Fully genetically encoded low-molecular-weight protein tags with defined shapes for direct molecular identification by cryo-electron tomography

11 Cryo-electron tomography (cryo-ET) enables three-dimensional visualization of cells in near-12 native states, but direct identification of specific proteins in situ remains challenging due to 13 crowded cellular environments and the low intrinsic contrast of most proteins smaller than ~500 14 kDa. Consequently, molecular identification often relies on indirect labeling strategies or bulky 15 probes that can perturb native structures. Here we present a “shape-as-signal” strategy that 16 uses fully genetically encoded protein tags with defined shapes as a molecular signal for direct 17 identification by cryo-ET. We designed two single-chain, monomeric, low-molecular-weight tags: 18 an extended V-shaped tag (62 kDa) and a compact triangular tag (85 kDa). Both adopt rigid 19 geometries validated by cryo-electron microscopy and remain compatible with fluorescence 20 microscopy when fused to fluorescent proteins. Their characteristic shapes are readily 21 recognized and computationally detected in vitro. In cells, the V-shaped tag yields clear, non-22 disruptive signals at native locations. These results demonstrate that low-molecular-weight 23 protein tags can be unambiguously detected and assigned in situ within crowded cellular 24 environments. This single-step genetic tagging strategy enables seamless dual fluorescence 25 and electron microscopy without exogenous probes, challenging the assumption that small 26 protein tags are unsuitable for direct cryo-ET identification. More broadly, this approach 27 establishes a scalable and minimally perturbative framework for visual proteomics and paves 28 the way for multiplexed, shape-encoded molecular mapping in intact cells. 29 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 2 Main 30 Light microscopy (LM) and electron microscopy (EM) reveal how proteins are organized and 31 move in cells. Fluorescence microscopy (FM)—including modern super-resolution methods—32 can localize specific targets with 10-100 nm precision1–7, owing to protein and small-molecule 33 fluorophores that enable selective labeling. EM provides a complementary view at higher spatial 34 resolution. In particular, cryo-electron tomography (cryo-ET) visualizes cellular ultrastructure in 35 three dimensions (3D) under near-native, vitrified conditions, resolving membranes, cytoskeletal 36 elements, and large protein assemblies in intact cells8–14. However, cryo-ET is fundamentally 37 limited by molecular identification. Only large, structurally distinctive complexes—such as 38 ribosomes (>2.5 MDa), the 26S proteasome (~2 MDa), mitochondrial respiratory 39 supercomplexes (>1 MDa), and cytoskeletal polymers—are recognized directly in tomograms15–40 20. These complexes can be determined in situ at sub-nanometer resolution by subtomogram 41 averaging (STA) only when they are abundant in cells. However, most proteins are <70 kDa and 42 present at low abundance, making them difficult to identify effectively and precisely. As a result, 43 cryo-ET often reveals where a structure resides in the cell, but not what it is. Precisely and 44 unambiguously identifying these smaller, low-abundance proteins remains a central limitation of 45 cryo-ET. 46 Multiple strategies have attempted to overcome this barrier by attaching high-contrast or 47 physically large (>10 nm) markers to proteins of interest through affinity targeting or chemically 48 induced coupling. Recent examples include nanogold particles21, iron-loaded ferritin cages22,23, 49 DNA origami “signpost” scaffolds24, and multimeric protein tags such as genetically encoded 50 multimeric particles (GEMs)25. These methods generate visible landmarks principally but 51 accompanied with practical constraints: they often require post hoc labeling, have limited 52 efficiency in cells, and can generate false positives through off-target binding, therefore limiting 53 their general use. Cryogenic super-resolution fluorescence imaging improves localization of 54 tagged proteins beyond the diffraction limit26–34, but correlation with cryo-ET is still only precise 55 to tens of nanometers due to high fluorescence background and alignment error at cryogenic 56 temperatures—typically not sufficient to assign identity to individual molecules26,27,35,36. 57 Here, we addressed this molecular identification limitation with a “shape-as-signal” strategy: 58 developing a new class of genetically encoded, shape-defined protein tags that are directly 59 visible by cryo-ET. Rather than attaching heavy metal particles or bulky scaffolds, we 60 engineered low-molecular-weight, single-chain proteins that fold into rigid, geometrically 61 distinctive 3D shapes intended to be recognizable by morphology (size and shape) alone. We 62 engineered two single-chain, monomeric, shape-defined tags—a V-shaped protein (62 kDa) and 63 a triangular protein (85 kDa)—whose rigid architectures were verified by single-particle cryo-EM. 64 Using ferritin as a visibility benchmark in vitro, both tags produced clear densities in 3D cryo-65 tomograms, and STA resolved ferritin cages and tags. Inside the Escherichia coli (E. coli), 66 extended V-shaped protein architectures are inherently easier to distinguish around target 67 assemblies, whereas more compact triangular designs, although still detectable, are more likely 68 to be confused with surrounding punctate densities. In HeLa cells, fusion of both tags 69 respectively to TOM70NTD targeted them to the mitochondrial outer membrane without 70 detectable trafficking or morphological defects; cryo-ET revealed a characteristic V-shaped 71 density for enabling unambiguous, molecular-resolution (<2 nm) localization using standard 200 72 keV cryo-EM, while triangular tag signals were subtler but size-consistent. Fusion of GFP to the 73 V-shaped tag provided dual optical and ultrastructural visibility, enabling broadly applicable 74 correlative light and electron microscopy (CLEM) without the need for additional physical 75 correlation steps. Together, these results establish a fully genetically encoded strategy for direct 76 protein identification in cryo-ET and point toward a modular toolbox of shape-defined tags. 77 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 3 This strategy establishes a fundamentally new route for efficient, direct in situ protein labeling 78 and opens opportunities to tackle an expanding set of scientific questions that demand both 79 precise molecular mechanisms and intact ultrastructural context—critical areas that have long 80 lacked practical solutions. It enables, for example, mapping the spatial arrangement of adaptor 81 proteins or even individual protein isoforms at nascent membrane structures such as vesicles, 82 distinguishing closely spaced paralogs within a membrane complex, and tracking assembly 83 intermediates inside cells. Because the tags are fully genetically encoded, they can be 84 seamlessly combined with standard molecular perturbations (mutants, truncations, rescue 85 constructs), enabling coordinated functional and structural analyses and pushing visual 86 proteomics toward single-molecule-level imaging in the native cellular context. 87 Design and in vitro validation of V- and Δ -shaped tags 88 To generate EM-visible protein tags with special geometries, we engineered an extended V-89 shaped tag (Fig. 1a-d) and a compact triangular tag (Fig. 1e-h). 