A peptide tag-specific nanobody enables high-quality labeling for dSTORM imaging

Dense fluorophore labeling without compromising the biological target is crucial for genuine super-resolution microscopy. Here we introduce a broadly applicable labeling strategy for fixed and living cells utilizing a short peptide tag-specific nanobody (BC2-tag/bivBC2-Nb). BC2-tagging of ectopically introduced or endogenous proteins does not interfere with the examined structures and bivBC2-Nb staining results in a close-grained fluorophore labeling with minimal linkage errors. This allowed us to perform high-quality dSTORM imaging of various targets in mammalian and yeast cells. We expect that this versatile strategy will render many more demanding cellular targets amenable to dSTORM imaging.

F luorescence-based super-resolution microscopy (SRM) is becoming increasingly applied in cell biology. Singlemolecule localization microscopy (SMLM) techniques, such as (direct) stochastic optical reconstruction microscopy ((d) STORM) provide outstanding spatial resolutions and have enabled unprecedented insights into the organization of subcellular components [1][2][3] . However, the quality and value of SMLM imaging can be limited due to poor photon emission or detection efficiency, low fluorophore labeling densities, linkage errors or steric hindrances [4][5][6] . Most current SMLM labeling approaches employ antibodies or recombinant proteins either fused to photoactivatable fluorescent proteins (FPs) or fluorogen-labeling enzymes, such as the Halo-, CLIP-, or SNAP-tag [7][8][9][10] . While conventional antibodies introduce significant linkage errors by displacing the fluorophore from the target, large protein/enzyme tags can affect expression, cellular localization, folding and/or function of the respective fusion protein [11][12][13] . Although small peptide tags, such as FLAG-, HA-, or Myc-tag [14][15][16] are available, those epitopes often have to be arranged in multiple arrays to recruit medium-affine binding antibodies 17 and thus do not provide dense labeling sufficient for high-quality SRM.
Instead of using antibodies, a 15-amino-acid peptide-tag can be visualized by high-affinity fluorescently labeled monomeric streptavidin 18 , which, however, can be affected by the binding of endogenously biotinylated proteins. Alternatively, reversibly onand off-binding labels in point accumulation for imaging of nanoscale topography (PAINT) microscopy allow for a continuous and therefore ultra-high density readout as they are not limited by a predefined fluorophore tagging pattern 19 . Yet, this approach can only be used for distinguishable structures like membranes or DNA combined with illumination-confined arrangements, such as in surface-near or lightsheet illuminations 20 . The visualization of other structures by PAINT approaches relies on a specific labeling commonly achieved by DNA-PAINT 21,22 .
As a promising substitute for conventional antibodies, smallsized nanobodies (antibody fragments derived from heavy-chainonly camelid antibodies) coupled to organic dyes were recently introduced for SRM. Nanobodies targeting native proteins, such as components of the nuclear pore complex, tubulin, or vimentin were described for dSTORM imaging [23][24][25] . Despite their capability to directly probe endogenous antigens, the de novo generation of gene-specific nanobodies and their validation for SRM imaging purposes is cumbersome and time-consuming 26,27 , which is reflected by the fact that only a very limited number of SRM-compatible nanobodies are available by now 25 . Due to their applicability for nanoscopy of widely used FP-fusions, GFP-, and RFP-nanobodies became very popular tools for SMLM 28,29 . However, this strategy relies on the correct expression of FPfusions and does not cope with problems arising from mislocalization or dysfunction 12,13,30 . Thus, nanobodies directed against short and inert tags might prove advantageous for SRM.
Here we introduce a versatile labeling and detection strategy comprised the short and inert BC2 peptide-tag (PDRKAAVSHWQQ) and a corresponding high-affinity bivalent nanobody (bivBC2-Nb) for high-quality dSTORM imaging. We demonstrate the benefits of our approach for close-grained fluorophore labeling with minimal linkage error of various ectopically introduced and endogenous targets in fixed and living cells.

Results
Development of a dSTORM suitable BC2-tag/bivBC2-Nb system. As originally described, we first labeled the BC2-Nb at accessible lysine residues by N-hydroxysuccinimide (NHS) ester fluorophores, such as Alexa Fluor 647 (AF647) 31 . While BC2-Nb AF647 (NHS) is sufficient for wide-field microscopy (Fig. 1a, left panel, Supplementary Fig. 1a, b), dSTORM imaging of BC2tagged proteins revealed a rather low-staining efficiency resulting in inferior structural labeling coverage (Fig. 1b, left panel). Thus, we analyzed the binding properties of a bivalent format of the BC2-Nb (bivBC2-Nb) (Fig. 1a, right panel). We assessed its binding kinetics by biolayer interferometry (BLI) and observed a considerably reduced dissociation rate compared to monovalent BC2-Nb ( Supplementary Fig. 1c). Notably, this decrease in dissociation rate is not caused by simultaneous binding of the bivBC2-Nb to two BC2 epitopes as confirmed by a BLI assay using a tandem-BC2-tag of two consecutively linked BC2 epitopes (BC2-BC2-tag) ( Supplementary Fig. 1d).
Nevertheless, antigen labeling using the bivBC2-Nb conjugated to NHS-ester fluorophores (bivBC2-Nb AF647 (NHS) ) did not yield the expected visual improvement of staining specificity (Fig. 1a, right panel). Considering the crystal structure 31 , we designed a site-directed, enzymatic coupling strategy, which should not affect the paratope and binding properties of bivBC2-Nb. Using the Sortase A system, we linked peptides conjugated to a single-AF647 fluorophore in a defined 1:1 ratio to the C-terminus of bivBC2-Nb (bivBC2-Nb AF647 (sort) ) 32,33 . This approach significantly improved the staining specificity by a factor of two compared to bivBC2-Nb AF647 (NHS) (Fig. 1a, right panel, Fig. 1c and Supplementary Fig. 1a, b, e). An exemplary dSTORM image of a HeLa cell transiently expressing vimentin BC2T and stained with the bivBC2-Nb AF647 (sort) illustrates the remarkable quality of the BC2-tag/bivBC2-Nb labeling strategy (Fig. 1b, right panel). For a better understanding bivBC2-Nb AF647 (sort) is referred to bivBC2-Nb AF647 in the following.
