A split protease-E. coli ClpXP system quantifies protein–protein interactions in Escherichia coli cells

Characterizing protein–protein interactions (PPIs) is an effective method to help explore protein function. Here, through integrating a newly identified split human Rhinovirus 3 C (HRV 3 C) protease, super-folder GFP (sfGFP), and ClpXP-SsrA protein degradation machinery, we developed a fluorescence-assisted single-cell methodology (split protease-E. coli ClpXP (SPEC)) to explore protein–protein interactions for both eukaryotic and prokaryotic species in E. coli cells. We firstly identified a highly efficient split HRV 3 C protease with high re-assembly ability and then incorporated it into the SPEC method. The SPEC method could convert the cellular protein-protein interaction to quantitative fluorescence signals through a split HRV 3 C protease-mediated proteolytic reaction with high efficiency and broad temperature adaptability. Using SPEC method, we explored the interactions among effectors of representative type I-E and I-F CRISPR/Cas complexes, which combining with subsequent studies of Cas3 mutations conferred further understanding of the functions and structures of CRISPR/Cas complexes.

P roteins serve essential functions in nearly all biological processes in living organisms. It has been revealed that over 80% of proteins exist in complexes 1 , suggesting that protein is normally not carrying out its function alone. Therefore, characterizing the protein-protein interactions (PPIs) could help reveal the functions of the proteins, and also prompt the understanding of biological processes which they are involved in. Many methods have been developed for characterizing the PPIs, including bacteria/yeast two-hybrid (B2H/Y2H) 2,3 , affinity-based techniques 4,5 , quantitative proteomics 6 , and split protein complementation assays 7 . For example, Y2H has developed 30 years ago, which has been used to confirm suspected interactions, identify protein-drug interaction, and discover inhibitors of the PPIs 8 . In addition, quantitative proteomics is a newly developed method, which was recently used to map receptor-ligand interactions in the activation mechanism of class B G-protein-coupled receptors (GPCRs) upon interaction with peptide hormones in signal-transduction pathways of mammalian cells and elucidate the protein interactome in chromatin biology 9 .
Besides these methods, split-protein assays, also termed protein fragment complementation assays (PCAs), are another set of methods for characterizing PPIs, including split-GFP 10 , splitluciferase 11 , split-ubiquitin 12 , and split-TEV protease 13 et al. PCAs have been widely applied in both prokaryotic and eukaryotic species to characterize transient and irreversible PPIs, and manipulate signal pathways, which consist of modular elements to prompt the system flexibility and functionality under in vivo environment. For example, split TEV protease assay was used to monitor the neuregulin-induced ErbB2/ErbB4 receptor tyrosine kinase heterodimerization 13 , and GPCR activities caused by druginduced changes 14 . Although PCAs has been proven to be a robust method for explore PPIs, designing functional split proteins is a very challenging task, which needs to maintain the stability as well as the functionality of the split proteins to ensure the recovered biological activity of the reassembled proteins 15 . Many split proteins have been identified so far, among which split protease has the advantages of signal amplification and clean background because of its high catalytic activity and substrate specificity. Small amount of reassembled proteases could cleave large amount of its substrates, thus generating enhanced signals to increase the systematic sensitivity.
E. coli is the dominant model species in biological research due to its easy manipulation and clear genetic background. Recently, engineered E. coli was also used for eukaryotic protein studies 16,17 , and the good adaption to low temperature further favors its use in academia and industry. Therefore, it is attractive to develop a sensitive PCA for characterizing PPIs in E. coli within a broad temperature range. Through combining the ClpXP-SsrA protein degradation machinery 18 , sfGFP fluorescence reporter, we developed a split human Rhinovirus (HRV) 3 C protease mediated PCA in E. coli cells, termed as Split Protease-E. coli ClpXP (SPEC) system. Our studies showed that splitting at K82 position of HRV 3 C protease could generate a high effective split HRV 3 C protease, and its incorporation to the SPEC system can make it characterize PPIs in E. coli quantitatively with high sensitivity, easy manipulation, and broad temperature adaptability from 18 to 37°C. Using SPEC method, we further analyzed the interactions of effects of type I-E and I-F CRISPR/Cas to understand its possible complex structure and functions.