90 For the V-shape, we built on a three-helix bundle scaffold37 and introduced rigid turn inspired by 91 sterile α motif (SAM)38 to stabilize the angular junction between two bundles. Two bundles were 92 connected by a rigid α -helical linker to form an extended V structure with a defined angle 93 (Extended Data Fig. 1a). AlphaFold239,40 predicted four V-shaped designs with ~12 nm arms and 94 inter-arm angles of approximately 60°, 72°, 90°, and 140° (Extended Data Fig. 1b). To preserve 95 structural rigidity while preventing undesired interactions, we neutralized SAM oligomerization 96 residues and turned surface hydrophilicity (Extended Data Fig. 2a). For the triangular tag, we 97 adopted a C3-symmetric oligomeric motif41 as a structural template to generate an equilateral 98 triangle with ~6 nm sides (Fig. 1e). Surface residues were optimized for hydrophilicity to 99 maintain solubility (Extended Data Fig. 2b). Thus, the V-shaped protein (~62 kDa) forms 100 extended 12 nm structures, whereas the triangular protein (~85 kDa) adopts a compact ~6 nm 101 triangle (Figs. 1a, e and Extended Data Fig. 1b). 102 We expressed and purified all five designed constructs for single-particle cryo-EM analysis 103 (SPA) to assess whether they folded as intended (Extended Data Fig. 3). Synthetic genes were 104 cloned and expressed in Escherichia coli (E. coli); proteins were purified by Ni-NTA immobilized 105 metal affinity chromatography and analyzed by size-exclusion chromatography (SEC) to 106 determine oligomeric state. Raw 200-keV cryo-EM micrographs showed well-defined particles 107 for the 72° V-variant and the triangular construct; individual V- and triangular shapes were 108 directly visible despite their low mass (Fig. 1b, f). As observed by EM and consistent with SEC, 109 both proteins behaved as monomer with no detectable oligomerization (Fig. 1b, f). The other 110 three V variants (60°, 90°, 140°) did not fold into the intended architectures. We designate the 111 12-nm 72° V-variant as V12 and the 6-nm triangular construct as Delta6 (Δ 6). SPA 112 reconstructions closely matched the designed models (Fig. 1c, d, g, h and Extended Data Fig. 113 4), confirming that both tags fold as intended and demonstrating our design strategy in which 114 low-molecular-weight proteins are engineered to adopt defined geometries that enhance EM 115 visibility. As is common for purified proteins, the samples exhibited preferred orientation on EM 116 grids, with both V12 and Δ 6 appearing predominantly in a “top” view (Fig. 1b, c, f, g). 117 Tagging apoferritin cages in vitro 118 As a visibility benchmark, we fused V12 or Δ 6 tag to E. coli ferritin (FtnA), which naturally 119 assembles into a ~12-nm nanocage composed of 24 subunits42. To minimize potential steric 120 stress, we designed constructs containing two FtnA copies fused to either V12 (Extended Data 121 Fig. 5a) or Δ 6 (Extended Data Fig. 5c), such that a fully assembled cage could carry up to 12 122 tags. We expected peripheral tag densities surrounding the apoferritin cage both in vitro and in 123 situ (Extended Data Fig. 5b, d). This design also enabled a direct test of whether tagging 124 perturbs apoferritin cage assembly. 125 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 4 Tomograms of purified V12-ferritin revealed spherical cages with additional peripheral densities 126 attributable to V12 (Fig. 2a). To further validate tagging, we evaluated several automated 127 particle-picking pipelines43–48 and performed STA46,47,49. Notably, current cryo-ET picking and 128 STA workflows are largely developed and optimized for large complexes (typically >500 kDa), 129 which limits performance on low-molecular-weight features. Even so, using the deep-learning 130 based program crYOLO43,44 and the template matching and correlation-based package 131 PyTom45,50 (Extended Data Fig. 6a, b), particle picking followed by STA yielded independent 132 averages for the apoferritin cage and V12 (Fig. 2b, c). The apoferritin cage was readily 133 reconstructed to high resolution (5 Å), consistent with its size (~12-nm outer diameter; ~8-nm 134 cavity; ~465 kDa) and high symmetry (octahedral symmetry, 432 point group). By contrast, only 135 16.3% of V12 picks contributed to a low-resolution average (Fig. 2c, d), underscoring a known 136

existing cryo-ET picking/STA pipelines, tuned for larger assemblies, struggle with low-137 molecular-weight targets like V12 due to low signal to noise ratio (SNR) and orientation 138 ambiguity, as well as the inherent missing-wedge in cryo-ET43–45,49,51. 139 In 2D tomographic slices, only views aligned near the V apex display a clear V (Fig. 2a, c); 140 whereas most other orientations appear as two dots or a short line (Fig. 2a, c). In 3D, however, 141 the V shape is evident: the averaged apoferritin cage and V12 volumes fit unambiguously into 142 the tomographic densities, producing a coherent structural model (Fig. 2e and Supplementary 143 Video 1), that confirms intact cage assembly and direct detectability of V12. Slice-wise densities 144 agree with 2D projections of the fitted model (Fig. 2e, f), demonstrating that nearly the entire 145 tags are visualized across orientations—further clarifying why existing particle picking and STA 146 algorithms struggle with V12 despite its clear visibility in tomograms. 147 For Δ 6-ferritin, tomograms likewise showed peripheral tag densities (Fig. 2g). Automated 148 picking (crYOLO, PyTom; Extended Data Fig. 6c, d) and STA yielded independent averages for 149 the cage and Δ 6 (Fig. 2h, l), with a usable-particle fraction of 71.5% for Δ 6 and 43.0% for the 150 cage (Fig. 2j), underscoring the compact tag’s strong in vitro performance. Relative to V12-151 ferritin, the lower cage fraction of the cage in Δ 6–ferritin datasets suggests that the compact Δ 6 152 density may influence apoferritin cage picking. The averages recapitulated the expected 153 geometries and fit perfectly into tomographic densities (Fig. 2k, l and Supplementary Video 2). 154 Slice views revealed triangular densities in top views and one or two discrete spots in side views, 155 consistent with Δ 6 orientation (Fig. 2i, l). 156 Tagging apoferritin cages in E. coli 157 We next examined whether the V12 and Δ 6 tags were detectable in situ. V12- and Δ 6-tagged 158 ferritin were expressed in E. coli, and 80–250-nm thick lamellae were prepared by a cryogenic 159 focused ion-beam scanning electron microscope (cryo-FIB-SEM) (Extended Data Fig. 7a, b). 160 For V12-tagged ferritin, tomograms reconstructed with missing-wedge compensation and 161 denoising using IsoNet52 revealed membranes, ribosomes, and numerous ~12-nm nanocages 162 (Fig. 3a, b). Template-based particle picking using PyTom (Extended Data Fig. 7c) followed by 163 STA identified apoferritin cages in situ (Fig. 3d). Only a small fraction of particles contributed to 164 the final average (Fig. 3e), underscoring the difficulty of detecting small features in crowded 165 tomograms. Notably, close inspection of individual cages revealed extended densities 166 consistent with the expected V-shaped geometry despite the tag’s modest mass (62 kDa). As 167 anticipated, existing algorithms did not reliably detect or reconstruct the low-molecular-weight 168 V12 tag in this context. Nevertheless, manual inspection consistently revealed V-shaped 169 densities adjacent to nanocages—matching the in vitro structures and demonstrating direct 170 recognition of V12 in situ (Fig. 3c, f, g and Supplementary Video 3). 171 For Δ 6-tagged ferritin, high quality of tomograms of cryo-FIB-milled E. coli likewise revealed 172 nanocages (Fig. 3h, i). Around the cages, ~5–6 nm dot-like densities were frequently observed 173 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 5 (Fig. 3j-l and Supplementary Video 4), consistent with the compact triangular geometry of Δ 6 174 and matching the in vitro structures (Fig. 2g-l). However, because similar punctate features are 175 abundant throughout the cytoplasm, individual Δ 6 tags, while detectable, were more prone to 176 misidentification with surrounding densities. 