Since the BC2-Nb was originally developed against β-catenin 34 , we assessed the influence of the background staining of endogenous β-catenin on the labeling quality. To distinguish background due to general unspecific staining from additional β-catenin staining, we compared HeLa cells (not expressing any GFP epitope) stained with a GFP-targeting nanobody (GFP-Nb AF647 ) to HeLa cells stained with bivBC2-Nb AF647 . Further, we performed bivBC2-Nb AF647 staining in HeLa cells transiently expressing the non-structural, autophagosomal marker protein LC3B fused to the BC2-tag ( BC2T LC3B), which is -in the absence of autophagy-homogeneously distributed throughout the cytoplasm. By analyzing the dSTORM data using DBSCAN clustering 35 , we measured a slightly increased level of 1.7 (±0.3 S.D.) nanobodies per square micrometer for bivBC2-Nb AF647 compared to the unspecific background staining of 0.61 (±0.03 S.D.) GFP-Nb AF647 per µm 2 . However, this level is considerably lower compared to 7.2 (±1.3 S.D.) bivBC2-Nb AF647 per µm 2 which we obtained for the staining of BC2T LC3B expressing cells (Fig. 1d, Supplementary Fig. 2a). We then compared signal intensities derived from bivBC2-Nb AF647 -stained HeLa cells, which were either left untreated or incubated with CHIR99021 (CHIR) to accumulate endogenous β-catenin 34 . While immunolabeling with a β-catenin-specific antibody showed a strong enrichment in CHIR-treated cells ( Supplementary Fig. 2b), dSTORM imaging revealed only a minor increase of bivBC2-Nb AF647 localizations ( Supplementary Fig. 2c, left panel). Moreover, in CHIR-treated HeLa cells transiently expressing vimentin BC2T , the nanobody signal was almost exclusively detectable at vimentin fibers ( Supplementary Fig. 2c Fig. 2d). From this we conclude that even if present at high levels, the BC2-epitope of β-catenin has a negligible impact on staining of ectopically introduced antigens.
For a stoichiometric quantification of the labeling quality of the BC2-tag/bivBC2-Nb detection system, we utilized the Escherichia coli protein ferritin (FtnA) recently described as a homooligomeric protein standard of 24 subunits 36 . We expressed BC2-tagged, as well as SNAP-tagged FtnA-24mers in U2OS cells and performed dSTORM imaging on cell lysates immobilized on coverslips 36 . By measuring single-AF647 blinking events, we obtained the parameters of the corresponding log-normal distribution (µ = 5.68, σ = 0.4), which describes the probability distribution of single-molecule fluorescence intensities (Supplementary Fig. 3a). We then measured the fluorescence intensities of immobilized FtnA oligomers labeled with the BC2-tag/bivBC2-Nb or SNAP-tag system. We compared these distributions to the expected fluorescence intensity distributions of fully labeled FtnA-24mers, calculated from the single-molecule fluorescence intensity distribution and the degree of labeling of each component (Methods section). As a result, the BC2-tag/bivBC2-Nb FtnA-oligomer staining revealed a completeness of labeling of Nanobodies per μm 2 61.4% which competes with the covalent SNAP-tag FtnA-staining efficiency of 64.7% (Fig. 1e, Supplementary Fig. 3b) and outcompetes photoactivation/photoconversion efficiencies of fluorescent proteins 37 . Notably, our observation of a rather low efficiency of about 65% for the SNAP-tag labeling is in agreement with reported assessments 36,38 , and conference presentations by K. Yserentant (2017).
Comparison of different labeling strategies for SMLM. As genetic tagging of structural proteins like vimentin often impairs their structure and function 39,40 , we evaluated the influence of the short BC2-tag on vimentin structure formation and compared the BC2-tag/bivBC2-Nb detection system with established strategies focusing on image quality and apparent organization of the vimentin network. We performed SMLM on native vimentin in comparison to vimentin fused to photoactivatable mCherry (PAmCherry-vimentin), eGFP (GFP-vimentin), or the BC2-tag (Fig. 2a, b). For our studies, we transiently expressed the corresponding proteins for 24 h in HeLa cells followed by chemical fixation of the cells. Native vimentin was visualized with the recently described vimentin-specific nanobody bivVB6-Nb AF647 25, 41 , while PAmCherry-vimentin was mapped directly. The other constructs were labeled with the nanobodies GFP-Nb AF647 or bivBC2-Nb AF647 , respectively. Image analysis of the vimentin network visualized by the different labeling strategies revealed considerable phenotypic differences (Fig. 2b Fig. 4e and Methods section). 94% of native vimentin fibers labeled with the bivVB6-Nb AF647 showed widths lower than 150 nm. In contrast, cells with incorporated PAmCherry-vimentin were smaller and showed a high percentage (20%) of thick vimentin bundles above 150 nm width, while in GFP-vimentin expressing cells more than 96% of all detectable vimentin fibers had widths below 75 nm (Fig. 2c). Compared to the N-terminally labeled counterparts, cells expressing vimentin-PAmCherry displayed a highly similar phenotype whereas cells expressing vimentin C-terminally fused to GFP (vimentin-GFP) showed an even more severely fragmented vimentin network ( Supplementary Fig. 5). Obviously, both type and position of the FP affects the formation of the vimentin network and induce altered cellular phenotypes. The various observed morphological alterations are likely caused by several mislocalization and self-oligomerization artifacts induced by the different FP moieties derived either from jelly fish (GFP) or red corals (DsRed) 40,42 . Notably, no phenotypic changes or significant differences in the abundance of fiber widths were detected between native and BC2-tagged vimentin (96% of all fibers below 150 nm, 4% above 150 nm; Fig. 2b, c and Methods section).
We then assessed the SMLM image quality achievable by the different labeling approaches. The quality is dependent on two main factors; (i) the optical resolution dictated by the precision with which fluorescent spots can be localized, and (ii) the structural resolution determined by the labeling density (coverage) and the physical distance between fluorophore and target (linkage error). We assessed these parameters for each analyzed fiber individually. The localization precision was calculated by a Nearest Neighbor based Analysis (NeNA) 43 , the labeling density was determined by the lengthwise fluorescent signal coverage along each fiber, and the linkage error by quantifying the apparent width of fibers of the smallest fiber category. For further comparison, we calculated the Fourier Image REsolution (FIRE) values 44 (see Supplementary Note 1). Since the readout of all three nanobody labeling strategies relies on the same bright fluorophore (AF647), NeNA yielded the same optical resolution statistics with a mean NeNA localization precision of about 9-12 nm. The fluorescentprotein PAmCherry has a lower photon yield and achieves an average NeNA value of 17 nm ( Supplementary Fig. 6). The structural resolution as assessed by the different labeling coverage statistics revealed significant differences (Fig. 2c, Supplementary Fig. 6 and Methods section). For PAmCherryvimentin, we observed the lowest coverage among all labeling strategies for thin fibers, and a maximum coverage of~75% for thick fibers, which is likely due to inefficient chromophore formation and photoactivation. The low coverage of~50% for the GFP-Nb is more likely explained by a steric hindrance in incorporating GFP-tagged molecules into the native vimentin network, which is in line with our observation of only thin fibers. The highest labeling coverage was observed for bivBC2-Nb with a coverage of~80% for fibers below 75 nm width, and nearly full coverage of fibers exceeding a width of 150 nm. For thin fibers it exceeds the coverage obtained with the bivVB6-Nb for native vimentin, which might be due to a reduced accessibility of the native epitope within assembled vimentin filaments. To assess the impact of the size of the labeling probe on the structural resolution, we compared our bivBC2-Nbbased approach with conventional, monoclonal antibody staining ( Supplementary Fig. 7a). Antibody labeling resulted in nearly complete coverage of thin vimentin fibers (>75 nm) ( Supplementary Fig. 7b), and the AF647-based readout resulted in the same localization precision and optical  Supplementary Fig. 1). d Assessment of bivBC2-Nb AF647 staining of endogenous β-catenin. Bar chart summarizes measured nanobody per µm 2 values for untransfected chemically fixed HeLa cells stained with GFP-Nb AF647 or bivBC2-Nb AF647 in comparison to chemically fixed HeLa cells transiently expressing BC2T LC3B stained with bivBC2-Nb AF647 , errors given as standard deviation (S.D.), N = 3 cells for each condition (Methods section, Supplementary Fig. 2). e Quantification of completeness of labeling for the bivBC2-Nb and SNAP-tag labeling systems using FtnA-oligomers of 24 subunits. Bar chart summarizes median values of FtnA-24mer fluorescence intensities as percentage of theoretical maxima (Methods section, Supplementary Fig. 3) resolution statistics ( Supplementary Fig. 7c). Despite the highlabeling coverage, the antibody-mediated displacement of the fluorophore led to an increased linkage error. Accordingly, we measured an average width of~55 nm for thin vimentin fibers probed with the antibody, whereas a smaller apparent width of 40 nm was observed with bivBC2-Nb ( Supplementary  Fig. 7c).
