A deep proteomics perspective on CRM1-mediated nuclear export and nucleocytoplasmic partitioning

1The nuclear envelope (NE) separates the cell nucleus from the cytoplasm. Although its lipid bilayers are impermeable for macromolecules, embedded nuclear pore complexes (NPCs) allow an exchange of material between these compartments. The NPC permeability barrier controls this exchange. It grants small molecules a free passage, but becomes increasingly restrictive as the size of the mobile species approaches or exceeds a diameter of ≈ 5 nm. Shuttling nuclear transport receptors (NTRs) are not bound by this restriction (for review see: ). They can traverse NPCs by facilitated translocation and have the capacity to ferry even large cargoes, such as newly assembled ribosomal subunits, across the barrier.

2Active transport of cargoes against concentration gradients requires an intact NE and NPC-barrier for retaining already transported cargoes in the destination compartment. In addition, it requires an input of metabolic energy, typically by the RanGTPase system. The corresponding duty cycles include not only a switch of Ran between its GDP- and GTP-bound states, but also one round of Ran-shuttling between nucleus and cytoplasm. The energetic coupling occurs through a primary RanGTP-gradient, which is generated through NTF2-mediated import of RanGDP, nucleotide exchange by the nuclear RanGEF (RCC1) and cytoplasmic RanGTP-depletion by RanGAP and RanBP1/RanBP2. This gradient then directly fuels importin- and exportin-mediated transport cycles.

3Exportins (reviewed in ) bind their cargo molecules cooperatively with RanGTP inside nuclei, carry them as trimeric RanGTP·exportin·cargo complexes to the cytoplasm, where GTP-hydrolysis triggers release of cargo and Ran. The free exportin can then re-enter nuclei and export the next cargo.

4CRM1, also called exportin 1 or Xpo1, is the most conserved nuclear export receptor. It is structurally well characterised, whereby structures of free CRM1, of certain cargo•CRM1•RanGTP complexes as well as their assembly- and disassembly-intermediates have been solved.

5CRM1 recognises its cargoes through short linear nuclear export signals (NESs), which comprise 4–5 critical hydrophobic (Φ) residues with characteristic spacings. CRM1 is essential for viability and is the target of the potentially lethal bacterial toxin leptomycin B, which blocks NES-binding by covalently modifying a conserved cysteine residue within the NES-binding site. CRM1 is known to carry multiple cargoes, including newly assembled 40S and 60S ribosomal subunits, the signal recognition particle SRP, U snRNAs, or the genomic RNA of HIV1 via the Rev protein. It is also known to regulate key cellular events by conditional export of transcription factors or cell cycle regulators from the nucleus (see e.g.;; ).

6Cell nucleus and cytoplasm are prime examples for the division of labour in a eukaryotic cell. The cytoplasm hosts the machineries of the secretory pathway, many metabolic activities as well as the cytoskeletal structures that account for cell motility and long-range transport. It has also specialised in protein synthesis and de novo protein folding.

7The nucleus lacks protein synthesis and thus depends on protein import from the cytoplasm. It has specialised in DNA replication and repair, nucleosome assembly, transcription, ribosome assembly, as well as in mRNA splicing and processing. Such specialisation critically relies on a spatial separation of interfering activities: Intranuclear protein synthesis, for example, would be a particularly wasteful process, because ribosomes would then also translate unspliced or incompletely spliced mRNAs, consequently read into introns, add inappropriate residues to the nascent chains, eventually encounter premature stop codons and thus produce truncated protein fragments. Such aberrant translation products would not only be non-functional, but probably also toxic, because they fail to fold, or act in a dominant-negative fashion.

8It is thus not very surprising that eukaryotic cells have implemented several lines of defence against intranuclear translation, whereby the NE acts as a primary barrier to keep cytoplasmic translation activity out of nuclei. Likewise, even though the 40S and 60S ribosomal subunits assemble inside the nucleus, they gain full translation competence only following late maturation steps in the cytoplasm (reviewed in; ).

9A very general problem is, however, that the NPC barrier is not perfect. Instead, also objects larger than the nominal exclusion limit can leak—albeit slowly—into the nucleus. Such slow mixing of nuclear and cytoplasmic contents would become a problem if the leaked-in proteins would interfere with nuclear functions or dysregulate cellular activities. Countermeasures might be selective degradation or inhibition in the inappropriate compartment, or, when mis-localised to the nucleus, recognition by an exportin and subsequent retrieval to the cytoplasm.

10Indeed, precedents for such exportin-mediated back-sorting of cytoplasmic proteins from the nucleus are already known. Animal Xpo6, for example, keeps actin out of the nucleus, while Xpo4 and Xpo5 do the same for the translation elongation factors eIF5a and eEF1A respectively. CRM1 was shown to expel several cytoplasmic factors from the nuclear compartment, including the RanGTPase system components RanBP1 and RanGAP as well as the translation factor subunits eIF2β, eIF5, eIF2Bε and eRF1. The full extent of active cytoplasmic confinement has, however, not yet been assessed.

11We report here global scale analyses of nucleocytoplasmic partitioning in Xenopus oocytes and of CRM1-mediated nuclear export. According to stringent criteria, we identified ≈ 1000 potential CRM1 cargoes from Xenopus laevis oocytes, ≈ 1050, from human HeLa cells, and ≈ 700 from the yeast S. cerevisiae. We tested a subset of cargo candidates for CRM1-dependent nuclear export in cultured human cells and thereby validated a majority of them as true cargoes. For a subset, we also confirmed direct CRM1-interaction and mapped the corresponding NESs, some of which turned out to have unusual features. The majority of identified CRM1 cargoes are proteins and protein complexes with a very strong bias towards a cytoplasmic localisation, suggesting that their active back-sorting from the nucleus is a major cellular activity. This applies to nearly all translation factors (including the largest translation factor complexes like eIF2B), to factors involved in final ribosomal maturation steps, which might prevent ribosomes from acquiring translation competence already in the nuclear compartment, as well as to regulatory proteins, autophagy-linked factors, peroxisome biogenesis factors and to centrosomal proteins, in both, humans and frogs. Another major functional group of CRM1 cargoes with perfect nuclear exclusion is represented by vesicle coat proteins, which points to a strong evolutionary pressure to preclude the budding of vesicles from the inner nuclear membrane. We also identified numerous new instances, where CRM1 appears to act in regulatory circuits. More generally, these data represent a very rich resource for other researchers seeking information about nucleocytoplasmic distribution and CRM1-controlled localisation.

12We were interested in a global view of how soluble proteins and protein complexes partition between the nucleus and the cytoplasm. In order to tackle this question, we applied a deep proteome analysis to the isolated compartments. A problem for such endeavour is that standard cell fractionation procedures rely on shearing forces, often combined with hypotonic lysis or even treatment with detergents (see e.g.; ). All these treatments compromise the integrity of the NE. Nuclear proteins, which are not firmly associated with solid structures like chromatin, will then leak out and contaminate the cytoplasmic fraction—just as the nuclear fraction will be contaminated by cytoplasmic components.