177 Together, these results indicate that both tags could be detected in the crowded bacterial 178 cytoplasm, with the extended V12 tag providing a more distinctive and recognizable shape cue 179 than the compact Δ 6 tag. 180 Display on the mitochondrial surface in HeLa cells 181 Having confirmed the visibility of both tags in bacteria, we next tested their labeling performance 182 on the mitochondrial surface in mammalian HeLa cells. To target a native membrane, V12 or Δ 6 183 was fused to the N-terminal targeting fragment of (TOM70NTD)25,53 and appended GFP for 184 fluorescence readout (Fig. 4a, b). Western blotting with anti-GFP confirmed robust expression of 185 tagged constructs (Fig. 4c), and GFP fluorescence colocalized with the mitochondrial marker 186 Hsp60 (Fig. 4d, m). Consistent results from anti-HA antibody and Mito-Tracker Red staining in 187 HeLa cells, together with western blotting in HEK293T cells, confirmed proper expression and 188 mitochondrial localization for both tags without detectable interference (Extended Data Fig. 8). 189 HeLa cells were transiently transfected; and GFP-positive cells were isolated by fluorescence-190 activated cell sorting (FACS), allowed to attach onto EM grids, and plunge-frozen for cryo-FIB 191 milling (Extended Data Fig. 9). Cryo-fluorescence imaging of the resulting lamellae guided cryo-192 ET data acquisition and tracking of the tag (Fig. 4e, Extended Data Fig. 9). We reconstructed 193 high-quality tomograms; after missing-wedge compensation and denoising with IsoNet52, V12-194 expressing cells showed well-resolved mitochondria, ribosomes, and vesicles (Fig. 4f, g). Cryo-195 fluorescence correlated with the 3D tomograms, revealing the signal on the mitochondrial 196 surface (Fig. 4e). On the mitochondrial outer membrane, extended densities with the 197 characteristic V-shaped geometry were clearly visible and annotatable, enabling precise 3D 198 mapping of tag distribution (Fig. 4g, h-l and Supplementary Video 5). 199 In mito-Δ 6-GFP expressing cells, ~6 nm dot-like densities were observed on the mitochondrial 200 surface and colocalized with fluorescence, with orientation-dependent appearances consistent 201 with a compact triangular tag (Fig. 4n-u). However, these features were less distinct than those 202 of V12 and difficult to assign unambiguously without reference. No V-shaped densities were 203 detected in mito-Δ 6 tomograms, further underscoring the uniquely identifiable morphology of the 204 V12 tag. 205 Together, these results demonstrate that the V12 tags produce clear, detectable densities on 206 mitochondrial surface in mammalian cells, correlates well with the GFP fluorescence signal. In 207 contrast, the smaller and more compact Δ 6 tag is challenging to resolve in situ without 208 supporting experiments or subtomogram averaging results, consistent with the relative 209 detectability observed in bacterial cells. 210

211 We introduce a fully genetically encoded, shape-defined tagging strategy based on a “shape-as-212 signal” principle, enabling direct identification of specific proteins in cryo-electron tomograms 213 without post hoc labeling or chemical targeting. By encoding geometry rather than contrast, 214 these low-molecular-weight, single-chain, monomeric tags form rigid, distinctive densities that 215 are recognizable by eye at the electron microscope and amenable to computational validation. 216 This creates a direct link between molecular identity and ultrastructural context—an essential 217 step toward molecular-resolution maps of macromolecular organization in intact cells and 218 toward routine in situ counting and positioning of individual proteins. 219 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 6 Compared with existing approaches such as nanogold labeling21, DNA-origami scaffolds24, or 220 multimeric particles like GEMs25, our tags are fully genetically encoded, small enough to 221 minimize perturbation of trafficking or localization, and engineered to fold into unambiguous 3D 222 geometries. Their visibility arises not from contrast enhancement but from distinctive shape, 223 analogous to how cytoskeletal filaments or membrane structures can be recognized in 224 tomograms by morphology (size and shape) alone. Tagging ferritin or the mitochondrial outer 225 membrane with either tag did not introduce detectable defects in protein assembly, trafficking, or 226 morphology (Figs. 2-4). Notably, V12 is clearly visible on a standard 200-keV cryo-TEM 227 (Glacios) in purified samples, and mammalian cells (Figs. 1, 4)—particularly in 3D tomograms—228 broadening accessibility and underscoring its potential for widespread application. 229 The two prototype designs illustrate a tunable design space. The extended V12 tag produces a 230 characteristic V-shaped density that is readily detectable in situ on the mitochondrial outer 231 membrane and in the cytoplasm of bacteria. The more compact Δ 6 tag, although less visually 232 striking in cells, is robustly identifiable in vitro. Together, these results suggest that tag geometry 233 can be tailored to experimental needs—for example, maximizing detectability in crowded 234 cytoplasm, minimizing footprint on a sensitive target protein, or introducing asymmetry so that 235 the tagged terminus (N- or C-terminal) can be unambiguously assigned. 236 Beyond manual annotation, these tags have the potential to support automated analysis. In 237 vitro, tagged complexes could be detected by both template matching45,50 and deep-learning 238 based particle picking43, demonstrating feasibility for computational identification (Figs. 2, 3, and 239 Extended Data Figs. 6, 7). Extending these approaches in situ should enable automated 240 recognition of specific tagged molecules directly in cells. In particular, developing 3D (not merely 241 2D) detection algorithms specialized for V-shaped densities would improve recall and precision 242 for low-molecular-weight features and accelerate both particle picking and subtomogram 243 averaging, enabling automated detection and statistical analysis without requiring subtomogram 244 averaging. 245 In cells, V12 could be directly recognized in tomograms and correlated with fluorescence signals 246 from fusion to a fluorescence protein (e.g., GFP), allowing precise 3D mapping of its distribution 247 on mitochondria. This ability to annotate the tagged protein’s position within its native 248 ultrastructural environment creates a route to follow how localization changes across conditions 249 such as signaling states, metabolic stress, or disease-associated mutations. More broadly, this 250 bridges the LM-EM resolution gap: light microscopy provides temporal context and molecular 251 specificity, while cryo-ET supplies molecular-resolution ultrastructure in the same cell with the 252 exact same V12-FP fusion tag, without relying on CLEM post hoc physical correlation. 253 Looking forward, the protein-origami design framework is inherently extensible. Engineering 254 additional tags with distinct, non-overlapping geometries would enable multiplexed labeling of 255 different proteins in the same cell, allowing simultaneous mapping of multiple targets in 3D. In 256 parallel, incorporation of heavy-atom clusters or tailored mass distributions could further 257 improve detectability and support automated in situ identification. 