Detection of various cellular targets with the bivBC2-Nb. Next we analyzed whether the BC2-tag/bivBC2-Nb detection system is transferable to other structural proteins. To test whether orientation of the BC2-tag affects the incorporation of the recombinant protein into endogenous structures, we transiently expressed cDNAs of mouse TUBA1B, human LMNB1, or ACTB either comprising the BC2-tag on the N-or the C-terminus in different cell lines followed by detection with bivBC2-Nb AF647 . As exemplarily shown for tubulin alpha-1B, C-terminal addition of the BC2-tag yielded more distinct microtubule structures compared to the N-terminally tagged version ( Supplementary Fig. 8). For lamin B1, we observed no differences regarding the tag position whereas ectopically expressed β-actin is only incorporated into the actin cytoskeleton when the BC2-tag is located at the Nterminus, which is in accordance to previously tested tagging approaches 45 . dSTORM imaging of HeLa cells expressing either lamin BC2T or BC2T actin, as well as U2OS cells transiently expressing tubulin BC2T revealed that BC2-tagged proteins are efficiently incorporated in the corresponding structures and could be imaged at high-resolution reaching localization precisions of 9-12 nm as previously shown for vimentin BC2T (Fig. 3a-c, Supplementary Fig. 9). As individual microtubules have a defined diameter of 25 nm these structures serve as an experimental benchmark for SRM 24,28,46 . Simulations on nanobody labeling of microtubules using a maximal probe displacement of 5 nm and a localization precision cutoff of 10 nm have yielded an apparent fiber width of about 40 nm 24 , which is in perfect agreement to our measured fiber width of 38.2 ± 9.2 nm ( Supplementary Fig. 9a). Moreover, a detailed analysis of individual actin fibers comprising transiently expressed BC2T actin showed comparable labeling densities as previously obtained for vimentin BC2T (Supplementary Fig. 9b-d).
Additionally, we used our approach to visualize non-structural proteins, namely the autophagosomal marker protein LC3B and the extracellular membrane marker GFP-GPI 47,48 . To monitor induction of autophagy, we co-expressed BC2T LC3B and GFP-LC3B in HeLa or A549 cells followed by incubation with DMSO or rapamycin to induce autophagosome formation. Wide-field imaging of chemically fixed cells, stained with bivBC2-Nb AF647 , showed a clear co-localization of GFP and nanobody signals at defined spots in rapamycin-treated cells, indicating correct localization of BC2-tagged LC3B at autophagosomes 47, 49  Fig. 10a). Further, dSTORM imaging and subsequent DBSCAN cluster analysis of newly formed autophagosomes in BC2T LC3B expressing cells after incubation with rapamycin revealed diameters of~0.3-1.0 µm for these foci (Fig. 3d, Supplementary Fig. 11, Methods section), which is in accordance to previous findings 50, 51 . For BC2-tagged GFP-GPI ( BC2T GFP-GPI), we observed a clear co-localization of the nanobody and the GFP signal at the plasma membrane in chemically fixed HeLa cells ( Supplementary Fig. 10b). Notably, with dSTORM, we detected a defined spatial organization of BC2T GFP-GPI, e.g., the formation of small clusters and enrichment of BC2T GFP-GPI molecules at cell-cell contacts, compared to a diffraction-limited, homogenous distribution observable by wide-field microscopy ( Fig. 3e, Supplementary Fig. 10b).
Visualization of endogenous proteins with the bivBC2-Nb. To utilize the BC2-tag as an endogenous marker under native promotor expression, we first replaced the gene coding for the nuclear DNA-binding protein cbp1 at its endogenous loci in the fission yeast Schizosaccharomyces pombe (S. pombe) by a C-terminally BC2-tagged version (cbp1 BC2T ). Cells expressing cbp1 BC2T show growth rates comparable to wild type (wt) and exhibit no morphological changes ( Supplementary Fig. 12a, b). As S. pombe possesses a thick cell wall and a highly packed cellular environment, any immunofluorescence-based SRM approach suffers from high-unspecific background and low-staining quality. Notably, by staining endogenously expressed cbp1 BC2T utilizing the bivBC2-Nb AF647 we were now able to visualize an endogenous nuclear protein in S. pombe by dSTORM imaging (Supplementary Fig. 12a). Second, we stably introduced the coding sequence of the BC2-tag under the native β-actin promotor at the 5′-end of the first exon of endogenous β-actin in HeLa and A549 cells using the CRISPR/Cas9 technology. After monoclonal selection of cells exhibiting a heterozygous integration of BC2T actin (HeLa-BC2T ACTB; A549-BC2T ACTB, Methods section) we treated both cell lines with transforming growth  Supplementary Fig. 9a), c HeLa cell expressing BC2T actin (coverage statistics in Supplementary Fig. 9b), d HeLa cells expressing BC2T LC3B either left untreated or treated with rapamycin. Bar charts represent the degree of clustering, given as a relative fraction of cluster points versus noise points, errors given as standard deviation (S.D.). Histograms represent cluster diameters as determined by DBSCAN analysis with a bin size of 100 nm (x-axis) plotted against relative fraction (y-axis). Full data are represented underneath the histograms as box + scatter plots with the same x-axis. The box marks the three quartiles and the whiskers mark 95% of all the data. Total number of clusters n = 342 in non-treated cells, n = 405 in treated cells, N = 3 cells for untreated cells and N = 4 cells for rapamycin-treated cells ( Supplementary Fig. 11), and e HeLa cells expressing BC2T GFP-GPI. All cells were stained with bivBC2-Nb AF647 (Methods section). Scale bars, images 5 µm, insets 1 µm. Crossed out rectangles mark the position of fiducial markers used for drift correction. Image reconstruction details are given in Methods section factor β (TGFβ) and monitored the induction of actin stress fibers by co-staining with bivBC2-Nb AF647 and phalloidin AF555 . As expected, we detected the formation of stress fibers only in A549 wt and A549-BC2T ACTB cells, which are described to respond to TGFβ 52 (Fig. 4a). dSTORM imaging further allowed a detailed insight into the non-disturbed actin network (Fig. 4b). These findings indicate that BC2-tagging of endogenous proteins is a viable approach for SRM studies to visualize cellular targets at endogenous levels and with minimal functional interference.