Results
Development of SPEC system. The SPEC system consists of two sub-parts, the ClpXP-SsrA protein degradation machinery mediated sfGFP fluorescence reporting sub-system and the split-HRV 3 C protease-mediated protein interacting sub-system (Fig. 1a). In E. coli, ClpXP protease can recognize the SsrA sequence (AAN-DENYALAA), thus causing the degradation of SsrA-tagged protein in cells 18 . When the prey protein and bait protein do not interact with each other, sfGFP-SsrA is degraded through the ClpXP-SsrA mediated protein degradation pathway accompanied with a low cellular sfGFP fluorescence intensity. However, when the prey protein and bait protein interact with each other, their fused Nterminal and C-terminal split HRV 3 C proteases can re-assemble to a functional HRV 3 C protease, cleaving the SsrA peptide off sfGFP-SsrA to cause an increased sfGFP fluorescence intensity.
We first validated the principle of SPEC system in E. coli BL21 (DE3) by the designed SPEC-VA and SPEC-VB constructs ( Supplementary Fig. 1a), which express the sfGFP-LEVLFQGP-SsrA polypeptide and intact HRV 3 C protease under the control of T7 and lac promoters, respectively. Our results showed that cells presented almost no sfGFP fluorescence when bearing the SPEC-VA vector only, while exhibited strong recovered sfGFP fluorescence when carrying both SPEC-VA and SPEC-VB vectors (Fig. 1b). Subsequently, we compared the functionality of Tobacco Etch Virus (TEV) and human Rhinovirus 3 C (HRV 3 C) proteases in SPEC system. Both are virus-derived proteases possessing high proteolytic activity and substrate specificity, with HRV 3 C protease exhibiting better proteolytic efficiency and toleration to low temperature 19 . Four common temperatures (37°C, 30°C, 25°C, and 18°C) were firstly used for system evaluation, because they are normally used for heterologous protein expression in E. coli. Our results showed that HRV 3 C protease presented up to 4 folds stronger sfGFP fluorescence intensity than using TEV protease, especially at low temperatures (Fig. 1c). With HRV 3 C protease, the highest levels of sfGFP fluorescence intensity in E. coli at 37°C, 30°C, 25°C, and 18°C were 2.0 ± 0.1 × 10E4, 1.8 ± 0.1 × 10E4, 1.6 ± 0.1 × 10E4 and 1.7 ± 0.1 × 10E4, which was obtained after 4 h, 6 h, 10 h and 20 h IPTG induction, respectively.
Identification of an optimized split HRV 3 C protease in SPEC system. Next, we identified the split HRV 3 C protease modules for SPEC system (Fig. 2a, b). The solved HRV 3 C protease structure indicated that it was constituted of two characteristic six-stranded β-barrel subdomains, connected by a random coil 20 . This structure is similar to the structure of TEV protease, which can also be dissected into two characteristic N-terminal and C-terminal subdomains. The reported split TEV protease modules have the split sites located on the random coil connected N-terminal and Cterminal subdomains 13 . Therefore, K82, L94, and N107, which were located on the random coil and the following β-sheet of the structure of HRV 3 C protease, were selected for generating the split HRV 3 C protease modules. At the same time, two protein pairs, Cas1/Cas2-3 from Type I-F CRISPR/Cas complex of prokaryotic Zymomonas mobilis 21 and Yae1/Lto1 from Fe-S protein assembly (CIA) machinery of eukaryotic Saccharomyces cerevisiae 22 , were used to evaluate the effectiveness of split HRV 3 C protease and generality of the SPEC system (Fig. 2b, Supplementary Fig. 1b, Supplementary Fig. 1c). All three split-sites could prompt a reassembled functional HRV 3 C protease, with K82 being the best. A total of 85-97% cells expressing split HRV 3 C (K82) protease showed sfGFP fluorescence, which is close to the 98% cells expressing the HRV 3 C protease (Fig. 2b). In comparison, onlỹ 40% activity of the TEV protease could be recovered with the reported split TEV protease in a previous report 13 . Moreover, split HRV 3 C (K82) protease outperformed split TEV protease in SPEC system under different temperatures (Fig. 