258 This work is primarily a proof-of-concept demonstration of shape-defined, genetically encoded 259 EM tags, and we do not yet use the approach to derive new biological insights. Our experiments 260 establish feasibility in selected test systems, but each future application will require empirical 261 optimization of tag placement, linker design, and expression levels, as well as functional 262 controls to verify that the fusion does not perturb the behavior of the protein of interest—263 analogous to the validation routinely performed for fluorescent protein fusions. In addition, V-264 tags are currently identified mainly by visual inspection in cells and by simple template matching 265 or deep-learning based particle picking in vitro. Nonetheless, our data show that V-shaped 266 densities are readily identifiable in 3D volumes in situ, suggesting that robust automated 267 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 7 detection is likely achievable. To fully realize large-scale, quantitative “visual proteomics,” 268 dedicated 3D detection algorithms tailored to V-shaped densities will need to be developed and 269 integrated into tomogram analysis pipelines. 270 In summary, these shape-specific, genetically encoded EM tags provide proof-of-principle for a 271 general strategy to assign molecular identity directly in cryo-electron tomograms, practically 272 bridging fluorescence imaging (temporal control, live-cell specificity, etc.) and cryo-ET 273 (molecular-resolution ultrastructure), enabling direct tracking of protein localization under 274 physiological and disease-relevant conditions. This approach opens a route to quantitative, 275 context-aware maps of protein localization, organization, and interaction networks inside intact 276 cells, laying the groundwork for truly integrative, in situ structural and functional proteomics. 277 278

& Correspondence 279 Supplementary Information is available for this paper. 280 Correspondence and requests for materials should be addressed to Qiangjun Zhou 281 ([email protected]). 282 Peer review information includes the names of reviewers who agree to be cited and is 283 completed by Nature staff during proofing. 284 Reprints and permissions information is available at www.nature.com/reprints. 285 286 Data and code availability 287 All cryo-EM/cryo-ET data will be deposited in EMPIAR (accession to be provided upon 288 acceptance). The density maps and structure coordinates have been deposited in the EMDB 289 and PDB under accession numbers EMD-73933 and 9Z9D (V12 tag) and EMD-73947 and 9Z9I 290 (Δ 6 tag). The original and/or analyzed data sets generated during the current study are 291 available from the corresponding author upon reasonable request. 292 This paper does not report original code. 293 Any additional information required to reanalyze the data reported in this paper is available from 294 the lead contact upon request. 295 296

297 We are grateful to Drs. David Miller, Ege Kavalali, Lisa Monteggia, Borden Lacy, Hassane 298 Mchaourab, Ian Macara (VU), Eric Skaar (VUMC), Z. Hong Zhou (UCLA) and Stella Sun (Pitt) 299 for insightful discussions. We also thank Drs. Yun-Tao Liu and Hongcheng Fan (UCLA) for their 300 support with IsoNet processing. EM data collection was performed at the Center for Structural 301 Biology Cryo-EM Facility at Vanderbilt University. We acknowledge use of the Glacios cryo-302 TEM, which was acquired under NIH award S10 OD030292. Flow cytometry experiments were 303 carried out in the VMC Flow Cytometry Shared Resource, which is supported by the Vanderbilt 304 Ingram Cancer Center (P30 CA68485) and the Vanderbilt Digestive Disease Research Center 305 (DK058404). Cryo-FIB milling was conducted at the Vanderbilt Institute of Nanoscale Science 306 and Engineering with technical support from Dr. James McBride. Cryo-CLEM and cell imaging 307 studies were performed in part through the Vanderbilt Cell Imaging Shared Resource, supported 308 by NIH grants CA68485, DK20593, DK58404, DK59637, and EY08126. This work was 309 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 8 supported by CDB Destination Postdoc Award to F.L., and grant from the National Institute of 310 Health (R01MH132918 to Q.Z.). 311 Author contributions 312 Conceptualization: F.L., Q.Z.; Methodology: F.L., R.S., O.C., P.L., Q.Z.; Investigation: F.L., R.S., 313 P.L., Q.Z.; Visualization: F.L., Q.Z.; Funding acquisition: Q.Z.; Project administration: Q.Z.; 314 Supervision: Q.Z.; Writing – original draft: F.L., Q.Z.; Writing – review & editing: F.L., R.S., O.C., 315 P.L., Q.Z. 316 317 DECLARATION OF INTERESTS 318 Authors declare that they have no competing interests. 319 320

321 Protein Design and Computational Modeling of V- and Δ -shaped Tags 322 All shaped tags were designed as single-chain proteins with rigid, predefined geometries. We 323 used an iterative, AlphaFold2-guided protein-engineering workflow (“protein nanoblocks”/Lego 324 strategy): initial designs were modeled in AlphaFold240,54, inspected for geometry and 325 confidence, and refined through successive design–prediction cycles (Extended Data Fig. 3). All 326 surface residues were tuned for hydrophilicity. Electrostatic surface potentials were calculated in 327 PyMOL (APBS plugin) 55 to verify balanced charge distribution across exposed surfaces and to 328 reduce the risk of nonspecific interactions or oligomerization. 329 For the V-shaped protein, the V scaffold was derived from a three-helix-bundle (PDB: 4TQL) 330 with the two bundles connected by a rigid turn inspired by sterile α -motif (SAM) domains38,56,57 331 and a de novo-designed mini-protein motif58. AlphaFold2 predicted four candidates with inter-332 arm angles of ~60°, 72°, 90°, and 140° (Extended Data Fig. 1). To maintain solubility and 333 prevent oligomerization or undesired interactions, SAM-interface residues were neutralized. 334 For the Δ -shaped protein, we used the same design strategy, Δ 6 was built from a C3-symmetric 335 trimeric scaffold (C3triangle120_C3_A) to form an equilateral triangular assembly (~6 nm per 336 side)41. Two short linkers were engineered to concatenate three repeats into a single chain, 337 preserving the C3 geometry. 338 Protein Expression and Purification 339 For V12 and Δ 6 proteins, codon-optimized genes encoding V12 and Δ 6 were cloned into 340 pET27b vectors with N-terminal His6 tags for expression in E. coli BL21(DE3) (NEB). Cultures 341 were grown in LB at 37°C to OD600 ≈ 0.6, induced with 0.1 mM isopropyl-β -D-thiogalactoside 342 (IPTG), and incubated for 12 h at 20°C. Cells were pelleted and resuspended in lysis buffer (20 343 mM Tris-HCl pH 8.0, 300 mM NaCl, 10 mM imidazole) supplemented with a protease inhibitor 344 cocktail tablet (Roche). After sonication and centrifugation (18,000 × g, 60 min) at 4°C, 345 supernatants were purified by Ni–NTA affinity chromatography (Ni-NTA Agarose, Qiagen), 346 anion-exchange chromatography (Resource Q, Cytiva), and size-exclusion chromatography 347 (Superdex 200 Increase 10/300 GL, Cytiva) in 20 mM Tris-HCl pH 8.0, 300 mM NaCl. Protein 348 fractions were verified by SDS-PAGE and concentrated to ~0.5 mg/mL for cryo-EM. 349 For V12-ferritin and Δ 6-ferritin nanocages, the E. coli ferritin (ftnA) gene was fused at its N 350 terminus to either V12 or Δ 6 via a flexible linker and were cloned into pJ414 vectors with N-351 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 9 terminal His6 tags. Cultures were grown and induced with as above but harvested after 4 h at 352 20°C. A portion of each culture (1 mL) was used directly for plunging freezing and cryo-FIB 353 milling. The remaining cells were pelleted, resuspended in lysis buffer (20 mM Tris-HCl pH 7.4, 354 300 mM NaCl, 10 mM imidazole), supplemented with a protease inhibitor cocktail tablet (Roche) 355 at 4°C. The cells were lysed by sonication, and clarified by centrifugation (18,000 × g, 60 min) at 356 4°C. Purification followed the same chromatography workflow as above with the buffer at pH 357 7.4. Purified samples were verified by SDS-PAGE and concentrated to ~0.5 mg/mL for cryo-EM. 358 