13In order to avoid these problems, we turned to Xenopus laevis stage VI oocytes. These cells measure ≈1.3 mm in diameter and have nuclei of ≈450 µm. Such very large dimensions allow for a manual oocyte dissection into nuclear and cytoplasmic fractions with exceptionally little cross-contamination (see e.g. ). These oocyte nuclei are also special with their volume being 100,000 times larger than that of average-sized cells with a G2 DNA contents. The chromatin should therefore make no more than a negligible contribution to nuclear retention of proteins. Instead, the nucleocytoplasmic distribution of a given protein or protein complex in these cells should be solely determined by its passive diffusion properties and by their potential to access active nuclear import and/or export pathways. In addition, oocytes are very long-lived cells that grow over months to their final size, which implies that even slow partitioning processes are likely to have reached a steady state.

14As a standard experiment, we dissected 60 oocytes, cleared the cytoplasmic fractions off yolk, normalised the nuclear and cytoplasmic fractions to their respective volumes, and identified proteins in three biological replicates by mass spectrometry. Proteins of two replicates were separated by SDS-PAGE (Figure 1A and 1B) and in-gel digested with trypsin. As a complementary approach, proteins of the third replicate were digested in solution. Resulting peptides were separated by reverse phase chromatography at pH 10 and obtained fractions analysed by LC-MS/MS. The raw data were searched against a comprehensive Xenopus laevis database. In this way, a total of 9573 proteoforms were identified, 7015 in isolated nuclei and 7036 in the cytoplasmic fractions. The intersection set comprised 4478 proteoforms (Figure 1C).

15Xenopus laevis is pseudotetraploid; thus it shows greater allelic diversity than other species, which represents a challenge for peptide-based protein quantification. We therefore treated allelic isoforms not as separate proteins, but mapped all recognizable allelic forms down to unique protein species. This ‘mapping’ was guided also by comparisons to the human and the (diploid) Xenopus tropicalis proteomes (see Materials and methods).

16We aimed at a high-quality dataset for those proteins that can actually pass through NPCs and partition between the nucleus and cytoplasm. We therefore tried to minimize the number of ‘contaminants’ in the list, which represent, for example, the endoplasmic reticulum (which co-purifies with nuclei in the form of the NE) or the mitochondrial cloud that is tightly associated with the outside of these nuclei. We subsequently tested which proteins disappeared from the nuclear fraction when the NE was manually removed. Such proteins were discarded from the list if their sequence features qualified them also as integral membrane, ER-luminal or mitochondrial proteins. This left us with a total of 5006 unique proteins. Of these, 1126 were identified only in the nuclear fraction, 1059 only in cytosolic fraction and 2821 in both compartments (Figure 1D, and Supplementary file 1).

17In order to quantify each of these proteins in the nuclear and cytoplasmic compartments, we employed the iBAQ strategy combined with an internal universal protein standard (UPS2; see Materials and methods). We estimate that our quantitation was reliable over a range of 5 orders of magnitude in abundance and accurate within a factor of 2.3 (see Figure 1—figure supplement 1).

18To validate the quality of the obtained nuclear and cytoplasmic fractions, we first evaluated the partitioning of the previously described marker proteins nucleophosmin (also called B23 or NO38) and gelsolin, which are localised exclusively to nucleus or cytoplasm, respectively. We chose highly abundant proteins, because these yield a larger number of unique, quantifiable peptides and thus allow for a more precise quantification. For nucleophosmin (a histone chaperone), we measured a nuclear concentration of ≈ 2.3 µM, a cytoplasmic concentration of 8 nM, and hence a nucleocytoplasmic (N:C) partition coefficient of 300:1. For gelsolin (a factor that stabilizes cytoplasmic actin in its G-form) we measured a nuclear concentration of ≈ 0.01 µM, a cytoplasmic of ≈ 4 µM, and thus an N:C ratio of 1:400. This suggests that the obtained nuclear and cytoplasmic fractions show only very limited cross-contamination and also suggested what range of partition coefficients should be expected also for other proteins.

19Supplementary file 1 contains complete and simplified data sets for the nucleocytoplasmic distribution of 5006 individual proteins and 9573 proteoforms. It lists the measured concentrations in nucleus and cytoplasm as well as ratios on a log10 scale. To make the data as accessible as possible to other researchers, we included not only unique identifiers for each protein hit, but also clickable annotation links to the corresponding UniProt entries for the identified Xenopus proteins (X. laevis if available, otherwise X. tropicalis) as well as to the human orthologues, where annotations as of now are more complete.

20When broken down to functional groups, it becomes evident that the various cellular processes represent quite characteristic N:C distribution patterns. As these patterns are in many cases tightly linked to the activity of CRM1, we will discuss these after describing our approach of mapping the CRM1-dependent nuclear exportome.

21An apparently CRM1-independent aspect relates to the energy supply of the giant oocyte cell nucleus. Efficient duty cycles of numerous enzymes, such as Ran, require a high NTP:NDP ratio, which seems hard to maintain inside these nuclei if ATP were produced only outside (i.e. by mitochondria or cytosolic glycolysis). The problem arises from the short half-life of ATP in living cells (≈ 1 min), which is less time than an ATP molecule would typically need to diffuse into and across such large nucleus. We now found evidence in the oocyte for two parallel solutions to this problem.

22The first is a ‘creatine-creatine phosphate energy shuttle’, which uses diffusion of creatine phosphate (CP) instead of ATP for a long-range transport of energy equivalents (reviewed by ). It exploits the ≈160 times higher phosphorylation potential of CP as compared to ATP, and operates by creatine kinase (CK) synthesising CP from ATP near ATP sources and the reverse reaction at sites of ATP consumption. Oocytes enable such CP shuttle by maintaining high CK concentrations in both, cytoplasm and nucleus (3 µM and 1.5 µM, respectively; Supplementary file 1).

23The second solution concerns the glycolysis pathway. The oocyte nucleus contains all glycolytic enzymes, except for hexokinase and phosphofructokinase that catalyse the ATP-consuming preparatory steps. The hexokinase level is generally very low in oocytes (because they produce glucose 6-phosphate by other means; see and Supplementary file 1), while 6-phosphofructokinase (PFK) is confined to the cytoplasm (N:C ≈ 1:2000; Supplementary file 1). This enzyme distribution predicts a directed flux of fructose 1,6 bisphosphate into the nucleus. It further suggests that oocyte nuclei produce the high-energy compounds phosphoenolpyruvate and 1,3 bisphoglycerate locally and use them for synthesizing ATP and other NTPs.

24The exportin CRM1 is known to be essential for viability and to account for nuclear export of numerous targets (see e.g.; for a comprehensive summary of so far identified substrates). However, it has been unclear how broad the cargo spectrum really is and which set of cellular processes are directly or indirectly controlled by CRM1-dependent nuclear export.