2c, d). It is also worth noting that the highest level of cellular sfGFP fluorescence intensity with re-assembled split HRV 3 C (K82) protease at 30°C, 25°C, or 18°C, was more than 4 folds higher than that of the re-assembled split TEV protease (Fig. 2c, d). coli ClpXP (SPEC) system. sfGFP is fused with an HRV 3 C protease cleavage sequence (LEVLFQ↓GP) at its C-terminus followed by a SsrA sequence (AANDENYALAA), forming a sfGFP-LEVLFQGP-SsrA cassette. The prey protein and bait protein are fused to the N-terminal and C-terminal split-HRV 3 C proteases, respectively. Through ClpXP-SsrA degradation pathway, highly expressed sfGFP-LEVLFQGP-SsrA cassette is fast degraded in E. coli cells, thus very low green fluorescence is detected. Nevertheless, when prey protein interacts with bait protein, the N-terminal split and C-terminal split-HRV 3 C proteases will re-assemble to a functional HRV 3 C protease, which will cleave off the SsrA peptide from the sfGFP-LEVLFQ↓GP-SsrA cassette, leading to an accumulation of sfGFP for increased fluorescence intensity. b Validation of the ClpXP-SsrA protein degradation machinery mediated sfGFP degradation in E. coli. The total cellular sfGFP fluorescence intensity (sfGFP F. I.) of cells bearing different plasmids were quantitated by flow cytometry (detected with FITC channel, 525/40 nm BP Construction of the SPEC system in the recombinant E. coli BL21 (DE3)-SPEC cells. To simplify the manipulation of SPEC system and increase its stability in E. coli, we used the modified no-SCAR (Scarless Cas9 Assisted Recombineering) technology 23 to integrate the sfGFP-LEVLFQGP-SsrA into E. coli chromosome followed by temperature sensitive plasmid curing 24 , generating the recombinant E. coli BL21(DE3)-SPEC strain (Fig. 3a). Additionally, the prey protein fused N-terminal split HRV 3 C (K82) protease and bait protein fused C-terminal split HRV 3 C (K82) protease were separated into SPEV-VD (Kan R ) and SPEC-VE (Amp R ) vectors, respectively ( Supplementary Fig. 1d). Our results showed a very low sfGFP background fluorescence in BL21(DE3)-SPEC cells, while cells bearing the SPEC-VB vector exhibited a high sfGFP fluorescence intensity (Fig. 3b). Similar high sfGFP fluorescence intensity was also obtained in BL21 (DE3)-SPEC bearing both SPEC-VD and SPEC-VE vectors, validating the effectiveness of this system constituting of BL21 (DE3)-SPEC strain and SPEC-VD/SPEC-VE vectors. Subsequently, we evaluated this final SPEC system using Cas1/Cas2-3 and Yae1/Lto1 protein pairs in BL21(DE3)-SPEC at different temperatures (Fig. 3c). Similar tendencies were obtained that split HRV 3 C (K82) protease conferred a higher sensitivity than that of split TEV protease, especially under low temperatures (Fig. 2c, d).  2 Identification of a highly efficient split HRV 3 C protease and its comparison with split TEV protease in the SPEC system. a Designing split HRV 3 C protease based on its structure (PDB: 2B0F) with three split positions (K82, L94, N107). Blue: N-terminal HRV 3 C protease, Red: C-terminal HRV 3 C protease, Purple: the random coil between N-terminal and C-terminal HRV 3 C protease, Cyan: the designed split sites. b Characterizing different split HRV 3 C proteases in SPEC system using pSPEC-N107/L94/K82-PP and pSPEC-N107/L94/K82-EP vectors, respectively. Cells bearing pSPEC-VB vector containing full length HRV 3 C protease was used as a positive control (right panel), and cells bearing only pSPEC-VA vector was used as a negative control (Fig. 1b). The total cellular sfGFP F. I. of cells bearing different vectors (pSPEC-K82-EP to pSPEC-N107-PP, Supplementary Table 1) were quantitated through flow cytometry. Upper-left panel, representative flow cytometry histograms of the total cellular sfGFP F.I. of cells expressing different split HRV 3 C proteases fused with Z. mobilis Cas1/Cas2-3 protein pair. Lower-left panel, representative flow cytometry histograms of the total cellular sfGFP F.I. of cells expressing different split HRV 3 C proteases fused with