AlphaFold2-guided engineering of split-GFP technology enables labeling of endogenous tubulins across species while preserving function

Dynamic properties are essential for microtubule (MT) physiology. Current techniques for in vivo imaging of MTs present intrinsic limitations in elucidating the isotype-specific nuances of tubulins, which contribute to their versatile functions. Harnessing the power of the AlphaFold2 pipeline, we engineered a strategy for the minimally invasive fluorescence labeling of endogenous tubulin isotypes or those harboring missense mutations. We demonstrated that a specifically designed 16-amino acid linker, coupled with sfGFP11 from the split-sfGFP system and integration into the H1-S2 loop of tubulin, facilitated tubulin labeling without compromising MT dynamics, embryonic development, or ciliogenesis in Caenorhabditis elegans. Extending this technique to human cells and murine oocytes, we visualized MTs with the minimal background fluorescence and a pathogenic tubulin isoform with fidelity. The utility of our approach across biological contexts and species set an additional paradigm for studying tubulin dynamics and functional specificity, with implications for understanding tubulin-related diseases known as tubulinopathies.


Introduction
The microtubule (MT) cytoskeleton serves as a dynamic framework of protein assemblies, endowing cells with both structural robustness and functional plasticity [1,2].Assembled through the polymerization of tubulin monomers into cylindrical arrays [3], MTs display a behavior known as dynamic instability-a meticulously orchestrated equilibrium between polymerization and spontaneous depolymerization phases [4].This regulated dynamic is instrumental for fulfilling diverse physiological functions of MTs [5].Consequently, the highresolution imaging of MT dynamics is pivotal for dissecting their mechanistic roles in cellular activities.
Over the past several decades, a myriad of methodologies has been developed to monitor MT dynamics in vivo.Conventional techniques have used microinjection of fluorescein-or rhodamine-labeled tubulins into cells or embryos to measure MT dynamics [5][6][7].Furthermore, green fluorescent protein (GFP)-tagged MT-associated proteins (MAPs), such as endbinding (EB) protein or the recent "Stable Microtubule-Associated Rigor-Kinesin" (Stable-MARK) [8,9], have been rigorously validated for noninvasive and in vivo visualization without disrupting native MT behavior.Complementary to these approaches, computational algorithms for quantification and automated trajectory analysis have been developed, thereby facilitating the acquisition of quantified datasets and imparting real-time measurement of MT dynamics in living systems [8].
Nevertheless, most existing techniques exhibit intrinsic limitations when delving into the functional specificity of tubulin isotypes, an indispensable aspect of MT biology that contributes to their diverse roles in cellular processes [1].Metazoan genomes encode an array of αand β-tubulin genes, each exhibiting unique spatial and temporal expression profiles and specialized posttranslational modifications within cellular compartments [1,10].For instance, the mammalian tubulin isotypes TUBB2 and TUBB3 are predominantly incorporated into neuronal MTs, playing an indispensable role in neurite outgrowth [11].Conversely, others (for instance, Caenorhabditis elegans TBA-5) are more prevalent in cilia and flagella with exclusive localization within specific ciliary segments, contributing to their unique structural and functional roles [12].This functional specificity of tubulin isotypes is expected to provide important regulatory mechanisms, empowering the MT cytoskeleton to various functionalities [1].Thus, the real-time interrogation of tubulin isotypes within living cells is foundational for deciphering the functional intricacies of MTs.
Mutations in tubulin-coding genes cause human diseases collectively designated as tubulinopathies such as cortical malformations, manifesting as microcephaly, lissencephaly, or polymicrogyria [13][14][15].The symptomatic spectrum of tubulinopathies extends from severe intellectual deficits to nuanced cognitive impairments, accentuating the indispensable role of tubulins in brain development [15].Also, tubulinopathies encompass ocular and renal anomalies [16].Notably, many of these mutations are de novo [16][17][18][19], underscoring the underestimated impact of tubulinopathies in human diseases.Indeed, AlphaMissense-based predictions revealed that over 80% missense mutations in tubulins (81.9% for α-tubulin TUBA1A and 82.5% for β-tubulin TUBB2B) are likely pathogenic, overwhelming those in KRAS (66.0%) or BRAF (57.7%)-2 hotspot proteins in cancer research (S1A Fig) [20,21].Comprehensive exploration of the functional specificities among divergent wild-type and mutant tubulin isotypes holds the promise of elucidating the underlying pathophysiological mechanisms [13], thereby facilitating the development of therapeutic strategies.Therefore, there is an imminent requirement to monitor the dynamics of individual tubulin isotypes in both wild-type and pathological forms.
Currently, GFP-tagged tubulins have been widely used to visualize tubulin isotypes across different species [22,23].However, C-terminal GFP tagging of tubulin impedes its interaction with MAPs or other regulatory elements, whereas N-terminal tagging obstructs the incorporation of tubulin into MT architecture [22,24].Moreover, endogenous (or knock-in (KI)) GFP labeling of tubulin produces more pronounced disruptions in MT functionality compared to overexpression (OE) [24].This could be attributed to the functional redundancy of tubulin isotypes and resilience exhibited by endogenous, nontagged tubulins.Nonetheless, ectopic OE of wild-type GFP-tagged tubulin isotypes has been widely adopted across diverse cellular environments, including the nematode [12], mammalian cell lines [25], or mouse embryos [26], with a fraction of GFP-tagged tubulins providing strong fluorescent signals allowing to visualized MTs.Despite those successful applications to mark wild-type MTs, GFP labeling fails to trace mutated tubulins associated with tubulinopathy with high fidelity.Instead, the visualization of mutated tubulins were accomplished by immunofluorescence (IF) staining in almost all situations [27][28][29][30][31]. Consequently, there is still a lack of unobtrusive, functional imaging of dynamic tubulin isotypes.
Guided by artificial intelligence-driven AlphaFold2 pipeline [32], we developed a novel strategy for the functional labeling of endogenous tubulins while preserving their inherent functionalities.We overcame 3 technical obstacles: the optimal site for tubulin labeling, the selection of fluorescent markers, and the constitution of the linker sequence.Employing the nematode C. elegans as a model system, we showed that a 16-amino acid (aa) linker, in conjunction with sfGFP11 from the split-sfGFP system and inserted at the H1-S2 loop of tubulin [33][34][35], enabled labeling of either α-or β-tubulin without compromising MT assembly, embryogenesis, or ciliogenesis.Extending this technique to tubulins in HeLa cells and murine oocytes, we achieved visualization of MT networks with significantly diminished background fluorescence.This suggests a majority of labeled tubulin is successfully integrated into MT assemblies, thereby implying that our approach is superior to existing technologies.These findings collectively underscore the extensive applicability of our strategy for functional tubulin labeling across diverse cellular milieus and species.