25In order to close this gap, we set out to identify cargoes in an unbiased manner from three different species and types of cells, namely: the already-mentioned Xenopus laevis oocytes, human HeLa cells and the yeast S. cerevisiae. We first prepared cellular extracts, used immobilised CRM1 as an affinity matrix and asked which proteins and protein complexes would bind in a RanGTP-dependent manner. As detailed below, CRM1 has many interaction partners of widely different abundance and affinity, which implies that the majority of them bind to the immobilized exportin in a highly sub-stoichiometric manner. The challenge thus is to cleanly distinguish such highly sub-stoichiometric bands from non-specific interactors. To this end, we optimised several parameters, such as the way of immobilisation, the exportin:extract ratios (to minimise competition between cargoes), buffer conditions, incubation time, etc., in order to maximise the signal-to-noise ratio of the affinity chromatography (for details see Materials and methods).

26Figure 2A shows the starting extracts and CRM1-bound fractions of such affinity chromatographies, and it documents indeed a very large number of protein species that bound to CRM1 in a RanGTP-dependent manner. Using the mass-spectrometric approaches mentioned above, we identified in the starting extracts ≈2800 (Xenopus), ≈3900 (human) and ≈2600 (yeast) proteins (Figure 2B). In the ‘CRM1+RanGTP’ samples, ≈2300, 3000 and ≈2000 proteins were identified. About 14% of these had not been detected in the total extracts and these probably represent low-abundance proteins that had been highly enriched during CRM1 affinity chromatography.

27The lists of proteins identified in the CRM1+RanGTP-bound fractions include not only true CRM1 binders and CRM1 export cargoes, but, for sure, also false-positive ones. To classify a given hit as a promising candidate, we therefore relied not only on its mere identification in this fraction. We also considered to which extent it became enriched from the input extract (‘input enrichment’), to which extent its CRM1-binding had been stimulated by RanGTP (‘RanGTP-stimulation’), as well as its absolute abundance in the ‘CRM1+RanGTP’-bound fraction (‘Molar fractions in CRM1+RanGTP’) (Figure 3), which affects the accuracy of quantification and thus the reliability of the first two numbers. As a resource for other researchers, we organised the quantitative data in Excel sheets (Supplementary files 2–4), which contain not only the just mentioned numbers, but also clickable links to the corresponding UniProt entries, as well as cellular localisations derived either from databases (human and yeast) or measured directly (Xenopus).

28We divided the identified proteins into several categories, using as a criterion species-specific thresholds (Supplementary files 2–4, and Figure 3). These thresholds had been adjusted to best match the behaviour of proteins, whose specific CRM1-interaction is already established beyond doubt, as well as of proteins that are known not to interact with the exportin (see Materials and methods). CRM1-binders of the category A1 not only had to pass the most stringent thresholds in terms of RanGTP-stimulation of CRM1-binding (≥500-fold in the case of Xenopus) and enrichment from the input extract (≥3-fold), but also belong to the most abundant proteins in the ‘CRM1+RanGTP’-bound fraction (Note that the threshold of a 500 times stronger signal than in the negative (minus RanGTP) control is far more stringent than the standard of ‘2-fold’, which is used in most proteomics studies). A2 had an even stricter threshold for ‘input enrichment’ (≥100), but a relaxed one for the apparent RanGTP-stimulation. It includes cargoes like snurportin, which bind CRM1 so strongly that the affinity is still high even in the absence of RanGTP. In category B, one of the three criteria was relaxed, while the category ‘low abundant’ includes proteins that were only detected in the ‘CRM1+RanGTP’-bound fraction, but were not sufficiently abundant to qualify for category A.

29We assume that most identified proteins from the categories A, B and ‘low abundant’ are direct CRM1 interactors that either carry a functional NES or occur in stable complexes with NES-containing proteins. Thus, we identified ≈1000 probable cargoes each in Xenopus and human as well as ≈700 in yeast. This is far more than identified for any other NTR (see e.g. ), and represents ≈20% of all detectable nuclear or cytosolic proteins, only a small fraction of which (≈ 10%) had been proposed to be associated with CRM1 before (see Supplementary file 5). This suggests that CRM1 serves a far larger number of cargoes than previously assumed. To estimate our positive discovery rate of direct CRM1 cargos, we tested a subset of these candidates below.

30Of course, nucleoporins, and FG Nups in particular, were also identified as CRM1 ligands. We consider them, however, as part of the transport machinery and not as cargoes. Some of them bind CRM1 very strongly, for example, the human or Xenopus Nup214•88•62 complex or Nup358 (see Figure 3b, and e.g.; ).

31On the other end of the distribution, we identified ≈700 proteins in Xenopus, ≈900 in human and ≈600 in yeast, which were strongly selected against in the ‘CRM1+RanGTP’-bound fractions (Supplementary files 2–4, Figure 3). These represent proteins, whose nucleocytoplasmic partitioning is probably not directly affected by CRM1. Metabolic enzymes (of e.g. glycolysis, the pentose phosphate pathway etc.), protein folding factors, and exclusively nuclear proteins are over-represented in this ‘non-binder’ category.

32In between, we found a broad zone of ‘ambiguous’ proteins, which actually represent a continuum. Some of them bound still very specifically to the export-form of CRM1 (according to the ‘minus RanGTP’-control); yet, the binding was weak. We assume that these proteins become only transiently CRM1 cargoes (e.g. in response to the addition or removal of a post-translational modification) or that they only transiently associate with bona fide CRM1-cargoes. On the other end of the ‘ambiguous’ category, there are proteins, which appear to be ‘CRM1 non-binders’. However, their low abundance in the starting extract precluded any reliable judgement of how strongly they were selected against in the ‘CRM1+RanGTP’-bound fraction.

33The data set contains most of the previously well-validated CRM1 cargoes, such as the nuclear import adapter snurportin, or the nuclear export adapters NMD3 and PHAX. The vast majority of hits (≥ 90%), however, were so far not linked to CRM1-mediated nuclear export. We therefore decided to verify a sample of candidates according to a common scheme.

34The first was the Xenopus translation termination factor eRF3a. Its GFP-fusion was exclusively cytoplasmic in transfected HeLa cells (Figure 4), which is consistent with its experimentally determined localisation in Xenopus oocytes (Supplementary file 1). A CRM1-block by 10 nM leptomycin B, however, caused the fusion to equilibrate between nucleus and cytoplasm, suggesting a significant nuclear entry rate and rapid CRM1-dependent retrieval in undisturbed cells.

35We also transfected a fusion that included an SV40 nuclear localisation signal (GFP-NLS-eRF3a) to enforce a faster nuclear entry, which made the fusion indeed exclusively nuclear following Leptomycin B-treatment. In undisturbed cells, however, we observed a still nearly exclusively cytoplasmic localisation, suggesting that the eRF3a NES confers a considerably faster export from nuclei than nuclear import mediated by the (very strong) SV40 NLS. We identified this NES within the N-terminal unstructured region of eRF3a and confirmed its nuclear export activity by transfection assays as well (Figure 4).

36eRF3a binds also purified CRM1 in a RanGTP-dependent manner (Figure 5), suggesting that the interaction is direct and not bridged by another factor. Due to its efficient binding from Xenopus oocyte extract to CRM1, eRF3a was classified as a ‘category A’ cargo. Next, we also confirmed a far lower scoring ‘category B’ candidate, namely the 1b subunit of the Arp2/3 complex, as a directly-binding, bona fide CRM1 export cargo with an NES at its extreme C-terminus. In the light of the rather weak CRM1-binding of Arp2/3-1b (Supplementary file 2), its NES turned out to be surprisingly strong. When fused to GFP, it behaved like a supraphysiological NES and produced pronounced transport intermediates at NPCs (best visible in weakly expressing cells). This difference is, however, plausible in the context of the Arp2/3 complex structure (pdb 1K8K; ), which shows this NES packing against the rest of the chain. The rather loose packing and high local B-factor suggest, however, that this NES region is sufficiently mobile to get transiently exposed for subsequent CRM1-binding.