S. cerevisiae Yae1/Lto1 protein pair. The experiments were performed at 25°C with 8 h induction. c Comparison of split HRV 3 C (K82) protease (Wine line, pSPEC-VC1) with split TEV protease (Dark blue line, pSPEC-VC5) using Cas1/Cas2-3 protein pair in SPEC system at 37°C, 30°C, 25°C, and 18°C, respectively. d Comparison of split HRV 3 C (K82) protease (Yellow line, pSPEC-VC2) with split TEV protease (Cyan line, pSPEC-VC6) using Yae1/Lto1 protein pair in SPEC system at 37°C, 30°C, 25°C, and 18°C, respectively. All the vector information can be found in Supplementary Table 1. Data are presented as mean ± SEM (n = 3 independent experiments) with Student's t test being performed, # P > 0.05, *P ≤ 0.05.
To further understand the difference between type I-E and I-F CRISPR/Cas complexes, we profiled the interactions of Cas3 mutants 30 with all other effectors (Fig. 4c, d). Surprisingly, the Cas3 mutants presented different effects in these two complexes. In type I-E CRISPR/Cas complex, most Cas3 mutants increased its interaction with Cas2 and Cas6, while mutations altering Mg 2+ -dependent HD-nuclease activity (H74A, D75A, K78A) and ATP-dependent helicase activity (K320N, K452N, S483A/T485A) either decreased or increased its interaction with Cas1, respectively. In comparison, most Cas3 mutants did not affects the interaction of Cas2-3 with other effectors in type I-F CRISPR/Cas complex. However, K127A significantly decreased the interaction of Cas2-3 with Csy1 (85%) and Csy2 (75%), while other mutants largely increased its interaction with Csy1 (>40%). These results indicated different roles of Cas3 in type I-E and I-F CRISPR/Cas systems, in which Cas3 might possess a more stable structure due to its fusion with Cas2 effector. Besides, this difference on effector interaction might lead to the different functional mechanisms between these two CRISPR/Cas systems.

Discussion
In our studies here, a designed split HRV 3 C protease was utilized in the SPEC system for quantitating protein-protein interaction from both prokaryotic and eukaryotic species. Compared to the widely used split TEV protease 13,31 , HRV 3 C protease is catalytically faster, thus producing further enhanced sfGFP signals in SPEC system (Fig. 1c). More importantly, HRV 3 C protease exhibit constantly high proteolytic activity in a broad temperature range from 18°C to 37°C (Fig. 1c). This property is very important because many heterologous proteins could only be expressed and folded well at low temperature in E. coli cells. Besides the common temperatures for E. coli cell growth, we also evaluated the SPEC system at 42°C and 12°C. Similar results were obtained that split HRV 3 C (K82) protease presented higher sensitivity than that of split TEV protease at 12°C, while their difference is not obvious at 42°C (Supplementary Fig. 4). These results further confirmed that split HRV 3 C (K82) protease possesses high sensitivity in a broad temperature range, which is better than the split TEV protease, especially at low temperature. Besides the advantageous adaption to low temperature, it is also noticed that the reassembled split HRV 3 C proteases presented almost the identical proteolytical activity to the intact HRV 3 C protease (Fig. 2b), while it was reported that reassembled split Fig. 4 Characterization of protein-protein interactions of effectors in the Type I-E and I-F CRISPR/Cas complex. a Protein-protein interactions among eight effectors (Cas1, Cas2, Cas3, Cas8, Cas11, Cas7, Cas5, and Cas6) of E. coli K12 Type I-E CRISPR/Cas complex was interpreted using the circular visualization R program 44 . b Protein-protein interactions among six effectors (Cas1, Cas2-3, Csy1, Csy2, Csy3, and Csy4) of Z. mobilis Type I-F CRISPR/Cas complex was interpreted using the circular visualization R program 44 . c The histogram of protein-protein interaction between Cas3 mutants (H74A, D75A, K78A, K320N, D452N, and S483A/T485A, respectively) and seven other effectors (Cas1, Cas2, Cas5, Cas6, Cas7, Cas8, and Cas11) of E. coli K12 