AlphaFold2-guided design of functional fluorescence labeling of tubulin
Leveraging the C. elegans system-renowned for its suitability for genetic manipulation and livecell imaging-we carried out initial investigations.The C. elegans genome harbors 9 α-tubulin and 6 β-tubulin isotypes [36].Supposing that regions of low sequence conservation might be more permissive for the addition of an epitope tag, we targeted the H1-S2 loop of the tubulin chain, a notably variable region (S1B Fig) .The H1-S2 loop of tubulin protrudes from the internal surface of hollow MTs [37], and previous studies have elucidated that insertion of up to 17 aa into this loop had no apparent functional disruption on tubulins [34].Therefore, affinity tags have been inserted into this loop for biochemical preparation or IF imaging of individual tubulin isotypes [38][39][40][41].To identify the most appropriate insertion site within this loop, we performed multiple sequence alignment (MSA) for each set of α-and β-tubulin isotypes to acquire the least conserved and most structurally flexible sites amenable for manipulation (S1C and S1D Fig) .Through MSA, the gap between the 43rd Gly and 44th Val in α-tubulin TBA-5 and the gap between the 37th Lys and 38th Gly in β-tubulin TBB-2 were the optimal candidates for insertion (S1C and S1D Fig).
Newly synthesized tubulin is captured by the chaperone prefoldin and subsequently delivered to the ring-shaped chaperonin TRiC / CCT complex [42,43], where it undergoes a stringent conformational folding process to form the mature protein.According to recent highresolution structural analyses of tubulin within the TRiC / CCT complex [44], the incorporation of a full-length fluorescent protein into 55-kDa tubulin may perturb this folding process, yielding improperly folded tubulin.To demonstrate this, we performed Rosetta relax runs for untagged tubulins as well as GFP-or mScarlet-tagged tubulins within TRiC / CCT complex [45,46].Full-length GFP or mScarlet was inserted directly into H1-S2 loop of α-tubulin TBA-5, which was subsequently deposited into TRiC / CCT complex for a sufficient relaxation of energy.After relaxation following Rosetta relax protocol in designated "rigid" TRiC / CCT complex, untagged tubulins remained relatively intact and structured in the chaperones (S2A Fig) .However, either GFP-or mScarlet-tagged tubulins were crushed thoroughly due to steric hindrance created by full-length fluorescent proteins (S2B and S2C Fig) , indicating GFP or mScarlet was not eligible for labeling tubulins within the H1-S2 loop.Therefore, we adopted the split-GFP system, inserting a short 16-aa GFP11 fragment directly into the H1-S2 loop, while coexpressing the complementary GFP1-10 fragment in the sensory neurons of C. elegans.Regrettably, this initial attempt yielded no detectable GFP signal at all (S4F Fig) .Next, we harnessed the AlphaFold2 pipeline to optimize the GFP11 insertion tag.According to predictions by AlphaFold2 (Fig 1A ), embedding GFP11 within the H1-S2 loop without an adjacent linker precluded its effective interaction with the complementary GFP1-10 fragment, likely due to the rigid topology of GFP11.To address this problem, we extended the GSlinker on both sides of GFP11 to lengths of 6 aa (GS-linker 1), 12 aa (GS-linker 2), or 16 aa (GS-linker 3), in sequential iterations.AlphaFold-based modeling indicated that a linker of at least 12 aa (GS-linker 2) permitted sporadic binding between GFP11 and GFP1-10 (observed in 1 out of 5 independent models), while a 16-aa GS-linker 3 should enable robust binding (observed in 4 out of 5 independent models) (Fig 1A