37In total, we tested 29 candidates from Xenopus, human and yeast, and validated 23 of them positive (Figures 4, 6, 7 and 8), suggesting that the majority of hits represent indeed CRM1 cargoes. Negative cases where, for example, the exclusively nuclear replication factor C (subunit 3) or xDDX6. Explanations could be an issue with NES-modulating post-translational modifications, a transport-independent interaction with CRM1 or that another subunit in a larger complex accounts for CRM1-binding.

38We reasoned that the latter scenario might apply to xDDX6, which has been reported to interact with Lsm14 (isoforms a and b;; ). Indeed, the transfected Lsm14b-GFP fusion was exclusively cytoplasmic, but shifted upon leptomycin B treatment to a nuclear localisation. Given that Lsm14 had even better CRM1-binding scores than xDDX6 and that the two factors were recovered in ≈1:1 stoichiometry within the ‘CRM1+RanGTP’-bound fraction (Supplementary file 2), it is most likely that Lsm14 is the direct CRM1-binder, while xDDX6 piggybacks. Similar considerations will probably apply to numerous other cargo candidates that occur in complexes with other proteins. Supplementary files 2–4 provide the information for interpreting such cases, because they list not only the already mentioned binding-scores, but also estimate the molar ratios in which cargo candidates were recovered in the ‘CRM1+RanGTP’-bound fractions.

39The sheer number of CRM1 cargoes already suggests a very broad impact of this nuclear export pathway on cellular physiology. Yet, there is a remarkable bias towards or against individual functional categories (Figure 9). Highly abundant metabolic enzymes (including glycolytic enzymes), for example, are greatly under-represented amongst the CRM1 cargoes. Here, it is remarkable that many of these CRM1 non-cargoes, including enzymes that are part of large complexes, show a rather even distribution between cytoplasm and nucleus of the Xenopus oocyte without having a detectable NLS (Supplementary file 2, sheets ‘glycolysis’ and ‘metabolic enzymes’). This suggests that a large size alone cannot guarantee a cytoplasmic confinement.

40Translation initiation factors represent the other extreme (Supplementary files 2–4, sheets ‘Translation factors’; Figure 9B). eIF2, eIF2B, eIF4B, eIF4G, eIF5 and eIF5B all behaved like high-scoring cargoes in either Xenopus, human or yeast, which coincides nicely with their complete exclusion from the nuclei of Xenopus oocytes. We assume that CRM1 serves here the purpose of suppressing intranuclear translation and possibly also avoiding an interference with ribosome biogenesis if translation factors bound pre-maturely to ribosomal assembly intermediates.

41Leakage from the cytoplasm into the nuclear compartment should be fast for small individual proteins, but slow for large entities. It is therefore remarkable that even very large translation factor complexes behaved like nuclear export substrates, examples being the 125 kDa eIF2αβγ complex (150 kDa with tRNA) or the 270 kDa eIF2Bαβγδε complex (Supplementary files 2–4, sheets ‘Translation factors’). This suggests that even the presumably slow leakage of large cytoplasmic assemblies into nuclei can be so deleterious that countermeasures are required.

42None of the translation elongation factors appeared to be a convincing CRM1 cargo, possibly because appending an NES to the elongation factors might be incompatible with efficient translation. Nevertheless, they are also subject to exportin-mediated nuclear export. Human EF1A was previously identified as the most prominent export cargo of Xpo5, while the elongation factor for proline-rich regions, eIF5A, is exported by Xpo4. It is thus well possible that other translation factors that fail to interact with CRM1 also use a more specialized nuclear export pathway.

43The translation termination factor eRF3a is again a bona fide CRM1 cargo (see Figure 4), as is eRF1 (Supplementary file 2 and ). Furthermore, CRM1-dependent nuclear depletion also applies to several translation-associated factors, such as the signal recognition particle SRP, where SRP54 appears to be the NES-carrying component (Figure 7), or the human start-methionine aminopeptidase MetAP2, which binds translating ribosomes and cleaves the start methionine from nascent polypeptides. The normally strictly cytoplasmic MetAP2 accumulates upon leptomycin B treatment in nucleoli (Figure 7), not only validating it as a true CRM1 cargo, but also suggesting that it can bind to assembling ribosomal subunits prematurely and possibly interfere with the maturation process.

44CRM1 is known to be essential for ribosome biogenesis. It exports pre-60S ribosomal subunits (rSUs) through the export adapter Nmd3, while Ltv1 and Rio2 behave like export adapters for yeast 40S rSUs. Yeast Arx1 and the Mex67•Mtr2 dimer also bind pre-60S rSU subunits and promote export through direct interactions with FG repeats. Furthermore, Ecm1, Bud20, Alb1, Tif6, Rlp24, Nog1, and Mtr4 are known to escort 60S rSUs, while Nob1 and Enp1 have been shown to accompany 40S rSUs to the cytoplasm, without remaining a constituent of mature ribosomes (reviewed in; ). We now found that yeast Enp1 (essential nuclear protein 1) behaves like an autonomous CRM1 substrate (Figure 8), suggesting that it might actually act as yet another adapter for CRM1-mediated export of 40S rSUs. The use of multiple export adapters, as in here, might not just be an issue of robustness and redundancies, but a more fundamental requirement for making such large cargoes sufficiently ‘soluble’ in the gel-like permeability barrier of NPCs.

45Furthermore, our analysis revealed that additional predominantly nuclear, ribosomal biogenesis factors behave like CRM1 cargoes, namely: Brx1, Pno1, Tsr1, Rpf1, Rlp7, Ytm1, Tsr3, Rix7, Erb1, Nsa1, Ssf1, Dim1, Nol10, Loc1, and Rpf2 (see Supplementary file 4 sheet ‘Ribosome biogenesis’). Thus, they might also escort rSUs to the cytoplasm, and possibly facilitate the export process by providing additional binding sites for CRM1.

46Newly exported 40S and 60S rSUs acquire translation competence only after a series of maturation steps in the cytoplasm. In the case of yeast 60S rSUs, this involves the activities of Drg1/ Afg2, Rei1, Reh1, Jjj1, Yvh1, Lsg1, Efl1/ Ria1 as well as of Sdo1 (reviewed in; ). While some of these might get loaded already inside the nucleus, it appears that Drg1, Rei1, Reh1, Lsg1, and Ria1 act exclusively in the cytoplasm. All these factors showed a strong and strictly RanGTP-dependent interaction with CRM1 (Supplementary file 4 sheet ‘Ribosome biogenesis’). This suggests that CRM1 keeps them cytoplasmic, possibly to avoid the occurrence of translation-active ribosomes inside the nucleus (at least Lsg1 and Rei1 behave like true CRM1 cargoes, see Figure 8). A cytoplasmic confinement of these factors appears crucial for yet another reason: They displace the export mediators Arx1, Nmd3, Mex67, and Mtr2 from newly exported 60S species. A premature intra-nuclear displacement by mis-localized cytoplasmic maturation factors could thus abort ribosomal export.