bivBC2-Nb visualizes its target proteins in living cells. To realize the advantages of the BC2-tag/bivBC2-Nb system also for live-cell applications, we first performed time-lapse imaging of HeLa cells transiently expressing BC2T GFP-GPI. After addition of bivBC2-Nb AF647 to the imaging medium, we observed a fast recruitment of the nanobody to its membrane-located antigen ( Supplementary Fig. 13a), with a saturation of the nanobody signal within 20-30 min (Supplementary Fig. 13b). Single-particle tracking dSTORM imaging further allowed us to trace the highly  Fig. 14b). Finally, we performed live-cell dSTORM imaging replacing AF647 with the fluorophore ATTO655 which is currently one of the few organic dyes exhibiting sufficient photoswitching under physiological intracellular conditions 8 . Two hours after nanobody transduction, dSTORM imaging of the bivBC2-Nb ATTO655 staining revealed the intracellular actin network of HeLa-BC2T ACTB cells (Fig. 5b) with subdiffraction details as previously seen for chemically fixed HeLa-BC2T ACTB cells (Fig. 4b). We further documented the photophysical differences in performance of the ATTO655 fluorophore, which is outcompeted by AF647 both in blinking statistics and brightness ( Supplementary Fig. 15, Supplementary Movies 5 and 6). In summary, these data demonstrate that the bivBC2-Nb is functional within living cells where it retains its outstanding binding capacities allowing for live-cell dSTORM imaging. Other recently reported protein transduction approaches offer numerous alternatives to introduce the bivBC2-Nb in various cell types for live-cell imaging of BC2-tagged proteins 41, 55-61 .

Discussion
In this study, we developed and extensively characterized a broadly applicable and transferable labeling strategy based on a structurally minimal tag in combination with the first peptidespecific nanobody suitable for SRM. In contrast to the widely established GFP/GFP-Nb system 28,62 for dSTORM imaging of fusion proteins comprising a large fluorescent moiety, the short and inert BC2-tag allows an efficient and dense incorporation of ectopically and endogenously expressed proteins into higherordered cellular structures in mammalian and yeast cells. In particular, it does not interfere with the native organization of structural proteins, such as vimentin, lamin, actin, and tubulin known to be easily compromised by large protein tags. We demonstrated that all tested BC2-tagged proteins are efficiently detected with a high affinity, bivalent nanobody format, robustly labeled in a one-to-one nanobody to fluorophore ratio exhibiting a high-labeling efficiency competing with covalent detection systems. Together, this enables high-labeling coverage with minimal linkage error and thus allows for genuine SRM studies of physiologically undisturbed cellular structures. Moreover, the bivBC2-Nb is functional in living cells where it retains its binding capacity and labels its target structures over extended time periods, which renders the BC2-tag/bivBC2-Nb labeling system into a versatile tool for SRM imaging of fixed and living cells. Methods Expression constructs. All primer sequences used for cloning are listed in a Supplementary Table 1. The expression construct coding for vimentin Nterminally fused to eGFP (GFP-vimentin) was previously described 41 . For generation of a photoactivatable (PA) mCherry (PAmCherry) fusion construct of vimentin (PAmCherry-vimentin) the coding sequence of PAmCherry was PCRamplified from the pBAD/HisB-PAmCherry1 vector using the following primer set: PAmCherry-F and PAmCherry-R. The PCR product was purified, digested with the restriction enzymes AgeI and BglII and ligated in the AgeI/BglII sites of a vector coding for mCherry-vimentin thereby replacing mCherry with PAmCherry. An expression construct coding for vimentin with a C-terminal BC2-tag (vimentin BC2T ) was generated by the replacement of the mCherry sequence from a mCherry-Vimentin-BC2T (mCherry-vimentin BC2T ) fusion construct previously described 31 . Thus, vimentin BC2T cDNA was PCR-amplified using the primer VIM-BC2T-for and ligated into NheI and BamHI restriction sites of the template construct. Constructs coding for vimentin C-terminally fused to PAmCherry (vimentin-PAmCherry) and vimentin C-terminally fused to eGFP (vimentin-GFP) were generated by Gibson assembly of the three following fragments: fragment 1-pEGFP-N1 vector backbone digested with NheI and BsrGI, fragment 2-vimentin amplified from vimentin BC2T with the primer set VIM-for and VIM-rev, fragment 3-PAmCherry amplified from PAmCherry-vimentin or eGFP amplified from GFP-vimentin using the primer set PAmCherry/eGFP-for and PAmCherry/eGFPrev. Fragments were assembled using the Gibson-Assembly Master Mix (New England Biolabs, cat. #E2611) according to the manufacturer's protocol. An expression construct coding for BC2-tagged β-actin ( BC2T actin) was generated by the combination of two PCR fragments derived from an eGFP-actin construct previously described in ref. 62 . The first PCR fragment was generated using primer BC2TActb(1)-for and BC2TActb(1)-rev. The second PCR fragment was generated using primer BC2TActb(2)-for and BC2TActb(2)-rev. Both DNA fragments were purified and ligated by compatible sticky ends generated by BssHII and SexAI restriction enzymes. To generate a BC2-tagged laminB1 ( BC2T lamin) expression construct the lamin B1 cDNA was PCR-amplified from a GFP-Lamin B1 DNA template 62 by PCR using primer BC2TLamin-for and BC2TLamin-rev and cloned into XhoI and NheI restriction sites of pEGFP-N1 vector thus introducing the BC2tag sequence. The expression construct coding for C-terminally BC2-tagged lamin (lamin BC2T ) was generated by the replacement of the vimentin cDNA of the above described vimentin BC2T by lamin cDNA PCR-amplified from GFP-Lamin B1 using the primer laminBC2T-for and laminBC2T-rev. DNA fragments were purified and ligated by compatible sticky ends generated by NheI and BssHII restriction enzymes. The expression construct coding for C-terminally BC2-tagged tubulin (tubulin BC2T ) was generated by substituting vimentin cDNA of the above described vimentin BC2T with the tubulin cDNA PCR-amplified from pPAmCherry-tubulin (addgene 31930) with primers tubulinBC2T-for and tubulinBC2T-rev using restriction enzymes NheI and BssHII.