GFP11-i labels endogenous mutated α-tubulins without affecting phenotypes
Next, we assessed the performance of the GFP11-i labeling approach on structural-sensitive tubulin variants harboring missense mutations, as these variants are analogous to diseaserelated mutated tubulins [13].Specifically, we examined a ciliary tubulin isotype, α-tubulin TBA-5, which is predominantly localized in the distal segments of cilia within the sensory neurons of C. elegans.Previous studies indicated that several mutations in TBA-5 would disrupt ciliary MTs (Fig 3A) [12].The A19V missense mutation in TBA-5 was hypothesized to disturb tubulin folding [12,49], as supported by significantly increased van der Waals overlaps (VDWoverlap � 0.8 Å) computed by ChimeraX software (S3A Fig) [50].Remarkably, deletion of TBA-5 (or loss-of-function TBA-5) has negligible effects on cilia due to the functional redundancy of tubulin isotypes [12,36], while A19V mutation transformed the configuration of TBA-5 to "gain-of-function," thereby destructing ciliary MTs upon its assembly into MTs, resulting in the loss of the ciliary distal segments (Fig 3A) [12].Furthermore, tba-5 (A19V) mutants exhibit temperature sensitivity, characterized by the absence of ciliary distal segments at 15˚C, while maintaining relatively intact cilia at 25˚C [12].This makes the system particularly suitable for investigating whether fluorescent labeling of such tubulin variants has any impact on their functional attributes.
The che-3::T2A::gfp1-10, a KI allele of the ciliated neuron-associated dynein-2 heavy chain CHE-3, served as donor of GFP1-10 fragments in ciliated neurons.Intriguingly, the cilia morphology and length observed in tba-5 (A19V) (gfp11-i) KI animals were indistinguishable from those of their untagged counterparts (Fig 3B -3E), albeit with a modest rescue of the Dyf phenotype (S4G Fig) .Consistent with tba-5 (A19V) mutant phenotype [12], we revealed that tba-5 (A19V) (gfp11-i) KI animals also possessed relatively intact cilia at 25˚C but were devoid of ciliary distal segments at 15˚C (Figs 3F and S5A), preserving the temperature-sensitive nature of TBA-5 (A19V) with respect to ciliary morphology.Additionally, we created wild-type tba-5 (gfp11-i) KI animals and did not observe any discernible ciliary defect (Fig 3G and 3H).Using the intraflagellar transport component DYF-11 as markers, we found the velocity of ciliary kinesins in GFP11-i-tagged TBA-5 animals was also indistinguishable from that of wild-type animals (S5B and S5C Fig), implying that GFP11-i labeling may not disturb interaction between MTs and MAPs.Taken together, GFP11-i is an effective tool for the functional labeling of α-tubulin TBA-5, especially applicable to mutated forms.
As GFP11-i is amenable to live-cell imaging, we investigated the dynamics of GFP11-i tagged wild-type TBA-5 or TBA-5 (A19V) through FRAP (fluorescence recovery after photobleaching) experiments.Photobleaching of ciliary distal segments expressing wild-type TBA-5 (GFP11-i) resulted in fluorescence recovery primarily at the ciliary tips, which represented the highly dynamic plus ends of A-tubules of axonemal MTs (Fig 3I and 3J) [12].The average recovery rate at the ciliary tips was 32.1 ± 7.3% after a 10-min recovery (Fig 3J), comparable to that (approximatelyAU : PleasenotethatasperPLOSstyle; donotusethesymbol � inprosetomeanaboutora 30%) measured in previous studies using overexpressed TBB-4::YFP [12].Given that TBA-5 (A19V) disrupted ciliary distal segments at 15˚C, we photobleached the remaining ciliary middle segments expressing TBA-5 (A19V) (GFP11-i) and