Type I-E CRISPR/Cas complex. Cas1/GST was used as a negative control ( Supplementary Fig. 2b), and Cas1/Cas2 was used as a positive control. Percentage numbers presented correspondent interaction intensities of protein pairs, which were normalized by the interaction between Cas3 and correspondent effectors ( Supplementary Fig. 2b). d The histogram of protein-protein interaction between Cas2-3 mutants (H123A, D124A, K127A, K458N, D608N, and S639A/T641A, respectively) and five other effectors (Cas1, Csy1, Csy2, Csy3, and Csy4) of Z. mobilis Type I-F CRISPR/Cas complex. Cas1/GST was used as a negative control ( Supplementary Fig. 3b), and Cas1/Cas2-3 was used as a positive control. Percentage numbers presented correspondent interaction intensities of protein pairs, which were normalized by the interaction between Cas2-3 and correspondent effectors ( Supplementary Fig. 3b). All the vector information can be found in Supplementary Table 1. Data are presented as mean ± SEM (n = 3 independent experiments).
TEV protease only recovers its 40% proteolytic activity 13 . The high activity and low-temperature adaptability of split HRV 3 C (K82) protease make it more efficient in SPEC system. SPEC systems have some unique advantages. Comparing to split GFP method, the split HRV 3 C protease can amplify the signals of PPIs, thus detecting the weak interactions. Some other PCAs can also amplify the signals of PPIs, such as split luciferase assay 32 . However, SPEC system has another advantage that it quantitates the PPIs through fluorescent intensity of cellular sfGFP, not requiring additional exogenous substrates. It also needs to point out that SPEC system also has limitations. Since SPEC system is an E. coli cell-based assay, it requires that both the prey and bait proteins can be expressed and functionally folded in E. coli cells. Therefore, it may not be suitable for complex eukaryotic proteins that cannot be folded in E. coli cells.
To stabilize the production of sfGFP in E. coli cells, we integrated the sfGFP gene into the E. coli genome in SPEC, which could minimize the heterogeneity caused by episomal plasmid (Fig. 3). The modular elements make the SPEC system highly flexible. Using other reporters, SPEC system may be performed with preferred signal readout. Additionally, owing to the sensitivity and flexibility of split HRV3C protease, we believe that it could be used in the future to design more useful biosensors, in particular for high-throughput applications requiring signal enhancement. In addition, SPEC system uses a two-plasmid strategy in the recombinant E. coli BL21(DE3)-SPEC strain. The separation of prey and bait proteins in different plasmids not only significantly simplifies the manipulation, but also facilitates the analysis of large-scale PPIs. Combined with the flow cytometry and high-throughput sequencing technology, it could be believed that SPEC system can be extended to characterize PPIs in a library against library set.
Combining our PPIs results with the previous reported structures of Cas1 4 -Cas2-3 2 complex 29 and resembled type I-E CRISPR/Cas complex 26 , we simulated the model structure scheme of type I-F CRISPR/Cas complex (Fig. 5). Our results showed that Cas2-3 presented strong interactions with all other effectors, and Cas1 presented a very strong interaction with Cas2-3, Csy2, and Csy4, but no interaction of Csy1 and very low interaction with Cys3 (Fig. 4b, Supplementary Fig. 3). We speculated that Cas1 effector is located outside of the whole complex, wrapping up the Cas2-3 effectors, with Csy1 and Csy3 effectors being located at the other side of the whole complex. This unique structure might be caused by the requirement of the "flexible" PAM recognition function of Csy1 33 , and the binding of the spacer region by Csy3 in type I-F CRISPR systems 34 . Additionally, our results also confirmed that Cas2 might function as adaption protein and the helicase and nuclease functions of the Cas3 domain were involved in interference process 35 .