The expression construct coding for N-terminally BC2-tagged tubulin ( BC2T tubulin) was generated by substituting actin cDNA of the above described BC2T actin with the tubulin cDNA PCR-amplified from pPAmCherry-tubulin (addgene 31930) with primers BC2Ttubulin-for and BC2Ttubulin-rev using restriction enzymes BssHII and BamHI. Mammalian expression construct coding for GFP-LC3B fusion protein, was generated by insertion of LC3B cDNA into BglII and EcoRI restriction sites of pEGFP-C1 expression vector. The expression construct coding for N-terminally BC2-tagged LC3B ( BC2T LC3B) was generated by substituting actin cDNA of the above described BC2T actin with the LC3B cDNA PCR-amplified from GFP-LC3B using primers BC2TLC3B-for and BC2TLC3B-rev using restriction enzymes BssHII and BamHI. BC2T-tagged GFP-GPI construct for mammalian expression was generated by insertion of synthetic DNA fragment with an ORF coding for signal peptide of human CD59 (aa 1-25), BC2-Tag, eGFP and amino acids 92-128 of huCD59, which contains the GPI attachment site at aa 102 into BglII and NotI restriction sites of pEGFP-N2 vector DNA. Plasmids coding for Ypet-FtnA and SNAP-FtnA were purchased from addgene (cat. #98280 and 98282). To generate the expression construct coding for BC2T Ypet-FtnA, BC2T was inserted to the N-terminus of Ypet-FtnA by PCR amplification of the complete Ypet-FtnA plasmid using BC2TYpet-for and BC2TYpet-rev. Subsequent recircularization of the amplified plasmid was performed using BssHII restriction sites.
Bacterial expression vectors coding for the GFP-nanobody (GFP-Nb), the BC2nanobody (BC2-Nb), and the bivalent BC2 nanobody (bivBC2-Nb) were provided by ChromoTek with a corresponding material transfer agreement. The bacterial expression construct of the bivalent VB6 nanobody (bivVB6-Nb) was previously described 41 . For all three constructs the original tag was replaced by a Sortase-tag (GSLPETG) upon PCR amplification of the full plasmid using the forward primer SorTag-Ins_for and SorTag-Ins_rev and subsequent recircularization using terminal AgeI restriction sites. The resulting expression constructs were confirmed by sequencing and bacterial expression followed by SDS-PAGE and immunoblot analysis using a C-terminal anti-His antibody (ThermoFisher Scientific, cat. #R930-25, dilution 1:1000). An expression vector coding for SortaseΔ59 (pET28a-SrtAdelta59) was a gift from Hidde Ploegh (Addgene plasmid #51138) 63 .
Degree of labeling (DOL, dye-to-protein ratio) was determined by absorption spectroscopy according to the instructions provided by ThermoFisher Scientific (stated for cat. #A20173). For BC2-Nb and bivBC2-Nb NHS-conjugated nanobodies DOLs of 1.8 ± 0.5 and 2.1 ± 0.7 were determined.
S. pombe strain construction. The cloning strategy for BC2-tagging of the CBP1 gene at the C-terminus was adapted from 65 . The Saccharomyces cerevisiae ADH1 terminator and kanamycin resistance gene were amplified from the PAW8 plasmid 66 using the following primer pair F_KanR_BC2 and R_KanR.~250 bp sequences up-and down-stream of the cbp1 gene were amplified from purified S. pombe DNA, with the primer pairs F1_cbp1, cbp1_BC2_R1, and F2_cbp1, R2_cbp1. Primers were designed to generate PCR products with overlapping regions of at least 20 bp. DNA fragments were assembled with overlap-extension PCR 67 , using melting temperatures of the overlapping regions as the annealing temperature. All PCRs were performed with Q5 High-Fidelity DNA polymerase (New England Biolabs, cat. #M0491L). Volume of 10 μl of the PCR product was transformed into wild-type S.pombe using the Frozen-EZ Yeast Transformation II Kit (Zymo Research, cat. #T2001), plated onto YES agar plates and incubated overnight at 30°C, then replica plated onto 200 μg ml −1 G418 (Thermo Fisher Scientific,) YES agar plates and incubated at 30°C until single colonies were visible. Genomic integration was confirmed by colony PCR and DNA sequencing (Eurofins).
Next, paired sgRNAs were cloned according to a previously described procedure into a plasmid harboring Cas9D10A and a puromycin resistance cassette 70 . Briefly, the sgRNA fragment were PCR amplified using sgRNA_fw and sgRNA_rev primer thereby adding 3′ and 5′ domains homologous to plasmid encoded hU6 promotor and gRNA scaffold sequences, respectively. The 148 bp PCR amplicon was gel-purified and ligated at a 3:1 ratio to the 415 bp fragment generated by BbsI digest of the pDonor_U6 plasmid (a gift from Andrea Ventura, Addgene plasmid #69312) 70 using the NEBuilder Cloning Kit (New England Biolabs, cat. #E5520S). After treatment with Exonuclease RecBCD (New England Biolabs, cat. # M0345L) the column purified DNA plasmid was digested over night at 37°C with BbsI. The linearized plasmid was then ligated into the BbsI-digested and dephosphorylated pSpCas9n(BB)-2A-Puro plasmid (a gift from Feng Zhang, Addgene plasmid #62987) 71 using T4 DNA ligase (NEB). Single-clone derived DNA plasmids were purified using QIAGEN Plasmid Midi Kit (Qiagen, cat. #12145) and verified by sequencing. The homology directed repair (HDR) template ACTB_HDR was synthesized as Ultramer DNA Oligonucleotides (IDT). HDR templates encoded for the HDR insert carrying the intended BC2-tag knock-in mutation flanked by left and right homology arms (each 50 bp) homologous to the ACTB target gene locus.
Generation of BC2-tag knock-in cell lines. 1 × 10 6 Hela Kyoto and A549 cells, respectively, were co-transfected at 50% confluency with 6.5 µg ACTB_HDR template oligonucleotide and 5.5 µg cloned Cas9N_Puro_ACTB_sgRNA expression vector construct or pEGFP plasmid (to control for transfection efficiency), respectively. For transfection of Hela cells Lipofectamine 2000 (ThermoFisher Scientific, cat. #11668019) and for A549 cells Lipofectamine LTX reagent (Thermo Fisher Scientific, cat. #15338100) was used. After 24 h, cells were trypsinized and re-plated in culture medium containing 1 µg ml −1 puromycin dihydrochloride (Sigma-Aldrich, cat. #P8833). Forty-eight hours later mock-transfected control cells were completely killed by the antibiotic. Puromycin-resistant cell pools were expanded for 1 week and subsequently used for detection of CRISPR/Cas9D10Ainduced BC2-tag knock-in by immunofluorescence staining using the bivBC2-Nb AF647 and genomic PCR. Monoclonal knock-in cell lines were derived from cell pools showing successful BC2-tag knock-in by limiting dilution.