47Cytoplasmic maturation also involves the incorporation of additional ribosomal proteins, a prominent example being rpS3. rpS3 is highly abundant in the ‘CRM1+RanGTP’-bound fraction, and thus perhaps hindered (directly or indirectly) by CRM1 to bind 40S rSUs prematurely already inside nuclei. The same might apply to rpL24a.

48These enzymes are essential for translation, and a cytoplasmic confinement of tRNA-charging would also contribute to the exclusion of intranuclear translation. Nevertheless, nuclear aminoacylation has been proposed as a proof-reading step for correct pre-tRNA processing and maturation prior to export. This proposal was based on the observations that blocking aminoacylation in yeast results in nuclear accumulation of tRNA and that a similar treatment in Xenopus oocytes prevents tRNAs, which had been injected into the nucleus, from reaching an exclusively cytoplasmic steady state distribution. There are, however, also arguments against this scenario, foremost that a loss of cytoplasmic retention of tRNA by translation elongation factor eEF1A (which binds only aminoacylated tRNA) would explain the phenotypes as well. Furthermore, a structural and functional analysis of tRNA•Xpo-t•Ran export complexes revealed that this exportin proof-reads a correct 3'-CCA end, but cannot sense aminoacylation. The latter applies also to the alternative tRNA exporter Xpo5.

49To clarify this issue at least for Xenopus oocytes, we analysed the nucleocytoplasmic distribution of the aminoacyl tRNA ligases and found a strong cytoplasmic bias (Figure 9C; Supplementary file 2, sheet ‘Aminoacyl tRNA ligases’). The cytoplasmic concentration exceeded the nuclear one by more than a factor of 100 in most cases, and even the least excluded ones had ≈10 times higher levels in the cytoplasm than in the nucleus. This makes is rather unlikely that nuclear aminoacylation is a general proof-reading criterion prior to tRNA export. At least the Ser, Thre, and Asn tRNA-ligases showed a strong RanGTP-dependent interaction with CRM1. We tested the Xenopus Asn tRNA ligase by transfection assays and observed a perfect CRM1-dependent nuclear exclusion (Figure 6). Human Ser tRNA ligase and the α-subunit of the yeast Phe tRNA ligase behave the same way (Figures 7 and 8), suggesting that cells make a true effort to keep tRNA aminoacylation cytoplasmic.

50We also tested the human Cys tRNA ligase that failed to interact with CRM1 (Supplementary file 3). Nevertheless, the transfected GFP-fusion protein showed a perfect nuclear exclusion, which was also not impaired by leptomycin B-treatment (Figure 7B). How this cytoplasmic confinement is maintained is still unclear, but the formation of larger complexes or alternative export pathways are plausible possibilities. In contrast to a CRM1-dependent confinement, the cytoplasmic localisation can, however, not be maintained against a fused SV40-type NLS. These transfection experiments also exemplify the validation of a ‘CRM1-non-binder’ (for more such validations, see below).

51We identified several high-scoring CRM1 cargoes amongst cytoplasmic mRNA degradation factors, components of P-bodies and stress granules. These include Upf1, Upf2 and Upf3 (Supplementary files 2–4 and Figure 5), which function in nonsense-mediated decay (NMD) of incorrectly spliced mRNA during a pioneering round of translation (reviewed by ), as well as the de-capping enzymes (DCP1 and DCP2) and enhancers of de-capping (EDC proteins), which typically reside within P-granules and initiate mRNA degradation (reviewed by ). Thus, CRM1 probably enables cells to control nuclear and cytoplasmic RNA turnover independently from each other. This CRM1 function appears very well conserved from yeast to human.

52So far, no connection has been made between CRM-mediated nuclear export and vesicle formation along the secretory pathway (reviewed by ). Yet, we observed that COPI and COPII coat proteins as well as, e.g., the AP-1 and AP2 adapter complexes or the AP180 clathrin coat assembly protein behave like CRM1 cargoes (Supplementary files 2–3, sheets ‘Vesicle coat proteins’; as well as Figures 4 and 6). They are also extremely well excluded from the oocyte nucleus (Figure 9E). A formation of intranuclear vesicles was so far observed only in oocytes of rather exotic species, such as the ascidian Botryllus schlosseri. The absence in other cell types can now be explained by an active depletion of the vesicular budding machineries from the nuclear interior.

53The data also suggest an unanticipated connection of CRM1 to autophagy (reviewed in ). Atg1, Atg13, Atg17, and Vps30 are all required for autophagy in yeast and all of them are high-scoring CRM1 cargoes (Supplementary file 4, sheet ‘Autophagy’). It thus appears as if CRM1 counteracted an initiation of autophagy from the nuclear interior. The situation is very similar in human cells, though the spectrum of CRM1-interacting autophagy components is slightly different (Supplementary file 3 sheet ‘Autophagy’). For example, here the ATG8-conjugating enzyme ATG3 is a major CRM1-cargo (Figure 7).

54Our data set also revealed high-scoring CRM1 cargoes that make an unexpected link between nuclear export and protein import into peroxisomes (reviewed in ): The peroxisomal targeting (PTS1) receptor Pex5 was not only identified as a high-scoring CRM1-binder, but also showed a strictly CRM1-dependent nuclear exclusion (Figure 7). This connection might point to a general challenge for post-translational transport from the cytosol, namely that diffusive transport will not necessarily lead to the destination organelle, but also to and possibly into nuclei. In the case of peroxisomal proteins, this poses a particular danger as many of them produce reactive oxygen species that might damage the genome. A first line of defence against such incidents is a trapping of peroxisomal proteins by dedicated targeting receptors. The resulting receptor•substrate complexes, however, still need to reach peroxisomes by diffusive transport. If this fails and the complex ends up inside nuclei, then CRM1-mediated export can rectify the problem and give the targeting complex another chance to reach its correct destination.

55PEX19 from yeast or human is another example, which behaves like a perfect CRM1 cargo and shows a CRM1-dependent nuclear exclusion (Figures 7 and 8). PEX19 targets membrane proteins to peroxisomes and allows pre-peroxisomes to bud from the ER (reviewed in ). A mis-targeting of PEX19 to nuclei by a fused NLS causes interesting consequences, namely nuclear accumulation of newly synthesised peroxisomal membrane proteins. It now seems very likely that a loss of CRM1-mediated export would have the same effect.

56Apart from a cytoplasmic confinement of e.g. VASP or the Arp2/3 complex (Supplementary file 3, Figure 5), it appears that CRM1 has only little direct impact on the actin cytoskeleton, which is consistent with the fact that metazoans possess a dedicated exportin (Xpo6) to deplete actin from their nuclei.