The simulated type I-F CRISPR/Cas complex structure (Fig. 5) exhibits high similarity with that of the type I-E cascade complex 26 . Interestingly, Cas2 and Cas3 only presented an interaction of 56%, while this interaction was significantly enhanced to 90% and 86% when co-expression of Cas1 and Cas8 (Fig. 4a, Supplementary Fig. 2), respectively. This might explain the characteristically fused Cas2-Cas3 domain in type I-F CRISPR/Cas complex, because it lacks Cas8 effector. Cas3 is the catalytic effector in type I CRISPR/Cas systems, and the fused Cas2-Cas3 domain is believed to maintain a more stable function of Cas3 in type I-F CRISPR/Cas complex, which is further supported in our studies of Cas3 mutants in both type I-E and I-F CRISPR/Cas complexes (Fig. 4c, d). Comparing to the Cas3 mutants in type I-E CRISPR/Cas complex, which exhibited promiscuous effect, the correspondent mutations in type I-F CRISPR/Cas complex all presented similar interactions with all other effectors, except the nuclease activity related Cas2-3 (K127A) mutant presented obviously decreased interaction with Csy1/Csy2 (from 63% to 10%, 88% to 21%, respectively). Besides, Cas2-3(H123A, D124A, K458N, and S639A/T641A) presented a significantly enhanced interaction with Csy1. All these Cas3 related mutations affect the interactions of Cas2-Cas3 with Csy1/ Csy2 effectors, which were located at N-terminus of crRNA-Cascade complex, and speculated to function as blocking nuclease recruitment, crRNA and target binding 36 .
In principle, any protein pairs that are able to be functionally expressed in E. coli cells at the temperature range from 12°C and 42°C. can be characterized in SPEC system. Considering the exist of similar ClpXP-SsrA mediated protein degradation machinery

Methods
Bacterial strains and reagents. E. coli XL10-Gold strain (Stratagene, San Diego, USA) was used for plasmid construction, and E. coli BL21(DE3) strain was used for protein expression and flow cytometry analysis. Culture media, chemicals, and restriction enzymes were purchased from BD Biosciences (San Jose, CA, USA), Sigma-Aldrich (St. Louis, MO, USA), and New England Biolabs (Ipswich, MA, USA), respectively.
Plasmid construction. pET28a and pUC19 plasmids were used as backbone to construct pSPEC plasmids in this study (Supplementary Fig. 1). Total of 64 constructs were generated and their vector information can be found in Supplementary  Table 1. The split HRV 3 C protease domain (N HRV 3 C or C HRV 3 C) was fused with different protein pairs (prey and bait proteins), forming a prey-N HRV 3 C protease or bait-C HRV 3 C protease cassette. HRV 3C protease substrate sequence was flanked with sfGFP and SsrA tag (AANDENYALAA) sequences, forming the sfGFP-LEVLFQGP-AANDENYALAA cassette. A well-characterized Tobacco Etch Virus (TEV) protease variant containing a S219P mutation with enhanced stability 38 was used and annotated as TEV Protease in our research. Similarly, TEV protease substrate sequence (ENLYFQS) was also flanked with sfGFP and SsrA tag sequences, forming the sfGFP-ENLYFQS-AANDENYALAA cassette. These cassettes were cloned into different pSPEC plasmids.
Plasmids for this study was constructed using T5 exonuclease DNA assembly (TEDA) assay with minor modification 39 . Firstly, 1 mL 5 × TEDA solution contained 0.5 M Tris-HCl pH7.5, 50 mM dithiothreitol, 0.25 g of PEG-8000, and 1 μL of 10 U/μL T5 exonuclease (New England Biolabs, USA) was prepared for cloning. For the experiment, 4 μL assembly mixture aliquot was thawed on ice, followed by adding 16 μL of DNA solution containing the linear vector and target inserts. The molar ratio of linearized vector to target insert was set as 1:3, with vector amount approximately 50 to 200 ng. Reactions were then performed at 30°C for 40 min, followed by immediate transformation into E. coli competent cells.