Genomic PCR of BC2-tag knock-in cells. Genomic DNA was isolated from puromycin-resistant cells using the QIAamp DNA Mini Kit (Qiagen, cat. #12125) according to manufacturer's instructions. Quantity 500 ng of purified genomic DNA was used for PCR amplification of the integrated BC2-tag sequence from the ACTB target gene locus using the primer ACTB_fw and BC2_rev. PCR products were separated on 1.5% agarose gels and visualized using ethidium bromide staining.
Twenty-four hours after transfection or in case of U2OS cells expressing BC2T tubulin, tubulin BC2T , Ypet-FtnA, or BC2T Ypet-FtnA cells 72 h after transfection, cells were washed twice with PBS and fixed with 3.7% w/V paraformaldehyde (PFA) in PBS for 10 min at RT. Fixed cells were washed three times with PBS and permeabilized with a 1:1 mixture of methanol/acetone for 5 min at −20°C. After three washing steps with PBS, cells were blocked with Image-iT FX signal enhancer (Thermo Fisher Scientific, cat. #I36933) for 30 min. Subsequently cells were washed with and stored in PBS until staining. For nanobody staining, GFP-Nb AF647 , BC2-Nb AF647 (NHS) , BC2-Nb AF647 (sort) , bivBC2-Nb AF647 (NHS) , bivBC2-Nb AF647 (sort) , or bivVB6-Nb AF647 was added with a final concentration of~50 ng ml −1 in 5% BSA in TBS-T (0.05% (w/V) Tween) and incubated overnight at 4°C. Unbound nanobodies were removed by three additional washing steps with TBS-T. Images were acquired with a MetaXpress Micro XL system (Molecular Devices) and ×40 magnification.
For staining of endogenous β-catenin,~5000 HeLa Kyoto cells per well were seeded in a µclear 96-well plate (Greiner Bio One, cat. #655090). Cells were either transfected with an expression plasmid coding for vimentin BC2T or left untransfected. Twenty-four hours after transfection cells were further left untreated or continuously cultured in the presence of 10 µM CHIR99021 for 16 h. Subsequently, cells were washed twice with PBS and fixed with 3.7% w/V paraformaldehyde (PFA) in PBS for 10 min at RT. After three washing steps with PBS, cells were permeabilized and blocked with 0.1% Triton X-100 in 5% BSA in TBS-T for 30 min. For detection of endogenous β-catenin, cells were incubated with an anti-β-catenin antibody (BD Biosciences, cat. #610154, dilution 1:200) followed by detection with an Alexa Fluor 488 labeled anti-mouse-antibody (Invitrogen, cat. #A10680, dilution 1:1000).
Staining of BC2T Ypet-FtnA and SNAP-FtnA lysates. Wells of an 8-well µ-slide (Ibidi, cat. #80826) containing lysates of BC2T Ypet-FtnA and SNAP-FtnA expressing U2OS cells were blocked with 10% BSA in PBS for 30 min, then with Image-iT FX signal enhancer (ThermoFisher Scientific, cat. #I36933) for an additional 60 min. Nanobodies were diluted to~0.5 µg ml −1 in staining (PBS, 10% BSA, 0.1% (V/ V) Triton X-100 (Sigma-Aldrich, cat. #T8787)). SNAP-Surface Alexa Fluor 647 (New England Biolabs, cat. #S9136S) was diluted to 0.1 µM in the same solution. Volume of 200 µl of the staining solution was added to each well and stained for 6 h at room temperature. After staining, wells were washed three times for 15 min with PBS containing 0.1% Tween-20. The nanobody staining was post-fixed with 4% PFA and 0.25% glutaraldehyde in PBS for 5 min to make the binding permanent. In the case of SNAP, the binding is already covalent. Wells were then washed an additional three times for 15 min with PBS and imaged overlaid with 300 µl of PBS.
Protein transduction. HeLa_ BC2T ACTB, or A549_ BC2T ACTB were plated at~5000 cells per well of a µclear 96-well plates (Greiner Bio One, cat. #655090) and cultivated at standard conditions. Next day, Nbs were transduced using Pro-DeliverIN (OZ Biosciences, cat. #PI10250) according to manufacturer's protocol. Per well of a 96-well plate 0.25 µl of Pro-DeliverIN was mixed with 0.75 µg Nb and incubated for 15 min at RT. Volume of 20 µl Opti-MEM (ThermoFisher Scientific, cat. #31985062) was added to the mixture and immediately transferred to the cell culture medium in the well. After 2 h, medium was replaced by imaging medium DMEM gfp -2 (Evrogen, cat. #MC102) supplemented with 10% FCS, 2 mM L-glutamine and cells were imaged.
Live-cell staining and imaging. HeLa Kyoto transiently expressing BC2T GFP-GPI, HeLa-BC2T ACTB, or A549-BC2T ACTB cells were plated at~5000 cells per well of a µclear 96-well plate (Greiner Bio One, cat. #655090) and cultivated at standard conditions. Next day, time-lapse imaging was performed in a humidified chamber (37°C, 5% CO 2 ) of a MetaXpress Micro XL system (Molecular Devices) at ×40 magnification. For live-cell staining of BC2T GFP-GPI, culture medium was replaced without washing by live-cell visualization medium DMEM gfp -2 (Evrogen, cat. #MC102) supplemented with 10% FCS, 2 mM L-glutamine and 1 µg ml −1 bivBC2-Nb AF647 . Time-lapse imaging with 4-5 min intervals was started immediately upon medium replacement. For live-cell staining of HeLa-BC2T ACTB and A549-BC2T ACTB upon protein transduction of nanobodies, cells were washed once with and placed in DMEM gfp -2 medium 2 h after addition of transduction mix (see "protein transduction" section above) and imaged in hourly intervals.