Single particle cryo-electron microscopy 359 Purified V12 and Δ 6 proteins were applied to glow-discharged Quantifoil R1.2/1.3 Cu 300-mesh 360 grids and vitrified using a Vitrobot Mark III (FEI) (95% humidity, 4°C, blot time 3 s). Data were 361 acquired on a 200-keV Thermo Fisher Glacios TEM equipped with a Falcon 4 direct detector at 362 120,000× magnification (pixel size 0.73 Å) with a total dose of 60 e⁻ /Å2 in EER format. Beam 363 induced motion-correction and dose-weighting to compensate for radiation damage over spatial 364 frequencies were performed using Patch Motion correction and Contrast Transfer Function 365 (CTF) estimation were performed in cryoSPARC59. Particle picking, two-dimensional (2D) 366 classification, and 3D refinement produced final reconstructions, reached overall resolutions of 367 5.7 Å for V12 and 6.8 Å for Δ 6 by gold-standard Fourier shell correlation (FSC) at the 0.143 368 criterion. Both datasets were processed without applying symmetry (C1), allowing unbiased 369 reconstruction of the full asymmetric architectures of the tags. 370 Mammalian cell culture, transfection, and FASC 371 HeLa (ATCC, no. CCL-2) and HEK293T (ATCC, no. CRL-3216) were cultured in DMEM (Gibco, 372 no. 31053028) supplemented with 10% (v/v) fetal bovine serum (FBS, Gibco, no. A5669701), 373 and 1% MEM nonessential amino acids (Gibco, no. 11140-050) at 37°C with 5% CO2. 374 For mitochondrial targeting, TOM70NTD-V12 and TOM70NTD-Δ 6 constructs tagged with GFP or 375 HA were cloned into pFUGW backbone under the UBC promoter. TOM70NTD corresponds to 376 residues 1-59 of human TOM70 protein, which mediates outer mitochondrial membrane 377 localization. 378 Cells were seeded into 10 cm dishes one day before transfection. At ~70% confluency, 379 transfections were performed using FuGENE 6 (Promega, no. F6-1000) with 5µg of plasmids 380 DNA and Opti-MEM (Gibco, no. 31985062) following the manufacturer’s protocol. Two days 381 post-transfection, GFP-positive cells were sorted by flow cytometry using a BD FACS Aria III. 382 Parallel transfections were carried out in 6-well or 24-well plates for immunoblotting and 383 immunofluorescence assays. 384 Cryo-ET sample preparation 385 For E. coli expressing V12-ferritin and Δ 6-ferritin, E. coli cultures (1mL) expressing V12-ferritin 386 or Δ 6-ferritin (described above) were centrifuged at 2500 × g for 5min, washed once with PBS 387 (pH 7.4) and resuspended into ~60 µL PBS. Cell suspensions were applied to glow-discharged 388 Quantifoil R2/2 Cu 200-mesh grids and plunge-frozen using a Vitrobot Mark III (FEI) at 95% 389 humidity and 24°C with a 3s blot time. 390 For HeLa cell preparation, Gold Quantifoil R2/2 SiO2 film grids were UV-sterilized for 30min per 391 side and coated with sterilized 0.05 mg/mL poly-L-lysine (PLL, Sigma-Aldrich, no. P2636-392 100MG) in 0.1M borate buffer (pH 8.5; Boric Acid, Sigma-Aldrich, no. B-0252; Borax, Sigma-393 Aldrich, no. B-9876) overnight at room temperature. Grids were rinsed 3 times with ddH2O and 394 equilibrated in culture medium. 395 After cell sorting, GFP-positive cells were pelleted with 200 × g for 5 min and resuspended in 396 medium containing 4 µM AraC (to prevent division) and HEPES and seeded onto 3-well dishes 397 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 10 (Culture-Insert 3 Well in 35 mm µ-Dish, ibidi, no. 80366 ) with the amount of ~1× 104 cells per 70 398 µL with 2 grids each well. Six hours after attaching, grids cultured with GFP-positive HeLa cells 399 were plunge-frozen in pre-warmed PBS using Leica EM GP2 with one side blotting at 37°C, 95% 400 humidity, 3 s blotting time. 401 Immunoblotting and immunofluorescence 402 Cells were lysed in RIPA buffer (25 mM Tris pH 7.6, 150 mM NaCl, 1% NP-40; Sigma, no. 403 R0278) supplemented with protease inhibitors. Lysates were separated by SDS-PAGE using 404 4%-20% Mini-PROTEIN TGX Precast Protein Gels (Bio-RAD, no. 4561094) and transferred to 405 PVDF membranes. Immunoblotted was performed with anti-GFP (Roche, no. 11814460001, 406 1:1,000) or anti-HA (Invitrogen, no. 26183, 1:5,000) primary antibodies, and GAPDH (Cell 407 signaling, no. 2118S, 1: 1,000) served as a loading control. IRDye secondary antibodies (LI-408 DOR) were used for detection, and signals were imaged with an Odyssey DLx system (LI-COR). 409 For immunofluorescence, cells were fixed with 4% paraformaldehyde (PFA), permeabilized with 410 0.1% Triton X-100, and stained with anti-HA (Invitrogen, no. 26183, 1:500; magenta), anti-411 Hsp60 (Cell signaling, no. 12165S, 1:200), MitoTracker Red CMXRos (Invitrogen, no. M46752), 412 and DAPI (blue). Images were acquired using a Nikon CSU-W1 SoRa confocal microscope and 413 Nikon SIM system. Colocalization with mitochondria was quantified in FIJI60 using Pearson’s 414 correlation coefficient. 415 Cryo-FIB lamella preparation 416 Cryo-focused ion beam (cryo-FIB) milling was performed using an FEI Helios NanoLab G3 CX 417 with a Quorum PP3010T cryo-SEM system at liquid nitrogen temperature. Prior to milling, 418 metallic platinum was deposited by sputter coating (10 mA, 20 s), followed by a protective layer 419 of organometallic platinum applied via the gas injection system (6 mm working distance, 25° 420 stage tilting angle and 8s injection). 421 Two notches were first created ~1 μ m away from the lamella to relieve mechanical stress and 422 prevent warping or bending during subsequent thinning and transfer. Cells were then milled to 423 ~1 μ m thickness at a 20° stage tilt using ion beam currents of 0.43 nA and 0.23 nA at 30 keV. 424 The stage was then tilted to 16º, and lamellae were thinned to a target thickness of 400–500 nm 425 using beam currents of 80 pA and 40 pA. Finally polishing was performed at 16° with cross-426 cleaning at 23 pA to achieve a final thickness of 100-250 nm. Before unloading, SEM overview 427 image of all lamellae and the corresponding grid was acquired to provide localization references 428 for subsequent cryo-CLEM. Finally, lamellae were sputter-coated with platinum (3 mA, 2 s) to 429 minimize charging and beam-induced drift during cryo-ET imaging. 430 Cryo-correlative light and electron microscopy (Cryo-CLEM) 431 Cryo-FIB-milled lamellae of HeLa cells expressing TOM70NTD-V12-GFP or TOM70NTD-Δ 6-GFP 432 were imaged using Leica STELLARIS Cryo-confocal microscope. FIB-milled grids were 433 transferred with a Leica EM VCM under fresh liquid nitrogen to limit ice containment. 434 Lamellae were first located in widefield mode based on overview SEM reference images. 435 Subsequently, z-stacks encompassing the entire lamellae and adjacent notches were acquired 436 in Lighting mode using 491 nm and 587 nm lasers to capture GFP fluorescence and 437 autofluorescence, respectively, for later correlation with TEM search maps. Z-stacks were 438 processed to generate sum-intensity projections. Correlation between cryo-fluorescence images 439 and low-magnification TEM search maps (lamella overviews) was performed using IMOD61,62 440 and FIJI60. 441 Cryo-ET image acquisition 442 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 11 For purified V12-ferritin and Δ 6-ferritin nanocages, the purified samples were applied to glow-443 discharged Quantifoil R2/2 Cu 200 mesh grids and plunge-frozen as described above. Tilt 444 series were collected from -55° to +55° in 5° increments with dose-symmetric tilt scheme (8e-/Å2 445 per tilt; total accumulated dose ~184 e-/Å2) on a 300 kV Titan Krios G4 microscope equipped 446 with a Gatan K3 detector and a BioQuantum energy filter. Data were acquired at a nominal 447 defocus of 3-4 µm, using Thermo Fisher Tomography software. 