57Xenopus and human septins, in contrast, are highly abundant amongst the CRM1-bound proteins. We validated Septin 2 and found that leptomycin B treatment of transfected cells caused indeed a shift from a cytoplasmic to a nuclear localisation (Figure 5). Septins interconnect actin and microtubule networks, function in cytokinesis, formation of cilia and defence against pathogens, and it appears that their nuclear accumulation needs to be actively suppressed.

58The data also suggests some contribution of CRM1 to nuclear exclusion of α- and β-tubulin in metazoan cells (Supplementary files 2 and 3). We further found high-scoring CRM1 cargoes amongst the components of the centrosomes (Figure 9F), which function as microtubule organizing centres (reviewed by ). These strong CRM1 interactors include the HAUS augmin complex, γ-tubulin and γ-tubulin complex components as well as a number of additional centriolar proteins (e.g. CEP41, CEP55, and CEP170). This could point not only to an active suppression of microtubule nucleation in metazoan interphase nuclei, but also to additional mitotic function of CRM1 (see )—namely at centrosomes.

59Almost every aspect of cellular physiology is under the control of protein kinases, phosphatases or the ubiquitin/ proteasome system. Our data now suggest that CRM1 is more heavily involved in such regulatory circuits than previously thought. Alone in HeLa cells, we found ≈ 70 kinases as high-scoring CRM1 cargoes (Supplementary file 3), only a fraction of which had been described as such before. The new cargoes include heavily studied kinases such as protein kinase A, the interleukin-1 receptor-associated kinase 1, the pro-apoptotic serine/ threonine kinase 3, the tank-binding kinase 1, Raf-1, and several isoforms of casein kinase I and II. In addition, we identified numerous kinase regulators as new CRM1 cargoes, examples being the already mentioned regulatory subunit of PKA (PRKAr2a; Figures 4 and 5) or the TSC1•TSC2 complex (Supplementary file 2), which is a key component of the mTOR signalling pathway. Such action of CRM1 will contribute to compartment-specific phosphorylation patterns, but in many cases also exert control by granting or denying those kinases access to their substrates.

60Likewise, numerous phosphatases and components of the ubiquitin or the ubiquitin-like modifier system showed up as potential new CRM1 cargoes, alone in human cells 22 and 57, respectively (Supplementary file 3, Figure 9H and 9K).

61CRM1-dependent nuclear export signals are usually short (9–15 residues long) peptides with 4–5 hydrophobic Φ residues that are spaced according to characteristic patterns. The most common one is a PKI-type NES. Several other spacings are also allowed, which can be explained either by such NESs binding in a different conformation to CRM1 (snurportin-type or Rev-type NES) or by skipping one Φ-residue in a 5Φ NES. A functional NES also needs to be solvent-exposed and not buried in a globular fold.

62We applied these criteria to identify NESs in a set of validated, new CRM1 cargoes. We were successful in six cases (Figure 5), but failed with some others (e.g. Pex5, rpS3, Haus1, Cop beta’, MetAP2, CDC37L, Enp1, Phe tRNA ligase α). These negative cases illustrate how difficult a reliable NES prediction still is. These difficulties originate from at least two problems. First, many published NESs turned out to be incorrect (see e.g. and discussion therein). NES prediction will therefore remain unreliable and produce frequent false-positive hits as long as it is based on an unreliable list of positive cases. Second, we probably miss true NESs, because we do not yet know all NES patterns.

63Our study should now provide solutions to these limitations. First, we provide large test sets for benchmarking current and future NES predictions, first of all a total of ≈1300 ‘category A cargoes’, which should contain an NES (in the case of oligomers at least one per complex). The subset of such true CRM1 cargoes, which lack a so far recognizable export signal, provides an ideal starting point for identifying NESs that conform to new patterns. We also found ≈2200 clear non-CRM1-binders that can serve as a negative control group for NES prediction.

64The just mentioned non-binders include several cases (11) that have been listed as CRM1-cargoes in the NESdb (Supplementary file 5; ). This could now point to the false-positives in the NESdb or to contaminating false-negatives in our dataset. To address this issue, we analysed seven of these conflicting cases and tested the behaviour of the corresponding GFP fusions in the HeLa cell transfection assay (Figure 7—figure supplement 1). All of them showed a considerable nuclear signal already in untreated cells and this nuclear signal did not further increase upon blocking CRM1 by leptomycin B. These seven re-tested candidates thus behaved as predicted from their classification as CRM1 non-binders, at least for the tested protein isoforms (cloned from HeLa cell cDNA), for the tested cell type (human HeLa cells) and under standard cell culture conditions.

65An interesting twist to CRM1-mediated export is that it can be regulated by post-translational modifications. Cyclin B, for example, is kept cytoplasmic by CRM1 until prophase, when the NES gets inactivated by phosphorylation of adjacent serine residues and the protein suddenly accumulates inside nuclei. PHAX, the export adapter for U snRNAs, on the other hand, requires phosphorylation for an efficient interaction with CRM1. Given these examples, it will now be very interesting to obtain a global view on how phosphorylation impacts individual cargo-CRM1 interactions, for example, by testing in how far CRM1 selects for or against the corresponding phosphoforms.

66Surprising outcomes of our study not only were the sheer number of CRM1 cargoes (≈ 1/4 of all detectable cytoplasmic and nuclear proteins), but also that the majority of cargoes are actually exclusively cytoplasmic proteins (Figure 9A; Supplementary files 2–4). This poses the questions of why there is a need to actively maintain a cytoplasmic localization and why evolution has not come up with a better barrier system?

67The perhaps best answer is that is impossible to preclude nuclear entry of unwanted proteins in the first place, because there are several leakage routes into the nucleus. First, the NPC permeability barrier is per se probably imperfect and allows leakage: It is based on the sieving effect of reversibly crosslinked FG repeat domains and probably represents compromise in the sense that a tighter barrier would also restrict facilitated transport (see and discussions therein).

68In addition, there are situations, where even the most perfect NPC barrier gets bypassed. The open mitosis in metazoans, for example, leads to a complete mixing of nuclear and cytoplasmic contents and requires an unmixing following reformation of the nuclear envelope. Nuclear proteins require re-import, while obligatory cytoplasmic proteins, such as translation factors, components of the vesicular transport machinery and many others, are obviously subject to active nuclear export. Another source of leakage might be NPC assembly, where inner and outer nuclear membrane might already fuse to a pore, before all FG Nups are in place to maintain a permeability barrier.

69Finally, cells should survive a temporary damage of their nuclear envelope, which is known to occur rather frequently in certain cultured cancers cells, but might also happen in normal cells, in particular when exposed to mechanical stress. Such incidents not only require a repair of the NE, but also a rapid unmixing of nuclear and cytoplasmic contents by active import as well as by active export. Thus, the barriers of the NE alone cannot maintain the compartment identities. Instead, a robust separation of nuclear and cytoplasmic contents requires active corrective mechanisms, whereby the exportin CRM1 appears to play a very central role.

70Manual microdissection of the oocytes was performed in ‘5:1/HEPES buffer’ (10 mM HEPES/ KOH pH 7.5, 83 mM KCl, 17 mM NaCl, supplemented with Roche complete protease inhibitor (EDTA-free) as previously described. Isolated nuclei were gently washed several times, and proteins were recovered by ethanol precipitation.