observed fluorescence recovery preferentially at the ciliary tips, which represented the plus ends of both Atubules and B-tubules of axonemal MTs (Fig 3K and 3L) [12,52].The average recovery rate at the ciliary tips was 60.0 ± 13.6% after a 10-min recovery in tba-5 (A19V) mutant (Fig 3L ), faster than that of middle (B-tubule, approximately 25%) or distal segment tips (A-tubule, approximately 30%) measured in wild-type phasmid cilia [12], implying that TBA-5 (A19V) was incorporated into both A-and B-tubules of ciliary MTs.These findings were previously inaccessible using either traditional fluorescent protein tagging or IF imaging, underscoring the exclusive advantage of GFP11-i labeling in exploring the dynamic functionality of tubulin isotypes or variants, a fundamental aspect of MT functionality.
To demonstrate that insertion of full-length fluorescent proteins into H1-S2 loops disrupts the functionality of tubulins as predicted (S2B and S2C
Using the ciliated neuron-specific labeling, we observed robust distribution of endogenous TBB-2 within amphid and phasmid ciliary MTs (Figs 4B and S6A).Previous attempts using Nterminal GFP::TBB-2 KI allele had masked this phenomenon [22], likely due to background signals emanating from extraneous tissues (S6A Fig) .Furthermore, we found that ciliary TBB-2 was predominantly localized within the middle segments of cilia, with diminished signals in the distal segments, thereby revealing a unique and regulated distribution pattern distinct from that of TBB-4 distributing along the full-length cilia [12].These new findings were previously inaccessible with traditional labeling strategies.
The hypodermis-specific labeling facilitated visualizing dynamic architecture of MT network in the hypodermal cell layers, effectively eliminating signal interference from adjacent tissues.By imaging of hypodermal MTs throughout the nematode, including the head, pharynx, vulva, and tail regions (Figs 4C and S6B), we captured MT dynamics-encompassing polymerization (pol), depolymerization (depol), and translocation (trans) via kymographs and time-lapse videography (Fig 4D and 4E and S3-S5 Videos).
The germline-specific labeling enabled tracking of the temporal behavior of mitotic and meiotic MTs within germline as well as in early-stage embryos, excluding background noise emanating from other unrelated tissues (Figs 4F and S6C).In the mitotic germline, the self-renewing mitosis of individual germline stem cell (GSC) was unambiguously captured, spanning from metaphase to  To investigate potential effects of GFP11-i on MT dynamics, we quantified astral MT growth rates during 1-cell stage embryonic mitosis.Previous studies employing EB protein EBP-2::GFP as plus-end tip markers revealed astral MT growth rates of 0.72 ± 0.02 μm/s during metaphase and reduced to 0.51 ± 0.02 μm/s in anaphase [56].Using endogenous EBP-2:: mNeongreen as markers (S8A Fig), we confirmed the rates were 0.68 ± 0.08 μm/s during metaphase and reduced to 0.55 ± 0.06 μm/s (N = 70) during anaphase (S8B Fig) .By quantifying these profiles using TBB-2 (GFP11-i) in tbb-2 (gfp11-i) embryos (S8C Fig), the astral MT growth rates were 0.69 ± 0.09 μm/s during metaphase and reduced to 0.55 ± 0.05 μm/s (N = 70) in anaphase (S8D Fig).These rates were commensurate with those measured via EBP-2::mNeongreen, which suggested that GFP11-i labeling may not perturb intrinsic MT dynamics.Collectively, we conclude that GFP11-i offers superior endogenous labeling of both α-and β-tubulin isotypes, outperforming existing fluorescent protein-based labeling methodologies in C. elegans.