448 For bacterial and mammalian lamellae, the stage was tilted by ±9º to compensate for the final 449 milling angle. Tilt series were collected from -60° to +60° using a dose-symmetric tilt scheme 450 with 2° increments (total dose ~183 e⁻ /Å2). The E. coli lamellae were imaged on a 300 kV Titan 451 Krios G4 microscope equipped with a Gatan K3 detector and energy filter, using a defocus of 3-452 5 µm and a calibrated pixel size of 1.6 Å. HeLa cell lamellae were first screened by collecting 453 low-magnification search maps for all existing lamellae. Cryo-fluorescence correlation with 454 CLEM data was performed as described above to identify regions containing both GFP signal 455 and mitochondria for targeted cryo-ET data acquisition. Tilt series were collected on a 200 kV 456 Thermo Fisher Glacios TEM equipped with a Falcon 4 direct detector, using 4-5 µm defocus, a 457 70 µm objective aperture, and a pixel size of 1.5 Å. 458 Cryo-ET data processing 459 For purified V12-ferritin and Δ 6-ferritin nanocages, tilt series were aligned and reconstructed in 460 RELION547,63 with integrated motion correction and CTF correction. Reconstructed tomograms 461 were binned fourfold and processed with IsoNet for missing-wedge compensation and denoising, 462 enabling improved model fitting and visualization. 463 Subtomogram averaging (STA) was performed using crYOLO43,44 for automatic ferritin cage 464 picking and PyTom45 for localization of smaller tag particles. Amond tested approaches, 465 crYOLO43,44 was most effective for large in vitro particles, whereas PyTom45 performed better for 466 small tag features in vitro and in situ cage detection. Independent refinements of cage and tag 467 subtomograms were carried out in RLION547,64,65, yielding final resolution of 5 Å and 22 Å for 468 ferritin cage and the V12 tag, respectively, and 6.7 Å and 7.3 Å for ferritin cage and Δ 6 tag, 469 respectively. Averaged densities were fitted into corresponding tomograms using UCSF 470 ChimeraX66 for visualization, tags detection and structural interpretation. 471 For Bacterial and mammalian cell tomograms, tilt series of E. coli and HeLa cell lamellae were 472 motion corrected with Motioncor367 and reconstructed using IMOD (weighted back-projection 473 mode)61,62 and binned fourfold, yielding final pixel size of 6.4 Å (E. coli) and 6 Å (HeLa). The 474 tomograms were subsequently processed with IsoNet52 for missing-wedge compensation and 475 denoising, using custom masks generated to focus on regions enriched in ferritin cages or 476 mitochondrial membranes and associated tags. Ribosomes, membranes, and ferritin nanocages 477 were segmented using AI-assisted tools in Amira (Thermo Fisher Scientific). 478 Tag-like densities were identified through manual inspection and validated by docking averaged 479 tag models obtained from purified samples into tomographic volumes using ChimeraX66. While 480 PyTom45 enabled efficient in situ cage picking, existing algorithms failed to reliably detect the 481 smaller tag densities due to the combination of the missing wedge and the crowded cellular 482 environment. STA of in situ ferritin cages, performed using Warp46 and RELION563,64, achieved a 483 final resolution of ~12 Å. 484 Current algorithmic limitations hinder robust automated identification of small, shape-defined 485 tags in situ. Ongoing efforts aim to develop new computational approaches tailored for these 486 geometrically defined tags to enhance their detection and verification within cellular tomograms. 487 Although technically challenging, such advancements are expected to substantially broaden the 488 applicability and usability of both tags in future studies. 489 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 12 Data analysis and visualization 490 All density maps were visualized in UCSF ChimeraX66 and segmented in Amira (Thermo Fisher 491 Scientific). Electrostatic potential surfaces were rendered in PyMOL with APBS55. Fourier shell 492 correlation (FSC) was used to estimate resolution68. For 3D modeling, structures were fitted into 493 tomograms using ChimeraX66. Figures were prepared in ChimeraX, PyMOL, BioRender 494 (Extended Data Figs. 3, 9A), and Adobe Illustrator. 495 496

497 1. Gustafsson, M. G. L. Surpassing the lateral resolution limit by a factor of two 498 using structured illumination microscopy. J Microsc 198, 82–87 (2000). 499 2. Wichmann, J. & Hell, S. W. Breaking the diffraction resolution limit by stimulated 500 emission: stimulated-emission-depletion fluorescence microscopy. Optics Letters, 501 Vol. 19, Issue 11, pp. 780-782 19, 780–782 (1994). 502 3. Klar, T. A. & Hell, S. W. Subdiffraction resolution in far-field fluorescence 503 microscopy. Optics Letters, Vol. 24, Issue 14, pp. 954-956 24, 954–956 (1999). 504 4. Betzig, E. et al. Imaging intracellular fluorescent proteins at nanometer resolution. 505 Science 313, 1642–5 (2006). 506 5. Rust, M. J., Bates, M. & Zhuang, X. Sub-diffraction-limit imaging by stochastic 507 optical reconstruction microscopy (STORM). Nature Methods 2006 3:10 3, 793–508 796 (2006). 509 6. Schermelleh, L. et al. Super-resolution microscopy demystified. Nature Cell 510 Biology 2019 21:1 21, 72–84 (2019). 511 7. Sahl, S. J., Hell, S. W. & Jakobs, S. Fluorescence nanoscopy in cell biology. Nat 512 Rev Mol Cell Biol 18, 685–701 (2017). 513 8. Lu č ić , V., Leis, A. & Baumeister, W. Cryo-electron tomography of cells: connecting 514 structure and function. Histochemistry and Cell Biology 2008 130:2 130, 185–196 515 (2008). 516 9. Turk, M. & Baumeister, W. The promise and the challenges of cryo-electron 517 tomography. FEBS Lett 594, 3243–3261 (2020). 518 10. Wan, W. & Briggs, J. A. G. Cryo-Electron Tomography and Subtomogram 519 Averaging. Methods Enzymol 579, 329–67 (2016). 520 11. Schur, F. K. Toward high-resolution in situ structural biology with cryo-electron 521 tomography and subtomogram averaging. Curr Opin Struct Biol 58, 1–9 (2019). 522 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 13 12. McCafferty, C. L. et al. Integrating cellular electron microscopy with multimodal 523 data to explore biology across space and time. Cell 187, 563–584 (2024). 524 13. Medalia, O. et al. Macromolecular Architecture in Eukaryotic Cells Visualized by 525 Cryoelectron Tomography. Science (1979) 298, 1209–1213 (2002). 526 14. Nicastro, D. et al. The molecular architecture of axonemes revealed by 527 cryoelectron tomography. Science 313, 944–8 (2006). 528 15. Medalia, O. et al. Macromolecular architecture in eukaryotic cells visualized by 529 cryoelectron tomography. Science 298, 1209–13 (2002). 530 16. Nicastro, D. et al. The molecular architecture of axonemes revealed by 531 cryoelectron tomography. Science 313, 944–8 (2006). 532 17. Waltz, F. et al. In-cell architecture of the mitochondrial respiratory chain. Science 533 387, 1296–1301 (2025). 534 18. Asano, S. et al. Proteasomes. A molecular census of 26S proteasomes in intact 535 neurons. Science 347, 439–42 (2015). 536 19. Xue, L. et al. Visualizing translation dynamics at atomic detail inside a bacterial 537 cell. Nature 610, 205–211 (2022). 538 20. Pfeffer, S. et al. Dissecting the molecular organization of the translocon-539 associated protein complex. Nat Commun 8, 14516 (2017). 540 21. Young, L. N. et al. ExoSloNano: Multi-Modal Nanogold Tags for identification of 541 Macromolecules in Live Cells & Cryo-Electron Tomograms. bioRxiv 542 https://doi.org/10.1101/2024.10.12.617288 (2024) doi:10.1101/2024.10.12.617288. 543 22. Wang, Q., Mercogliano, C. P. & Löwe, J. A ferritin-based label for cellular electron 544 cryotomography. Structure 19, 147–154 (2011). 545 23. Clarke, N. I. & Royle, S. J. FerriTag is a new genetically-encoded inducible tag for 546 correlative light-electron microscopy. Nat Commun 9, 1–10 (2018). 547 24. Silvester, E. et al. DNA origami signposts for identifying proteins on cell 548 membranes by electron cryotomography. Cell 184, 1110--1121.e16 (2021). 549 25. Fung, H. K. H. et al. Genetically encoded multimeric tags for subcellular protein 550 localization in cryo-EM. Nat Methods 20, 1900–1908 (2023). 551 26. Last, M. G. F., Voortman, L. M. & Sharp, T. H. Imaging intracellular components in 552 situ using super-resolution cryo-correlative light and electron microscopy. in 553