71The respective cytoplasmic fractions were diluted with ‘5:1/HEPES buffer’ and homogenized with a pestle. Pigments, yolk and membranes were removed from the extract by two rounds of centrifugation (17,000 g, 15 min, 4°C). Note that thereby also insoluble protein complexes, such as intermediate filaments were removed, and that our analysis explicitly aimed at the nucleocytoplasmic partitioning of soluble complexes and proteins. We therefore also excluded proteins that co-purified with nuclei through their association with the nuclear envelope (NE). Such proteins were identified and ‘flagged’ by a manual dissection of nuclei into nuclear interiors and crude NEs (kindly performed by Volker Cordes) and by asking which ‘nuclear’ proteins did not fractionate with the nuclear interiors.

72A total of 775 unique proteins (see below) were flagged as possibly not being cytosolic or intranuclear, 748 of them were removed from the list of Supplementary file 1, while 27 were kept in the final list because they were manually qualified as cytosolic or intranuclear proteins according to previous literature, or according to their subcellular localization at Human Proteome Atlas.

73Quantitative mass spectrometric analysis of the obtained nuclear and cytosolic fractions involved three biological replicates with two technical replicates each. For quantification of nucleocytoplasmic partitioning, we considered that a nucleus is 10-fold smaller in volume than a yolk-free cytoplasm. Therefore, we compared for each analysis ≈60 nuclei with ≈6 cytoplasms.

74In these experiments, we used biotinylated CRM1 versions carrying an Avi-tag, which is an optimised biotin-acceptor sequence for enzymatic biotinylation by BirA.

75HeLa S100 extract and Saccharomyces cerevisiae whole cell extract were kindly supplied by the group of Reinhard Lührmann. Xenopus laevis oocyte extract was prepared as described.

76Xenopus laevis oocyte, Saccharomyces cerevisiae and cytosolic HeLa extracts were diluted 1:5 in binding buffer (20 mM HEPES/ NaOH pH 7.5, 90 mM KAc, 2 mM MgOAc, 250 mM Sucrose, 5 mM DTT) and cleared by a 1 hr centrifugation step in a S55A rotor at 4°C and 37,000 rpm. Supernatants were incubated with Phenyl-Sepharose (low substitution) for the selective depletion of endogenous nuclear transport receptors as described previously. The flow-throughs were incubated with 0.1 μg/ml RNAse A for 20 min on ice (to detach ‘indirect’ cargoes that interact through RNA with direct ones) and then 100 units/ml RNasin (Promega) were added.

771 ml of a such treated extract was supplemented with 5 μM RanGTP (hsRanQ69L5-180), centrifuged in a S45A rotor for 1 hr at 4°C at 37,000 rpm. 500 pmoles mmCRM1 or scCRM1 were immobilized on 20 μl streptavidin agarose beads (Sigma Aldrich). Free biotin-binding sites were quenched thereafter with 1 mM biotin, and the beads were rotated for 3 hr at 4°C with the RanGTP-supplemented extract. Beads were then washed three times with 500 μl binding buffer. Bound material was eluted with 60 μl SDS sample buffer at 45°C. Minus CRM1 and minus RanGTP controls were processed analogously. Input extracts as well as elutions were analysed by SDS-PAGE followed by Coomassie-staining and/or analysed by mass spectrometry as described below.

78An overview of the sample preparation for mass spectrometry and LC-MS/MS instrumentation:

79To estimate absolute protein concentrations, Universal Proteomics Standard-2 (UPS2; Sigma-Aldrich) was added to the analysed samples (at a 1:10 (w/w) ratio between the standards and the total sample protein). Since UPS2 is based on human proteins, this standard was only employed for the non-human samples.

80In one workflow, proteins were first separated by SDS-PAGE (4-12% Bis/Tris gradient mini-gel, NuPAGE, Novex) and visualized by colloidal Coomassie-staining. Proteins were then in-gel digested as described before with minor modifications. Briefly, proteins were reduced with 10 mM DTT for 30 min at 55°C, and then alkylated with 55 mM iodoacetamide (IAA) in 50 mM ammonium bicarbonate (BC) for 20 min at 26°C in the dark. Protein digestion was performed overnight at 37°C at a 1:50 (w/w) trypsin (Promega #V5111) to protein ratio. Following digestion, peptides were extracted from the gel pieces, and concentrated by vacuum evaporation of the solvent in a SpeedVac to near dryness. Dried peptides were dissolved in 20 µl of 1% (v/v) formic acid, and 6 µl were analysed LC-MS/MS for each technical replicate.

81In the second workflow, ethanol-precipitated proteins were dissolved in 1% (v/v) RapiGest SF surfactant (Waters # 186002122) at 70°C for 10 min. Proteins were then reduced with 5 mM DTT in 50 mM BC for 30 min at 50°C, and alkylated with 10 mM IAA in 50 mM BC for 20 min. Excess IAA was reacted with an additional 5 mM DTT at RT for 20 min. Proteins were first digested with Lys-C (Roche, 1:100 enzyme to protein ratio) for 4 hr at 37°C, then overnight with trypsin at a final RapiGest concentration to 0.1%. Following digestion, the samples were acidified with trifluoroacetic acid (final concentration of 1%, v/v, 37°C, 1 h) to break down the RapiGest surfactant, the resulting by-products were pelleted by centrifugation (13,000 rpm, 15 min, RT), and the supernatant containing the digested peptides was transferred to a new tube. Peptides were desalted on reversed phase-C18 solid-phase extraction cartridges (SPE; SepPak, Waters) and concentrated in a SpeedVac to near dryness. Then, peptides were resuspended in 10 mM ammonium hydroxide (pH 10) and loaded onto a reverse phase HPLC column (XBridge C18, Waters, 3.5 µm, 1.0 mm x 150 mm) and eluted in a 5-35% (v/v) acetonitrile gradient at a flow rate of 60 µl/min. 45 initial fractions were collected, which were combined into 17 peptide pools. Each pool was concentrated as described above and dissolved in 20 µl 1% FA (v/v). 6 µl each were then analysed by LC-MS/MS for a technical replicate.

82First, extracted peptides were loaded onto an in-house packed C18 ‘trapping’ column (0.15 mm x 20 mm, Reprosil-Pur 120 C18-AQ 5 μm, Dr. Maisch GmbH, Germany). Then a second C18 column was connected in tandem (an analytical C18 capillary column; 0.075 mm x 250 mm column self-packed with 3 μm Reprosil-Pur 120 C18-AQ). Peptides were then eluted using a 105 min linear gradient (5– 35% acetronitrile in 0.1% FA at 300 nl/min) on an EASY nLC-1000 system in-line coupled to a Q Exactive hybrid quadrupole/orbitrap mass spectrometer (Thermo Scientific, Dreieich). The instrument was operated in data-dependent acquisition mode with a survey scan resolution of 70,000 at m/z 200 and an AGC target value of 1 x 106. Up to 15 of the most intense precursor ions with charge state 2 or higher were sequentially isolated at an isolation width of 2.0 m/z for higher collision dissociation (HCD) with a normalized collision energy of 25%. Dynamic exclusion was set to 30 s to avoid a repeating sequencing of the same precursor ion.