GFP11-i enables labeling of mammalian tubulins
To examine the applicability of GFP11-i labeling across multiple species, we tested this technique with the human α-tubulin isotype hTUBA1A, β-tubulin hTUBB8, and mouse α-tubulin TUBA4A (Fig 1D), all of which had been implicated in tubulinopathies [13].Equivalent quantities of CMV promotor-driven gfp-htuba1a or htuba1a (gfp11-i) constructs were transiently transfected into HeLa cell lines along with GFP1-10, individually (Fig 5A).Within GFP-positive cells, MTs labeled with GFP11-i exhibited significantly improved signal-to-noise ratio (mean MT / background fluorescence ratio = 3.3, N = 10 cells), nearly twice higher than that labeled with GFP (mean MT / background fluorescence ratio = 1.7,N = 10 cells) (Fig 5B and  5C).This improvement of the signal-to-noise ratio likely stemmed from the superior functionality of GFP11-i-labeled tubulins, which may facilitate more efficient assembly into MTs.Likewise, we effectively labeled β-tubulin hTUBB8 using GFP11-i tagging strategy (S8E Fig) .Importantly, the expression of GFP11-i-labeled hTUBA1A did not change the level of the αK40 acetylation, a posttranslational modification occurred within the lumen of MTs (S8F and S8G Fig) [57], which suggests that the property of the MT lumen may not be strongly affected by GFP11-i labeling.
Having demonstrated the effective labeling of wild-type tubulins using GFP11-i, we evaluated whether GFP11-i also allows for labeling of the human pathogenic tubulins with fidelity.The TUBA4A (E284G) de novo mutation was implicated in human infertility, and this pathogenic mutation could led to meiotic arrest in mouse oocytes in a dominant-negative manner  [28].The residue E284 locates in the M loop of α-tubulin, and, thus, this mutation could potentially attenuate polar lateral interaction with R121 or K124 in the H3 helix of neighboring α-tubulins (Fig 6A).To investigate the impact of E284G on MT stability, equivalent quantities of untagged, or GFP-tagged, or GFP11-i-tagged TUBA4A (E284G) were transfected into HeLa cell lines (Fig 6B).Expression of either untagged TUBA4A (E284G) or GFP11-i-tagged TUBA4A (E284G) disrupted cellular MT arrays (Fig 6B and 6C, GFP as cotransfection marker in untagged group), demonstrating the dominant-negative effect of this mutant.However, the similar expression of GFP-TUBA4A (E284G) failed to induce an apparent defect, likely due to its ineffective incorporation into MT network (Fig 6B and 6C).These findings underscored the fidelity of GFP11-i labeling for live-cell imaging of tubulin isotypes in pathological contexts.