in Cell Biology vol. 187 223–248 (Academic Press Inc., 2024). 554 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 14 27. Last, M. G. F., Voortman, L. M. & Sharp, T. H. Building a super-resolution 555 fluorescence cryomicroscope. in Methods in Cell Biology vol. 187 205–222 556 (Academic Press, 2024). 557 28. Chang, Y.-W. et al. Correlated cryogenic photoactivated localization microscopy 558 and cryo-electron tomography. Nat Methods 11, 737–739 (2014). 559 29. Liu, B. et al. Three-dimensional super-resolution protein localization correlated 560 with vitrified cellular context. Sci Rep 5, 13017 (2015). 561 30. Tuijtel, M. W., Koster, A. J., Jakobs, S., Faas, F. G. A. & Sharp, T. H. Correlative 562 cryo super-resolution light and electron microscopy on mammalian cells using 563 fluorescent proteins. Sci Rep 9, 1369 (2019). 564 31. Dahlberg, P. D. et al. Cryogenic single-molecule fluorescence annotations for 565 electron tomography reveal in situ organization of key proteins in Caulobacter. 566 Proc Natl Acad Sci U S A 117, 13937–13944 (2020). 567 32. Wolff, G., Hagen, C., Grünewald, K. & Kaufmann, R. Towards correlative 568 super‐ resolution fluorescence and electron cryo‐ microscopy. Biol Cell 108, 245–569 258 (2016). 570 33. Dahlberg, P. D. & Moerner, W. E. Cryogenic Super-Resolution Fluorescence and 571 Electron Microscopy Correlated at the Nanoscale. Annu Rev Phys Chem 72, 253–572 278 (2021). 573 34. Kounatidis, I. et al. 3D Correlative Cryo-Structured Illumination Fluorescence and 574 Soft X-ray Microscopy Elucidates Reovirus Intracellular Release Pathway. Cell 575 182, 515-530.e17 (2020). 576 35. DeRosier, D. J. Where in the cell is my protein? Q Rev Biophys 54, e9 (2021). 577 36. Klumpe, S. et al. A modular platform for automated cryo-FIB workflows. Elife 10, 578 (2021). 579 37. Huang, P. S. et al. High thermodynamic stability of parametrically designed helical 580 bundles. Science (1979) 346, 481–485 (2014). 581 38. Di Pietro, S. M., Cascio, D., Feliciano, D., Bowie, J. U. & Payne, G. S. Regulation 582 of clathrin adaptor function in endocytosis: novel role for the SAM domain. EMBO 583 J 29, 1033–44 (2010). 584 39. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. 585 Nature 596, 583–589 (2021). 586 40. Mirdita, M. et al. ColabFold: making protein folding accessible to all. Nat Methods 587 19, 679–682 (2022). 588 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 15 41. Huddy, T. F. et al. Blueprinting extendable nanomaterials with standardized protein 589 blocks. Nature 627, 898–904 (2024). 590 42. Stillman, T. J. et al. The high-resolution X-ray crystallographic structure of the 591 ferritin (EcFtnA) of Escherichia coli; comparison with human H ferritin (HuHF) and 592 the structures of the Fe3+ and Zn2+ derivatives11Edited by R. Huber. J Mol Biol 593 307, 587–603 (2001). 594 43. Wagner, T. et al. SPHIRE-crYOLO is a fast and accurate fully automated particle 595 picker for cryo-EM. Commun Biol 2, 218 (2019). 596 44. Wagner, T. & Raunser, S. The evolution of SPHIRE-crYOLO particle picking and 597 its application in automated cryo-EM processing workflows. Communications 598 Biology 2020 3:1 3, 1–5 (2020). 599 45. Hrabe, T. et al. PyTom: A python-based toolbox for localization of macromolecules 600 in cryo-electron tomograms and subtomogram analysis. J Struct Biol 178, 177–601 188 (2012). 602 46. Tegunov, D. & Cramer, P . Real-time cryo-electron microscopy data preprocessing 603 with Warp. Nat Methods 16, 1146–1152 (2019). 604 47. Bharat, T. A. M. & Scheres, S. H. W. Resolving macromolecular structures from 605 electron cryo-Tomography data using subtomogram averaging in RELION. Nat 606 Protoc 11, 2054–2065 (2016). 607 48. Rice, G. et al. TomoTwin: generalized 3D localization of macromolecules in cryo-608 electron tomograms with structural data mining. Nat Methods 20, 871–880 (2023). 609 49. Castaño-Díez, D., Kudryashev, M., Arheit, M. & Stahlberg, H. Dynamo: A flexible, 610 user-friendly development tool for subtomogram averaging of cryo-EM data in 611 high-performance computing environments. J Struct Biol 178, 139–151 (2012). 612 50. Chaillet, M. L. et al. Extensive Angular Sampling Enables the Sensitive 613 Localization of Macromolecules in Electron Tomograms. Int J Mol Sci 24, 13375 614 (2023). 615 51. Liu, G. et al. DeepETPicker: Fast and accurate 3D particle picking for cryo-616 electron tomography using weakly supervised deep learning. Nat Commun 15, 1–617 15 (2024). 618 52. Liu, Y . T. et al. Isotropic reconstruction for electron tomography with deep learning. 619 Nat Commun 13, 1–17 (2022). 620 53. Nguyen, T. T. & Voeltz, G. K. An ER phospholipid hydrolase drives ER-associated 621 mitochondrial constriction for fission and fusion. Elife 11, (2022). 622 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 16 54. Jumper, J. et al. Highly accurate protein structure prediction with AlphaFold. 623 Nature 596, 583–589 (2021). 624 55. Baker, N. A., Sept, D., Joseph, S., Holst, M. J. & McCammon, J. A. Electrostatics 625 of nanosystems: application to microtubules and the ribosome. Proc Natl Acad Sci 626 U S A 98, 10037–41 (2001). 627 56. Sayou, C. et al. A SAM oligomerization domain shapes the genomic binding 628 landscape of the LEAFY transcription factor. Nat Commun 7, 11222 (2016). 629 57. Trevelyan, S. J. et al. Structure-based mechanism of preferential complex 630 formation by apoptosis signal-regulating kinases. Sci Signal 13, (2020). 631 58. Rocklin, G. J. et al. Global analysis of protein folding using massively parallel 632 design, synthesis, and testing. Science 357, 168–175 (2017). 633 59. Punjani, A., Rubinstein, J. L., Fleet, D. J. & Brubaker, M. A. cryoSPARC: 634 algorithms for rapid unsupervised cryo-EM structure determination. Nat Methods 635 14, 290–296 (2017). 636 60. Schindelin, J. et al. Fiji: an open-source platform for biological-image analysis. Nat 637

9, 676–82 (2012). 638 61. Mastronarde, D. N. Accurate, automatic determination of astigmatism and phase 639 with Ctfplotter in IMOD. J Struct Biol 216, 108057 (2024). 640 62. Mastronarde, D. N. & Held, S. R. Automated tilt series alignment and tomographic 641 reconstruction in IMOD. J Struct Biol 197, 102–113 (2017). 642 63. Burt, A. et al. An image processing pipeline for electron cryo-tomography in 643 RELION-5. FEBS Open Bio 14, 1788–1804 (2024). 644 64. Bharat, T. A. M. & Scheres, S. H. W. Sub-Tomogram Averaging in RELION . 645 Nature protocols http://biorxiv.org/lookup/doi/10.1101/030544 (2015) 646 doi:10.1101/030544. 647 65. Scheres, S. H. W. RELION: Implementation of a Bayesian approach to cryo-EM 648 structure determination. J Struct Biol 180, 519–530 (2012). 649 66. Meng, E. C. et al. UCSF ChimeraX: Tools for structure building and analysis. 650 Protein Science 32, (2023). 651 67. Zheng, S. Q. et al. MotionCor2: anisotropic correction of beam-induced motion for 652 i mproved cryo-electron microscopy. Nat Methods 14, 331–332 (2017). 653 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 17 68. Rosenthal, P. B. & Henderson, R. Optimal Determination of Particle Orientation, 654 Absolute Hand, and Contrast Loss in Single-particle Electron Cryomicroscopy. J 655 Mol Biol 333, 721–745 (2003). 