490
The problem is then to find the optimal ordering that globally minimizes this criterion:    Table 2. We picked up all the 545 membrane holes which were visible in the EM tomograms and less than 80 nm in diameter.

546
Since for some pores the membrane contrast was too low to perform membrane profile analysis 547 due to high noise, we excluded them form the quantitative structural analysis and used them 548 only for measuring the pore density in Fig. 2d and Supplementary Fig. 5c. In addition, the 549 nuclear pores near gold particles were removed from the spectral ordering, since the gold 550 particles used for tomography alignment gave high electron density that interfered with the  Code availability 558 The computer codes used in this study are available on reasonable request from the 559 corresponding author.