Discussion
Our study presents advances in the functional fluorescence labeling of tubulins, addressing the long-standing challenge of preserving functionality while achieving robust labeling and realtime imaging.Utilizing the C. elegans model and AlphaFold2 pipeline, we successfully engineered an optimized GFP11-i construct that enabled both functional and fluorescent cytosol backgrounds.The entire meiotic spindle was calculated in each oocyte.N = 6 independent oocytes for each group.Numerical data for panels C and F are available in S1 Data.GFP, green fluorescent protein; GV, germinal vesicle; MT, microtubule.https://doi.org/10.1371/journal.pbio.3002615.g005characterization of endogenous tubulins.By expanding the utility of GFP11-i to different mammalian tubulin isotypes, we validated its applicability in both human cell models and murine embryonic systems.This underscores its considerable promise for understanding MT dynamics across various biological contexts, encompassing both normal physiology and tubulin mutation-associated pathologies.
Previous studies indicated that insertion of up to 17 aa (for instance, 6xHis tag) into the H1-S2 loop of tubulins did not disrupt the functionality of tubulins [34].Therefore, inserting other small tags into the H1-S2 loop would be also feasible for tubulin purification.However, it remained mysterious how a longer insertion tag would affect the functionality of tubulins.Our initial attempts employing 16-aa GFP11 tags yielded no fluorescence (S4F Fig) .It was only after harnessing the predictive capacity of AlphaFold2 that we were able to navigate these pitfalls by optimizing linker length for GFP11 incorporation (Fig 1A).This computational approach, previously untapped for tubulin engineering, provides additional insights into recombinant protein design and paves the way for similar strategies in labeling other structurally complex proteins.
In C. elegans, our RNAi assays and quantification of brood sizes substantiated the functionality of GFP11-i labeled TBB-2 (Fig 2A -2E).The data from GFP11-i starkly contrasted with embryonic lethality induced by conventional GFP tagging at either terminus, thereby affirming the pivotal advantage of GFP11-i in preserving protein functionality.The central thrust of our study established GFP11-i as a potent tool for the functional labeling of endogenous mutated α-tubulins without perturbing their molecular activities.In particular, we have investigated the structural and functional impact of GFP11-i labeling on the A19V mutant of the ciliary tubulin isotype TBA-5 in C. elegans.Despite structural sensitivity of TBA-5 (A19V) and its impairment of ciliary distal segments, GFP11-i labeling did not induce or exacerbate discernible morphological alterations, including the temperature-sensitive nature, highlighting its promise as an ideal labeling strategy for high-fidelity biological studies.Considering that pathogenic tubulins tend to lose their toxic effects on MTs when labeled with GFP (Fig 6 ), our developed method enables following the dynamic behavior of pathogenic tubulins in living organisms or cells while preserving their "toxicity" on MTs, thereby holding the promise to advance mechanistic understanding of ciliopathies and other tubulin-related diseases involving tubulin isotypes [13,15].
Our work has broader implications for the fields of cell biology and bioengineering.Beyond tubulins, our method can potentially be adapted to other proteins that are difficult to label without compromising their native functions.The success of our AlphaFold2-guided design provides a framework for leveraging computational tools in the de novo design of functional protein tags.It is tempting to speculate that similar pipelines could be employed to design novel probes for a multitude of applications, ranging from live-cell imaging to targeted therapeutics.
While our study offers promising insights, it also has some potential limitations.Whether GFP11-i in the MT lumen affects recruitment of intraluminal MAPs remains to be demonstrated [1,57].Although GFP11-i labeling did not compromise MT growth rates during embryonic mitosis in C. elegans (S8D Fig), additional parameters such as the catastrophe rates of MTs needs to be identified to further confirm the functionality of GFP11-i labeling.Moreover, a comparative study involving additional labeling techniques, isotypes, and species could further substantiate our conclusions.Future studies are required to explore generating cell lines or mouse models expressing GFP11-i labeled tubulins to investigate MT dynamics in various MT-based processes, such as mitosis or ciliogenesis.
In summary, our GFP11-i labeling technology represents an important advance in the pursuit of efficacious yet minimally perturbative labeling of endogenous proteins.It exhibits a higher signal-to-noise ratio and promotes MT assembly with remarkable efficiency.Importantly, the GFP11-i strategy enables the visualization of pathogenic tubulin dynamics with fidelity in living cells, thereby providing a substantial advantage over traditional GFP labeling strategies.The broad applicability, corroborated by its successful deployment in C. elegans as well as mammalian cells involving both wild-type and structural sensitive mutated tubulin isotypes, renders it suitable for investigations focused on elucidating the intricate dynamics of tubulin (and, potentially, other proteins) in both physiological and pathological contexts.