656 657 658 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 1 1 Fig. 1. Design and structural characterization of V- and Δ -shaped protein tags. 2 (a) Design of V-shaped tag (V12) predicted by AlphaFold2. Two three-helix arms connected by 3 a rigid SAM-turn motif, forming an angle of ~72° with an arm length of ~12 nm. 4 (b) Cryo-EM micrograph of purified V12 collected on a 200-keV Glacios cryo-TEM. Orange 5 boxes mark representative V-shaped particles; enlarged views are shown at right. 6 (c) Representative 2D class averages showing the characteristic V-shaped architecture, mostly 7 in top view. Scale bar, 10 nm. 8 (d) Cryo-EM density map of extended V12 (62 kDa) with the predicted model fitted into the 9 density. 10 (e) Design of the compact Δ 6 tag predicted by AlphaFold2, consisting of three copies of C3-11 symmetric trimeric motif assembly ~6 nm in diameter. 12 (f) Cryo-EM micrograph of purified Δ 6 collected on a 200-keV Glacios cryo-TEM. Blue boxes 13 mark individual triangular particles; enlarged views are shown at right. 14 (g) 2D class averages of Δ 6 showing compact triangular geometries. Predominant top views are 15 shown; side views are indicated by red boxes. Scale bar, 5 nm. 16 (h) Cryo-EM density map and fitted predicted model of the compact Δ 6 (85 kDa) reveal the 17 expected triangular architecture. 18 19 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 2 Fig. 2. In vitro visualization and analysis of V12- and Δ 6-tagged ferritin nanocages. 20 (a) Representative cryo-electron tomogram slice of purified V12-tagged ferritin nanocages. 21 Insets show enlarged regions highlighting individual cages and associated V-shaped densities. 22 Scale bar, 20 nm (left) and 10 nm (right). 23 (b, c) STA structures and representative orientated slice views of the ferritin cage (b) and the 24 V12 tag (c), each reconstructed independently from purified tomograms. 25 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 3 (d) Fraction of particles retained after classification for V12 (orange) and ferritin cages 26 (magenta), illustrating the challenge of identifying small V-shaped tags in crowded tomograms. 27 (e) Model of the ferritin cage and V12 tag obtained by STA and fitted into the 3D tomographic 28 density. 29 (f) Comparison of model and tomogram slices. Top, representative model slice corresponding to 30 the tomogram slice; middle, tomogram slice; bottom, fitted model slices showing close 31 agreement between model and density. Scale bar, 10 nm. 32 (g) Representative cryo-electron tomogram slice of purified Δ 6-tagged ferritin nanocages. Insets 33 show enlarged regions highlighting individual cage and associated compact, triangular densities 34 surrounding the cages corresponding to the Δ 6 tag. Scale bar, 20 nm (left) and 10 nm (right). 35 (h, i) STA structures and representative orientated slice views of the ferritin cage (h) and the Δ 6 36 tags (i) reconstructed independently from purified tomograms. 37 (j) Fractions of particles retained after classification for Δ 6 (blue) and ferritin cages (magenta) 38 showing that the compact triangular tags are more readily identified in vitro but may influence 39 the structural analysis of target protein. 40 (k)Model of the ferritin cage and Δ 6 tag obtained by STA and fitted into the 3D tomographic 41 density. 42 (l) Comparison of model and tomogram slices. Top, representative model slice corresponding to 43 the tomogram slice; middle, tomogram slice; bottom, fitted model slices showing close 44 agreement between model and density. Scale bar, 10 nm. 45 46 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 4 47 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 5 Fig. 3. In situ visualization of V12- and Δ 6-tagged ferritin cages in E. coli by cryo-ET. 48 (a) Representative cryo-tomographic slice of a FIB-milled E. coli cell expressing V12-tagged 49 ferritin nanocages. Orange boxes mark examples of nanocages. Scale bar, 20 nm. 50 (b) Automated segmentation with Amira showing apo-ferritin cages (pink), ribosomes (green) 51 within the cytoplasm and cell membranes (grey). 52 (c) Enlarged views of boxed regions in panel (a) showing peripheral extended densities 53 corresponding to V12 tags with annotated views at right. Scale bar, 10nm. 54 (d) STA structure of the apo-ferritin cage from in situ particle picking. 55 (e) Fraction of ferritin cage particles retained after in situ classification, illustrating the low yield 56 of usable particles in crowded cellular environments. 57 (f) Comparison of model and in situ tomogram slices. Top, representative model slice 58 corresponding to the tomogram slice; middle, tomogram slice; bottom, fitted model slices 59 showing close agreement between model and density. Scale bar, 10 nm. 60 (g) Model of the ferritin cage and V12 tag obtained by in vitro STA and fitted into the in situ 3D 61 tomographic density. 62 (h) Representative cryo-tomographic slice of a FIB-milled E. coli cell expressing Δ 6-tagged 63 ferritin nanocages. Blue boxes mark examples of nanocages. 64 (i) Segmentation highlighting apo-ferritin cages (pink). 65 (j) Enlarged views of boxed regions in (h) showing compact peripheral densities corresponding 66 to Δ 6 tags with the annotation at right. Scale bar, 10nm. 67 (k) Comparison of model and in situ tomogram slices. Top, representative model slice 68 corresponding to the tomogram slice; middle, tomogram slice; bottom, fitted model slices 69 showing close agreement between model and density. Scale bar, 10 nm. 70 (l) Model of the ferritin cage and Δ 6 tag obtained by in vitro STA and fitted into the in situ 3D 71 tomographic density. 72 73 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 6 74 Fig. 4. Mitochondrial surface display of V12- and Δ 6-tagged TOM70NTD fusion proteins in 75 HeLa cells. 76 (a, b) Schematic of TOM70NTD-V12-GFP and TOM70NTD-Δ 6-GFP constructs. The TOM70 N-77 terminal domain (TOM70NTD) anchors to the mitochondrial outer membrane (OM), positioning 78 the V12 or Δ 6 tags on the cytosolic face. 79 (c) Immunoblot of HeLa cell lysates expressing TOM70NTD-V12-GFP or TOM70NTD-Δ 6-GFP 80 probed with anti-GFP antibody. GAPDH served as a loading control. 81 (d and m) Confocal fluorescence images showing mitochondrial localization of TOM70NTD-V12-82 GFP (d) and TOM70NTD-Δ 6-GFP (m). GFP signal colocalizes with the mitochondrial marker 83 Hsp60 (Pearson’s R = 0.61 and 0.89, respectively). Scale bars, 10 μ m. 84 (e) Cryo-correlative light and electron microscopy (cryo-CLEM) of TOM70NTD-V12-GFP cell. 85 Fluorescence overlay shows GFP colocalized mitochondria on a FIB-milled lamella. 86 (f) Tomographic slice of the corresponding region showing mitochondria, ribosomes, and 87 cytosolic features; orange arrowheads indicate V-shaped densities. 88 (g) Segmented tomogram showing mitochondria (green), ribosomes (red), and V12-tag 89 densities (yellow). In panels e-g, purple asterisks (*) mark the same mitochondrial cristae, and 90 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint 7 purple hash symbols (#) mark the same vesicle. 91 (h-l) Enlarged tomographic slices showing surface V12-tag densities (orange arrowheads) along 92 the mitochondrial outer membrane (OM, green lines) and annotated V12 (yellow). IM, inner 93 membrane. Scale bar, 10 nm. 94 (n) Cryo-CLEM of TOM70NTD-Δ 6-GFP cell. Fluorescence overlay shows GFP colocalized 95 mitochondria on a FIB-milled lamella. 96 (o) Tomographic slice of the corresponding region showing mitochondrion and cytosolic 97 features; blue arrowheads indicate triangular-shaped densities. In panels N and O, purple 98 asterisks (*) mark the same mitochondrion. 99 (p-u) Enlarged tomographic slices showing compact Δ 6-tag densities (blue arrowheads) on the 100 mitochondrial outer membrane (green lines) of TOM70NTD-Δ 6-GFP cells. OM, outer membrane; 101 IM, inner membrane. Scale bar, 10 nm. 102 103 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted January 20, 2026. ; https://doi.org/10.64898/2026.01.16.700029doi: bioRxiv preprint