83The LC setup was as described above, but a Dionex Ultimate 3000 HPLC (Thermo Scientific, Dreieich) and a 350 mm capillary C18 column were used.

84Both mass spectrometers were operated in data-dependent acquisition mode with a survey scan resolution of 120 000 (for Orbitrap Fusion) or 60 000 (for Q-Exactive HF) at m/z 200, with an AGC target of 1 x 10e6. Up to 30 of the most intense precursor ions with charge state 2 or higher were sequentially isolated for HCD with normalized collision energy of 27% (Q Exactive HF) or 30% (Orbitrap Fusion), respectively. MS/MS scans were recorded in the Orbitrap for Q Exactive HF, and in the LTQ ion trap for the Orbitrap Fusion. Dynamic exclusion was set to 50 s.

85MS raw files were processed with the MaxQuant software package (version 1.5.0.30) and peak lists were searched with the in-built Andromeda search engine.

86FASTA Sequence Databases: for X. laevis samples an mRNA derived X. laevis protein database with 79,214 entries was used, for human samples a human UniProt FASTA database (download date: June 2014, 20,258 entries), and for yeast samples a S. cerevisiae Uniprot FASTA database (download date: June 2014, 6743 entries). These databases were supplemented with common contaminants (e.g. keratins, serum albumin) and with the reverse sequences of all entries for false discovery rate estimations.

87The Andromeda search engine parameters were: carbamidomethylation of cysteine was set as a fixed modification, whereas oxidation of methionine and N-terminal protein acetylation were set as variable modifications; tryptic specificity was considered with proline restriction; up to two missed cleavages were allowed; and the minimum peptide length was set to seven amino acids. The MS survey scan mass tolerance was set to 6 ppm, and MS/MS mass tolerances to 20 ppm (Orbitrap) and 0.5 Da (LTQ ion trap), respectively. The false discovery rate was set to 1% at both the peptide and the protein level.

88Several levels of criteria were applied for confident estimation of protein concentrations. First, peptides having posterior error probability (PEP) above 0.01 were excluded from the estimation of protein concentrations in the cytosolic and the nuclear fractions of X. laevis samples. Second, proteins identified with single ‘only identified by site’ peptides in only one compartment were excluded from the calculation of protein concentrations. The absolute protein concentration of proteins in the cytosolic and the nuclear fractions were estimated by correlation with the absolute concentrations of UPS2 standard proteins and their respective iBAQ intensities, assuming volumes of 50 nl for the nucleus and 500 nl for the yolk-free cytosol. A regression curve of the absolute amount of UPS2 standard proteins were plotted against their measured iBAQ intensity (log 10 scale) in each biological replicate to generate linear regression equations. These equations were then used to estimate protein concentrations in each biological replicate. Then, the nuclear-to-cytosolic (N:C) ratio was calculated from average protein concentration in the nucleus and the cytosol.

89Annotations were largely based on the UniProt database. For each hit, relevant UniProt data were fetched. These data were simplified to ‘simplified localization’ (Cytoplasm, Nucleus or Both) and Flags (transmembrane, mitochondrial, ER proteins).

90UniProt annotations of X. laevis proteins are still sparse. Therefore, human and X. tropicalis entries were used as additional references. Appropriate orthologues were mapped by blasting identified X. laevis contigs against human and X. tropicalis databases (with an E value cut-off of 10-11).

91Functional protein groups were acquired from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database if not stated otherwise.

92For assessment of protein complexes in Xenopus laevis oocyte and HeLa cells, human data sets from and were used. Yeast protein complexes were assigned according to. For previous CRM1 interaction data, BioGRID, the NESdb, and data from were used.

93Proteins identified with ‘only identified by site’ (MaxQuant version 15.0.30) within the ‘CRM1+RanGTP’ sample were excluded from further analysis. For X. laevis samples, these additional parameters were applied: peptides with PEP values higher than 0.01 were excluded, and a minimum of two unique peptides were required to consider a protein for quantification.

94For calculating the molar fraction of a given protein in the ‘CRM1+RanGTP’ sample, its iBAQ intensity was divided by the sum of iBAQ intensities of all proteins detected in this sample.

95‘Enrichment from input’ for a given protein was obtained by dividing its molar fraction within the ‘CRM1+RanGTP’ sample by its molar fraction in the input extract.

96The ‘RanGTP-stimulation’ for a given protein was calculated by dividing its iBAQ intensities in the 'CRM1+RanGTP’ and the ‘CRM1 w/o Ran’ samples.

97When a protein was not identified in the ‘CRM1 w/o Ran’ or ‘Input’ sample, then its iBAQ intensity was replaced by a conservative estimate, namely a baseline intensity for detectability (the median of iBAQ intensities of the least 30 abundant proteins in this sample). This was to avoid in the calculation of parameters divisions by zero and to avoid over-estimating the significance of low abundance cargo candidates.

98Based on these three parameters, distinct significance categories were constructed for three species as summarized in the ‘Category thresholds’ tables (Supplementary files 2–4). For proteins that are part of the nuclear transport machinery (Importins, exportins, Nups and related factors) separate categories (NUPs, NTRs, NPC, CRM1 cofactor) were assigned.

99mmCRM1, scCRM1 and hsRanQ69L5-180 were expressed with an N-terminal His14-ZZ-bdSUMO tag and purified by (i) immobilisation via Ni2+ chelate chromatography, (ii) on column bdSENP1 protease elution and (iii) gel filtration. Candidate cargoes and controls were expressed as His14-ZZ-bdSUMO fusions and purified via Ni2+ chelate chromatography and imidazole elution.

100250 pmoles of His14-ZZ-bdSUMO tagged cargo candidates and controls were immobilized on 20 μl anti-ZZ beads, and incubated with 300 pmoles mmCRM1 either in the presence or absence of 1500 pmoles RanGTP (hsRanQ69L5-180) in a volume of 500 µl. After 3 hr incubation at 4°C, bound material was eluted by adding bdSENP1 protease, and eluted fractions were analysed by SDS-PAGE and Coomassie-staining.

101Cargo and NES candidates were cloned behind either GFP or a GFP-NLS (SV40) module in modified pEGFP-C1 (Invitrogen) vectors. Also a control vector coding for a RFP-NLS-NES fusion was prepared for cotransfection with the GFP/GFP-NLS constructs. HeLa Kyoto cells were grown on coverslips in 24-well plates, and transiently cotransfected with FuGENE6 (Promega) according to manufacturer’s instructions. After 24 hr, cells were incubated with either 10 nM Leptomycin B (LMB, dissolved in DMSO) or DMSO alone for 3 hr. They were then fixed for 30 min with 3% paraformaldehyde and 0.1% glutaraldehyde, and the aldehydes were quenched with 1 mg/ml NaBH4. Fluorescence signals were recorded on an SP5 confocal laser scanning microscope (Leica), using sequential scans with excitations at 488 and 561 nm.

102The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD002899. The dataset of high confidence CRM1 interactions (cargo categories A, B, and ‘low abundance’) has been submitted to the IntAct databases with the identifier IM-24624.