Worm strains, culture, and genetic cross
C. elegans were cultured at 20˚C on standard nematode growth medium (NGM) plates and seeded with the Escherichia coli OP50 unless described otherwise.The wild-type strain was Bristol N2.Some strains were provided by the Caenorhabditis Genetics Center (CGC), and other strains were created using the CRISPR-Cas9-mediated genome editing technology.Strain SHX2025 was a gift from Dr. Suhong Xu (Zhejiang University, P.R. China).Strain DUP228 was a gift from Dr. Wei Zou (Zhejiang University, P.R. China).S1 Table summarizes the strains used in this study.For genetic cross, 20 μL suspended OP50 was dropped at the center of NGM plates to make a cross plate.AroundAU : PleasenotethatasperPLOSstyle; numeralsarenotal 10 to 15 males carrying him-5 (e1490) allele and 5 hermaphrodites were transferred to this plate for at least 24 h.F1 and F2 hybrid progenies were singled and screened by PCR using EasyTaqAU : PleasenotethatPLOSdoesnotallowtrademark 2× Super Mix (TransGen Biotech, #AS111-14).

Molecular biology
For KI plasmids, 20 bp CRISPR-Cas9 targets (S2 Table ) were inserted into the pDD162 vector (Addgene #47549) by linearizing this vector with 15 bp overlapped primers.The resulting PCR products were treated with DpnI digestion overnight and then transformed into E. coli Trans5α.The linearized PCR products were cyclized by spontaneous recombination in bacteria.The homology recombination (HR) templates were constructed by cloning the 0.9 to 1.2 kb upstream and downstream homology arms into pPD95.77vector (Addgene #37465) using In-Fusion HD Cloning Kit (Takara Bio, #639650).Subsequently, fluorescence tag coding sequences were directly inserted into their respective positions in the plasmids.Ultimately, CRISPR-Cas9-targetted sites in the templates were modified with synonymous mutations.For OE plasmids, cDNA construct of tba-5, tba-5 (qj14), or gfp1-10 was cloned into pDONR vector containing 445 bp dyf-1 promoter and unc-54 3 0 UTR.cDNA construct of human tuba1a or gfp1-10 was cloned into pLV vector containing CMV enhancer and CMV promotor.cDNA construct of mouse tuba4a or gfp1-10 was cloned into pET27b vector downstream of T7 promotor.Fluorescence tag coding sequences were inserted into their respective positions in those plasmids above.All plasmids were purified with AxyPrep Plasmid Purification Miniprep Kit (Axygen, #AP-MN-P-250) and PureLink Quick PCR purification Kit (Invitrogen, #K310001).mMESSAGE mMACHINE T7 Ultra Kit (Life Tech, #AM1345) was used for in vitro transcription of mouse tuba4a or gfp1-10 mRNAs, and MEGAclear kit (Life Tech, #AM1908) was used for mRNA purification before microinjection.S3 Table summarizes the plasmids and primers used in this study.

Genome editing
CRISPR-Cas9-mediated genome editing was used to create KI strains [47].Target sequences were selected by the CRISPR design tool (https://zlab.bio/guide-design-resources).The CRISPR-Cas9 constructs with targets and HR templates were coinjected into the gonads of young adult worms at 50 ng/μL with the rol-6 (su1006) selection marker (50 ng/μL).F1 transgenic progenies (roller) were singled and screened by PCR.All KI alleles were verified by the Sanger sequencing of the entire genes to guarantee that no other mutations were introduced.SunyBiotech generated some KI strains and alleles in this study, the name of these strains started with "PHX."The name of these alleles started with "syb."

Transfection and immunofluorescence
HeLa cells were grown in DMEM (Gibco) containing 10% fetal bovine serum (ExCell Bio) and 1% penicillin and streptomycin (Yeasen) at 37˚C with 5% CO 2 .HeLa cells were seeded on 35 mm glass-bottom confocal dishes the day before transfection.In

Mouse oocyte collection and mRNA microinjection
The protocol for mouse oocyte collection was approved by the Institutional Animal Care and Use Committee and Internal Review Board of Tsinghua University.GV oocytes were collected from the ovary of 3 to 4 wk old C57BL/6 females 48 h after PMSG injection and then were incubated in M2 medium (Sigma, #M7167) supplemented with 10 μM Cilostamide (Target-Mol, #68550) at 37˚C with 5% CO 2 .tuba4a-gfp (or gfp-tuba4a, or tubb5-gfp) and h2b-mcherry mRNAs were coinjected at concentrations of 100 ng/μL and 50 ng/μL into the mouse GV oocytes using Eppendorf FemtoJet and a Leica microscope micromanipulator.For GFP11-i, tuba4a (gfp11-i), gfp1-10 and h2b-mcherry mRNAs were coinjected at concentrations of 100 ng/μL, 200 ng/μL, and 50 ng/μL.

Dye-filling assay
The fluorescence dye DiI (1,1 0 -dioctadecyl-3,3,3 0 ,3 0 -tetramethylindocarbocyanine perchlorate, Sigma) filling assay was broadly used to assess the ciliary function and integrity [52,58].Dyefilling positive animals possess relatively intact ciliary structures, while dye-filling defective animals develop abnormal ciliary structures.Worms with tba-5 (qj14) background were cultured at 15˚C for at least 24 h before this assay [12].Young adult worms were harvested into 500 μL M9 buffer with DiI (4 μg/ml), followed by incubation at 15˚C in the dark for 1 h.Worms were then transferred to seeded NGM plates and examined for dye uptake 2 h later using a fluorescence compound microscope.We observed at least 100 animals of each strain from 3 independent assays.