Quantification of staining intensities. HeLa Kyoto cells were plated at~5000 cells per well of a µclear 96-well plates (Greiner Bio One, cat. #655090) and transfected with expression plasmid for mCherry-VIM BC2T . Next day, cells were fixed and stained with the same concentration (1 µg ml −1 ) of monovalent or bivalent BC2-Nbs conjugated to AF647 either by NHS conjugation or via sortase coupling. To assess staining q uality of the different nanobody formats we calculated the ratio of the staining intensity in mCherry-VIM BC2T expressing cells and in non-transfected cells (background). Staining intensities were determined using a custom-written cell identification algorithm (MetaXpress, Custom module editor). In brief, transfected cells were identified based on cell size parameters and a threshold setting for mCherry fluorescence intensity above local background. Background fluorescence was defined as the average fluorescence of the remaining image area precluding mCherry-VIM BC2T expressing cells. For statistical significance the average fluorescence intensity of a large number of transfected cells was determined (BC2-Nb AF647 (NHS) : n = 115; bivBC2-Nb AF647 (NHS) : n = 134; BC2-Nb AF647 (sort) : n = 150; bivBC2-Nb AF647 (sort) : n = 195).
dSTORM imaging and post-processing. A 1:5000 dilution of fluorescent beads (FluoSpheres 715/755, ThermoFisher Scientific, cat. #F8799) was sonicated to break up clumps of beads. Volume of~5 µl of the beads were added to the sample and allowed to settle and adhere for 15 min, to serve as fiducial markers for drift correction. Images were recorded on a customized Nikon Ti-Eclipse inverted microscope, equipped with a CFI Apochromat TIRF ×100 objective with a numerical aperture of 1.49 (Nikon) and an iXON ULTRA 888 EMCCD camera (Andor). AF647 was imaged in 100 mM MEA (Sigma-Aldrich, cat. #M6500-25G) with a glucose oxidase (Sigma-Aldrich, cat. #G2133, C100) oxygen scavenger system 73  PALM imaging and post-processing. For imaging of PAmCherry-vimentin, the sample was illuminated with an OBIS LX 561 laser (Coherent), filtered through a ZET 561/10 clean-up filter (AHD Analysentechnik, Germany) at an intensity of 800 W cm −2 and an OBIS LX 405 laser (Coherent) at intensites of~25-5000 mW cm −2 . The readout was collected through a 610/75 bandpass filter (AHF Analysentechnik, Germany). Twenty thousand imaging frames were recorded at an exposure time of 70 ms. The number of PAmCherry molecules activated each frame was kept at a steady rate by increasing 405 laser intensity until all of the PAmCherry was readout. All other microscope parameters and image postprocessing remained the same as for dSTORM imaging.
Super-resolution analysis of endogenous β-catenin staining. To evaluate the effect of endogenous β-catenin staining on dSTORM imaging of low-abundance non-structural proteins, non-transfected HeLa and BC2T LC3B expressing HeLa cells were fixed and stained with bivBC2-Nb AF647 or GFP-Nb AF647 for 48 h, imaged and post-processed as described in the "Immunofluorescence staining for dSTORM imaging" and "dSTORM imaging and post-processing" sections".
Analysis was performed on non-transfected HeLa cells stained with bivBC2-Nb AF647 or GFP-Nb AF647 , in comparison to BC2T LC3B-expressing cells stained with bivBC2-Nb AF647 . The GFP-Nb AF647 staining served as a baseline of non-specific nanobody binding. Three 15 × 15 µm ROIs for each condition were analyzed with DBSCAN (DBSCAN parameters: ε = 40 nm; MinPts = 6) 35 to identify single nanobodies. The density of nanobodies per µm 2 was calculated for each ROI and the means of each condition plotted as bar charts with S.D. as the error.
To evaluate the effect of endogenous β-catenin staining when imaging abundant structural proteins, non-treated and CHIR99021-treated non-transfected and vimentin BC2T expressing HeLa cells were stained with bivBC2-Nb AF647 and imaged as described in the "Immunofluorescence staining for dSTORM imaging" and "dSTORM imaging and post-processing" sections. Localization counts for three cells per condition were obtained with the RapidSTORM software 75 and cell areas were measured in Fiji 77 . Localizations per µm 2 were calculated and the means of each condition plotted as bar charts with S.D. as the error.
Imaging of BC2T Ypet-FtnA and SNAP-FtnA lysates. A 488 nm Sapphire laser (Coherent Inc., Santa Clara, California USA) was used to excite the Ypet and the readout was collected through a 525/50 single bandpass filter (AHF Analysentechnik AG). Otherwise, the microscope setup was the same as described in the dSTORM paragraph above. The red laser intensity was reduced to~0.4 kW cm −2 , to avoid pixel saturation when imaging stained FtnA oligomers while still allowing for the detection of single-AF647 molecule blinking events. In the case of BC2T Ypet-FtnA, fluorescent spots were identified by their Ypet signal. Since the SNAP-tagged FtnA oligomers lacked the Ypet signal, they were excited with very low 640 nm laser (>0.01 W cm −2 ) and chosen within 2 s to avoid photobleaching. Both were then imaged for 200 frames with an exposure time of 50 ms. BC2T Ypet-FtnA and SNAP-FtnA analysis. The fluorescence microscopy movies were analyzed with a custom ImageJ script in the open access software Fiji 77 . AF647 single-molecule blinking events were identified in the last 100 frames of the movies. Molecules that blinked at least twice in the last 100 frames were selected in different quadrants of the imaged area. A round ROI with a diameter of 10 pixels was drawn around each molecule and the intensity trace of the ROI for all 200 frames was extracted. The integrated intensity value was plotted over time using the software OriginPro (Origin Lab Corp.). Only blinking events with a clear jump in the integrated intensity values co-inciding with a visual blink in the movie were measured. Intensity of blinking events was measured by calculating the difference between baseline and peak intensity of the event (Supplementary Fig. 3a). A total of N = 157 blinking events were measured. The single-molecule integrated intensity values were binned with a bin size of 100 AD counts, plotted as a relative frequency histogram and fitted with the log-normal distribution function in OriginPro (Origin Lab Corp.): p e À ðln ÀμÞ 2 2σ 2 as described in ref. 78 yielding µ = 5.68 and σ = 0.4 ( Supplementary Fig. 3a). FtnAoligomer spots were analyzed in a similar manner, with the integrated intensity value of the diameter of 10 pixels ROI of the second frame used for further analysis. A total of 130 BC2T Ypet-FtnA spots and 89 SNAP-FtnA spots were analyzed. The integrated intensity data was binned with a bin size of 1000 AD counts and plotted as relative frequency histograms ( Supplementary Fig. 3b). Approximating the labeling of FtnA oligomers with AF647 molecules with a binomial distribution, with the degree of labeling (0.7 ± 0.15 in case of the bivBC2-Nb AF647 nanobody and 0.95 ± 0.05 in case of the AF647-BG) representing the p and the number of FtnA molecules in an oligomer representing the n parameter, we could calculate the probability for the occurrence of each labeling state (e.g., 24, 23, 22… AF647 per oligomer). Data sets assuming full labeling were simulated by calculating the linear combination of calibration distributions following the binomial mixture of visual oligomeric states due to the degree of labeling by recursively convolving the singlefluorophore log-normal distribution (µ = 5.68, σ = 0.4) using a MatLab (Math-Works, Natick, MA-US)-based tool, as published and described in ref. 78 . To evaluate the completeness of labeling with the bivBC2-Nb AF647 nanobody and the SNAP-tag, we then compared the medians of the simulated fully labeled scenario data sets with our measured distributions.