Knock-in of sfGFP-LEVLFQGP-SsrA fragment into the genome of E. coli. E. coli BL21(DE3) (lpxM:: sfGFP-LEVLFQGP-SsrA) strain, named as BL21 (DE3)-SPEC, was generated using CRISPR/Cas9 assisted λ-Red recombinase system 40 followed by temperature sensitive plasmid curing. Cas9 and gRNA were incorporated into pKD46, a temperature-sensitive vector 41 carrying λ-Red genes (γ, β, exo), which promotes recombination, to construct pKD46-cas9-gRNA plasmid ( Supplementary  Fig. 1). Subsequently, pKD46-cas9-gRNA vector DNAs targeting the lpxM gene in E. coli genome were transformed into E. coli BL21(DE3) followed by cultivation and induction with 0.2% L-arabinose in LB media containing 100 μg/mL ampicillin at 30°C to an OD600 of approximately 0.6. The electrocompetent cells were then made by centrifugation, followed by washing twice with ice-cold deionized water. Meanwhile, the donor fragment (sfGFP-LEVLFQGP-SsrA) was gel-purified, and 10 ng of donor DNA and 50 μL of electrocompetent cells were then mixed on ice for electroporation. One mL LB media were then added into the post-shocked cells immediately following by recovering at 37°C for 1 h, and then spreading on LB plates with chloramphenicol. E. coli transformants carrying sfGFP-LEVLFQGP-SsrA fragments in the genome were screened and selected by colony PCR and Sanger sequencing. The validated transformants were then grown in LB media at 42°C for 12 h to eliminate the temperature-sensitive pKD46-cas9-gRNA-donor plasmid, generating BL21(DE3)-SPEC strain for further experiments.
Protein-Protein Interaction analysis with CytoFLEX by SPEC assay. The pSPEC constructs were transformed into E. coli BL21(DE3) or BL21(DE3)-SPEC cells followed by cultivating at 37°C until OD 600 reached 0.8. The cells were then induced by 0.5 mM IPTG (final concentrations) at different temperatures ranging from 18°C to 37°C.
Two hundred µL induced cells were collected by centrifugation at 3,000 × g at 4°C for 5 min and then resuspended into 400 µL 1 × PBS buffer (200 mM NaCl, 66 mM Na 2 HPO 4 , 34 mM NaH 2 PO 4 . 2H 2 O, pH 7.4), followed by sfGFP fluorescent analysis using Beckman Coulter CytoFLEX Flow Cytometer (Beckman Coulter, USA) equipped with a 488 nm laser and a 525/40 nm band-pass filter. sfGFP fluorescent intensity (sfGFP F. I.) of the cells was measured by the mean fluorescence intensity of the fluorescent cells.
The protein-protein interaction (PPI) intensity of cells expressing different targeting protein pairs were normalized using the mean fluorescence intensity of the cells expressing wt HRV 3 C protease as a control. The normalized interaction intensity was calculated as: Structure modeling and circular visualization analysis for PPIs data. The structure modeling of six proteins (Cas1, Cas2-3, Csy1, Csy2, Csy3, and Csy4) from Z. mobilis Type I-F CRISPR/Cas complex were performed by using the I-TASSER on-line server 42 respectively. The docking analyses of type I-F CRISPR/Cas complex were performed by using the ZDOCK program in BIOVIA Discovery Studio client 2020 43 . The circular visualization map of protein-protein interactions among six proteins (Cas1, Cas2-3, Csy1, Csy2, Csy3, and Csy4) from Z. mobilis Type I-F CRISPR/Cas complex was generated by using circlize package in R 44 .
Statistics and reproducibility. sfGFP fluorescent analysis and the protein-protein interaction (PPI) intensity of cells assays were independently replicated three times and individual data points are reported for each experiment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The source data for the graphs in the main figures is available as Supplementary Data 1. The vectors in Supplementary Figure 1 have been deposited in to Addgene with ID numbers 171974-171989. The datasets generated and/or analyzed during the current study are available from the authors on reasonable request.