Rosetta relax runs
Structural model of tubulin within TRiC / CCT complex (7TUB) was downloaded from RCSB PDB database [44].CeTBA-5 (NP_492668.1),TBA-5 (GFP11-i), TBA-5 (GFP-i), and TBA-5 (Scarlet-i) were predicted by AlphaFold v2.2.0 to acquire their structures.These structures were aligned to template tubulins in 7TUB using the "Alignment" function of Pymol 2.5.5 and then substituted template tubulins.For each model generated after substitution, we run the Rosetta relax protocol multiple times to relieve the clashes between substituted tubulins and neighboring CCT subunits [46].CCT subunits were set to be rigid in each relax script.Relaxed structures with the highest scores assessed by Rosetta Energy Function 2015 (REF2015.wts)were analyzed and presented [61].

FRAP
FRAP experiments were carried out at 15˚C using Zeiss LSM900 with Airyscan2 confocal microscopy (Carl Zeiss).A 488-nm laser at 100% power was used for photobleaching, and images were acquired every 2 min.We measured fluorescence intensities at the ciliary tips before or after the bleach.The data were normalized to the fluorescence before the bleach.

RNAi
RNAi was carried out as in [62].Designed target sequence for tbb-1 could be found in [24]; this target was proved to inhibit expression of tbb-1 efficiently.Worms were fed with RNAi bacteria for at least one generation before analysis.To calculate embryonic lethality, 15 day 1 adult worms were transferred to freshly seeded RNAi plates to lay eggs for 1 to 2 h.Then, all adult worms were removed and the number of progenies was counted.Embryonic lethality was calculated by the ratio of unhatched progenies after 3 d to all progenies at the beginning.To calculate brood size, L4 worms were transferred singly to freshly seeded RNAi plates.Worms were repeatedly transferred to freshly seeded RNAi plates every 24 h until reproduction ceased.We counted the number of hatched progenies at their L4 stages.

Quantification and statistical analysis
The sample sizes in our experiments were determined from the related published analyses.All experiments were repeated at least 2 times from independent samples with identical or similar results.We used GraphPad Prism 9 (GraphPad Software) for statistical analyses.Quantification was represented by the mean value ± standard deviation for each group.Two-tailed Student t test analyses were performed to compare the mean values between 2 groups.N represents the number of samples in each group.Statistical significances were designated as ns P > 0.05, * P < 0.05, ** P < 0.01, *** P < 0.001, and **** P < 0.0001.
), which was further corroborated by predicted alignment error (PAE) plots (S1E Fig).These predictions were consistent with our livecell imaging results (S4F Fig), confirming the reliability of our structural predictions.Notably, tubulins tagged by GFP11 with GS-linker 3 possessed intact structures as well as untagged tubulins in the chaperones after sufficient relax runs by Rosetta (S2D Fig), implying the negligible effect of this tag on tubulin folding process.Consequently, a 16-aa GFP11 flanked by 16-aa GS-linker 3 (termed optimized GFP11-i), emerged as the most efficacious insertion tag for the functional labeling of tubulins (Fig 1B).Moreover, AlphaFold2 anticipated that GFP11-i possessed a broad labeling spectrum, effectively marking a broad array of tubulins explored in our studies, including the nematode β-tubulin TBB-2, human α-tubulin TUBA1A, and mouse α-tubulin TUBA4A (Fig 1C).A schematic of AlphaFold2 pipeline-guided design of functional protein labeling was shown in Fig 1D.

Fig 5 .
Fig 5. GFP11-i labeling of mammalian tubulins.(A) Workflow of transgenic expression of GFP or GFP11-i-tagged human α-tubulin TUBA1A in HeLa cells.The total quantity of pLV plasmids for transfection was 0.5 μg in each group.GFP11-i alone served as a negative control.(B) Representative images for transfection-positive cells in each group.Hoechst (blue) marked the nuclei of live cells.Images were not z-stacked.Scale bar, 10 μm.(C) GFP fluorescence ratio of MTs to their adjacent cytosolic backgrounds.The mean value of 5 independent MTs was calculated in each cell.N = 10 independent cells for each group.(D) Workflow of mouse α-tubulin TUBA4A mRNA microinjection into GV-stage mouse oocytes.Oocytes at metaphase I were analyzed.(E) Representative images for oocytes in metaphase I in each group.Edges of cells were depicted by gray dashed curves.Scale bar, 10 μm.(F) GFP fluorescence ratio of spindle MTs to