Image analysis of dSTORM images. A custom ImageJ script was used to measure the widths and coverages of vimentin, actin, and tubulin fibers on reconstructed non-blurred images. The segmented line tool in ImageJ was used to manually draw lines along vimentin filaments and line thickness was adjusted to fit the filament. Each image was divided into 10 × 10 µm sections and 15 random filaments were measured in each section. Line selections were straightened using the straighten tool and an intensity profile was plotted for each filament. To determine filament width, the intensity profile was fitted with a Gaussian curve and the resulting sigma value was multiplied with 2.35 to obtain the full width at half maximum (FWHM).
To determine lengthwise coverage the image was converted into a binary image and the fraction of covered area of the middle 3 pixels was calculated. To reduce measurement error, line selections were wobbled by 0.5 pixels in four directions and values from all 5 measurements were averaged. To estimate image resolution, custom written Python 2.7 (Python Software Foundation) and Fiji 77 algorithms utilizing code from the Lama software 76 and GDSC SMLM ImageJ plugin (http:// www.sussex.ac.uk/gdsc/intranet/microscopy/imagej/gdsc_plugins) were used. NeNA localization precision 43 and Fourier Image Resolution (FIRE) 44 were calculated for all analyzed regions. To calculate FIRE, localization files were split into two by alternating frames ("random split" option unchecked). The threshold was set to Fixed 1/7 and the Fourier image scale was set to a constant 16 for all analyzed regions. Other parameters were left as default (auto image scale = 2048, sampling factor = 1). The results were plotted using the software OriginPro (Origin Lab Corp.).
LC3B clustering analysis. Imaging and post-processing of BC2T LC3B cells was done as described in the "dSTORM imaging and post-processing" section. The obtained localization files were loaded into the MatLab-based software PALMsiever 79 . Density-based clustering analysis was performed with the density-based spatial clustering of applications with noise (DBSCAN) algorithm (DBSCAN parameters: ε = 50 nm; MinPts = 40) 35 . Three cells were analyzed for the untreated conditions and four cells for the rapamycin-treated conditions. To compare the degree of clustering, the ratio of non-clustered to clustered localizations was calculated for each cell. Points assigned as core and border points by DBSCAN were considered as clustered, while noise points were considered as non-clustered. Cluster size was calculated with a custom-written Python 2.7 script, measuring the average distance of cluster points from the cluster center of mass. Degree of clustering results were plotted as stacked bar charts, using S.D. as the error and the cluster sizes were plotted as histograms with a bin size of 100 nm, as well as individual data points and box + whisker plots, the ends of the box marking the 1st and 3rd quartiles, notch marking the median and whiskers encompassing 95% of the data in OriginPro (Origin Lab Corp.).
Live cell dSTORM imaging. HeLa Kyoto transiently expressing BC2T GFP-GPI and HeLa-BC2T ACTB were seeded in 8-well Ibidi µ-slides to an approximate density of 7500-15,000 cells per well, cultivated for 24 h at standard conditions then sealed in falcon tubes containing equilibrated culture medium. Upon arrival, cells were placed in an incubator for 3 h to recover and re-equilibrate. Meanwhile, 3 µl of Pro-DeliverIN was mixed with 9 µg of nanobody and incubated at RT for 15 min. After incubation, 60 µl of Opti-MEM were added to the mix and transferred into culture wells containing 300 µl of culturing medium. Cells were incubated for 2 h in the incubator, then washed two times with pre-equilibrated imaging medium DMEMgfp-2 without FCS. Volume of 300 µl of pre-equilibrated imaging medium without FCS, supplemented with the appropriate STORM buffer components were added to the wells. Well slide lids were sealed with parafilm and imaged on a custom built piezo-electric heating stage at 37°C. For imaging of BC2T GFP-GPI, cells were stained with the bivBC2-Nb AF647 and the imaging medium without FCS was supplemented with filter sterilized Tris-Hydrochloride to a final concentration of 100 mM and 10 mM MEA. The microscope setup remained as before. To record single-molecule tracks, 640 nm laser intensity was reduced to 1-2 W cm −2 and 20,000 imaging frames were recorded with an exposure time of 40 ms. For imaging of BC2T ACTB, cells were stained with the bivBC2-Nb ATTO655 and the imaging medium without FCS was supplemented with 50 µM ascorbic acid. ATTO655 was excited with 2-4 W cm −2 of 640 nm laser and 20,000 imaging frames were recorded with an exposure time of 50 ms. BC2T ACTB images were processed as described in the "dSTORM imaging and post-processing" section.
Live cell image processing and analysis. Single-BC2T GFP-GPI particles were tracked with the help of customized tracking software written in C++ and visualized by customized software written in C++, to filter for trajectories of at least 5 steps and group single-molecule localizations or trajectories by their apparent diffusion coefficient (as calculated by MSD analysis). Only trajectories with 5 or more steps were used for analysis and visualization. Single-particle trajectories were grouped into immobile (diffusion coefficient <0.02 µm 2 s −1 ) and mobile particles (diffusion coefficient >0.02 µm 2 s −1 ) and overlaid on top of a super resolution image reconstructed in Rapidstorm 3.2 75 .
General statistical analysis of vimentin, actin, and tubulin. All labeling methods were repeated at least twice and imaged on at least two different days using the same setup and imaging parameters. Five cells from each labeling method were chosen based on the quality of drift correction. Cells were divided into 10 × 10 µm quadrants and 15 filaments in each quadrant were randomly chosen for analysis by at least two different people. All measured values were plotted as individual data points and box + whisker plots, the ends of the box indicating the first and third quartiles and the notch indicating the median. The whiskers encompass 95% of all the data. To confirm that the results of the different labeling methods differ significantly, vimentin filament widths were divided into a thin (0-75 nm), medium (75-150 nm), and thick (150 nm and above) category on which we performed a χ 2test of homogeneity. The total number of measured filaments was high enough to ensure adequate test power (n-values for all conditions given in Supplementary  Fig. 4). χ 2 -values confirmed that the filament phenotypes for the labeling methods do differ at a 0.001 level of significance. To confirm that the ratio of fiber thicknesses does not significantly differ between bivVB6-Nb AF647 and bivBC2-Nb AF647 , a χ 2 homogeneity test was performed on the categories 0-150 nm width and 150 nm and above, yielding a 0.001 level of significance. To test for correlation between FIRE, NeNA and coverage values, linear regression was performed pairwise on FIRE/NeNA, FIRE/coverage, and NeNA/coverage for all labeling methods. For FIRE/NeNA pairs, the slope of the linear correlation fits for bivVB6-Nb AF647 , GFP-Nb AF647 , and bivBC2-Nb AF647 was significantly different from 0 at a 0.05 level of significance, while in the case of PAmCherry it was not. Correlating FIRE and NeNA to coverage, the slope of the linear fit was never significantly different from 0 at a 0.05 level of significance.
Data availability. The data that support the findings of this study are available from the corresponding authors upon reasonable request.