The CRISPRaTOR: a biomolecular circuit for Automatic Gene Regulation in Mammalian Cells with CRISPR technology

gene expression; mammalian; CRISPR-Cas; biomedical engineering; biomolecular circuit; control engineering. .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint

1 2 We introduce the CRISPRaTOR, a biomolecular circuit for precise control of gene expression 3 in mammalian cells. The CRISPRaTOR leverages the stochiometric interaction between the 4 artificial transcription factor VPR-dCas9, and the anti-CRISPR protein AcrIIA4, enhanced with 5 synthetic coiled -coil domains to boost their interaction, to maintain the expression of a 6 reporter protein constant across diverse experimental conditions, including fluctuations in 7 protein degradation rates and plasmid concentrations, by automatically adjusting its mRNA 8 level. This capability, known as Robust Perfect Adaptation (RPA), is crucial for the stable 9 functioning of biological systems and has wide -ranging implications for biotechnological 10 applications. The CRISPRaTOR belongs to a class of biomolecular circuits named antithetic 11 integral controllers, and it can be easily adapted to regulate any endogenous transcription 12 factor thanks to the versatility of CRISPR-Cas system. Finally, we show that RPA holds also in 13 cells genomically integrated with the CRISPRaTOR, thus paving the way for practical 14 applications in biotechnology that require stable cell lines. 15 .CC-BY-ND 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity.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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint

16 Robustness is a fundamental property that enables biological systems to maintain stability and 17 functionality despite fluctuations in biochemical reaction rates caused by environmental 18 perturbations. Robust Perfect Adaptation (RPA) is a specific form of robustness in which a 19 biological system returns to a baseline level of function despite external fluctuations, 20 providing a mechanism for homeostasis 1. RPA mechanisms are often found in cellular 21 signalling pathways, where they help maintain a consistent response to signals despite 22 changes in signal intensity. Well-studied examples of RPA in biomolecular processes are 23 calcium homeostasis within mammalian cells 2 and the chemotaxis in E. Coli 3,4, allowing 24 organisms to adapt to changing concentrations of nutrients, toxins, or hormones. 25 Control Engineering is a well -established discipline to build “controllers” to regulate the 26 behavior of a physical system by keeping its output constant across a range of operating 27 conditions, by dynamically adjusting the input. In the context of biological processes, the input 28 can be any molecular species (e.g. a small molecule, a metabolite, etc.) whose changes have 29 a measurable effect on the output of the biological process (e.g. a protein of interest). 30 A key theoretical result of Control Engineering is that negative feedback can endow systems 31 with robustness to perturbations and uncertainties 5. Specifically, a negative -feedback 32 controller relies on a sense and react paradigm, where the output is actively measured and 33 compared against a reference value. Depending on the difference between the two (control 34 error), the controller will dynamically adjust the input to minimize the control error. It can be 35 demonstrated that, under specific conditions, if the magnitude of the input is proportional to 36 the sum of the error over time (integral control), then the system will exhibit RPA6,7. 37 Thanks to recent advances in molecular biology and biomolecular control theory, building a 38 biomolecular integral controller to robustly regulate gene expression at a constant level has 39 now become feasible. A biomolecular implementation of an integral controller is shown in 40 Figure 1, and it has been named the Antithetic Integral Controller (AIC) 8. It consists of a n 41 activator (X), which drives the expression of the “output” species (Z), and an inhibitor (Y), 42 which stoichiometrically binds X in a one -to-one fashion and inactivates it. In this 43 configuration, if a decrease in Z occurs, because of an external perturbation, it will cause a 44 decrease in Y and thus free up more X to increase the level of Z back to its initial level. Similarly, 45 an increase in Z will indirectly decrease X via Y and thus reestablish the equilibrium. 46 The AIC has been experimentally implemented in bacteria using sigma /anti-sigma factors in 47 bacteria9, and more recently in mammalian cells by means of a pair of sense and antisense 48 mRNA10, or by protein splicing with inteins11. These implementations albeit successful, have 49 some limitations in mammalian systems: the use of sense -antisense RNA pairs could 50 potentially trigger toxicity because of the cell's innate immune response, as double-stranded 51 RNAs are generated during viral replication 12; the protein splicing approach, despite being 52 potentially compatible to any protein of interest, requires extensive engineering to ensure 53 correct inteins splicing and protein folding. Further more, these AIC implementations have 54 been tested by transient transfection, where plasmid molar ratio can be strictly controlled. 55 However, engineering cells for practical application would require stable integration of gene 56 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint circuits in the cells’ own genome, and proof that performances would still be the same are 57 lacking. 58 Here, we leveraged the proteins of the CRISPR/Cas family to implement an AIC in mammalian 59 cells. Specifically, we made use of the artificial transcription factor VPR -dCas913 and the 60 recently discovered anti -CRISPR protein AcrIIA4 14,15, augmented with synthetic coiled-coil 61 domains to enhance their binding affinity 16,17, to implement an AIC in mammalian cells. We 62 demonstrate its ability to confer RPA in both transient transfection and stable genomic 63 integration. The use of VPR-dCas9 makes our implementation very versatile, as it can be used 64 as a “plug -and-play” circuit to control any endogenous transcription factor of interest by 65 simply designing an appropriate guide RNA (gRNA). We named our implementation of the 66 AIC, the CRISPRaTOR. 67 Figure 1 – The Antithetic Integral Controller and Robust Perfect Adaptation. A) The Antithetic Integral Controller is a negative feedback loop (Closed Loop) where a constitutively expressed activator species X drives the expression of a species of interest Z (output). Z drives the expression of an inhibitor species Y , which binds and inhibits X. When the concentration of Z changes, so does Y thus causing X to change in an opposite manner to Z (e.g. if the concentration of Z decreases, active X will increase, and vice versa). This mechanism enables the Antithetic Integral Controller to dynamically adjust the concentration of Z (solid orange line) in the face of a perturbation (red line) and thus maintain Z constant over time. In the Open Loop configuration, Z is directly expressed from a constitutive promoter, and if its concentration decreases because of an external perturbation (red line), its concentration would not be constant over time (dashed orange line). B)The species Z i n our implementation i s itself a transcriptional activator and its concentration can be experimentally tracked over time either indirectly by placing the luminescence firefly Luciferase ( Fluc) under a promoter driven by Z, or directly by fusing the EGFP fluorescent reporter to Z itself. .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint

and Discussion 68 Experimental implementation of an Antithetic Integral 69 Controller (AIC) by means of the CRISPR-antiCRISPR system. 70 71 The experimental implementation of the AIC is shown in Figure 2. The nuclease-deficient Cas9 72 fused to the transactivation domains VP64, p65, and Rta (VPR) (VPR -dCas9)13 acts as species 73 X (Figure 1 ) and it is constitutively expressed from the pCMV promoter. This synthetic 74 transcriptional activator can drive transcription from any synthetic or endogenous promoter 75 of interest by simply choosing the cognate guide-RNA (gRNA). As species Z of the AIC (Figure 76 1), we chose the Reverse Tet TransActivator (rtTA) driven by the 7B_pMin promoter18, which 77 in the presence of the constitutively expressed gRNA_B, is activated by the VPR-dCas9-N8. In 78 the presence of Doxycycline, rtTA binds the pTRE3G promoter upstream of the anti -CRISPR 79 protein AcrIIA4, which acts as species Y since it can bind and inactivate the cognate Cas9 14,15. 80 We have previously demonstrated19 that the inhibitory activity of AcrIIA4 towards the VPR -81 dCas9 can be increased by more the 3 -fold by boosting their binding affinity thanks to the 82 fusion of two orthogonal synthetic Coiled Coils (CCs) domains (N7 and N816,17) thus giving rise 83 to two new moieties, namely AcrIIA4-N7 and VPR-dCas9-N8, as shown in Figure 2A. Our AIC 84 implementation, which we named the CRISPRaTOR, thus exploits the stoichiometric inhibitory 85 action of AcrIIA4-N7 to negatively regulate the transactivator VPR-dCas9-N8 and to give rise 86 to a negative feedback regulation of the rtTA protein and to Robust Perfect Adaptation (RPA). 87 For example, if the rtTA protein level decreases, then there will be less transcription from the 88 pTREG promoter, and hence less AcrIIA4 -N7 protein. This in turn will lead to an increase in 89 “free” VPR -dCas9-N8 (i.e. not bound by AcrIIA4 -N7) and consequently an increase in rtTA 90 transcription, thus eventually re-equilibrating the level of rtTA. To monitor the level of rtTA, 91 we used two alternative strategies as summarised in Figure 1, and detailed in Figure 2A and 92 Figure 3A. Specifically, we either cloned the firefly luciferase (Fluc) downstream of a pTRE3G 93 promoter, so that by measuring luminescence we have an indirect readout of rtTA level (Figure 94 2A), or alternatively we fused a fluorescence tag to the rtTA and we monitored green 95 fluorescence levels (Figure 3A). 96 97 The CRISPRaTOR confers Robust Perfect Adaptation in transient 98 transfection. 99 100 To test RPA, we fused the bacterial derived Destabilization Domain (DD) 20, a 12 -kDa (107 -101 amino-acid) tag based on a mutated FKBP, to the transcription factor rtTA to obtain DD-rtTA, 102 as shown in Figure 2A. By changing the concentration of the small molecule Shield1, it is 103 possible to change the stability of the DD -rtTA fusion protein and thus its amount . We first 104 assessed the degradation of DD-rtTA protein and its rescue by Shield1 while also confirming 105 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint that the DD-rtTA fusion protein still maintains its function as a transcriptional activator. To this 106 end, as shown in Supplementary Fig. 1A, we cloned the mCherry reporter gene downstream 107 of the pTRE3G promoter while constitutively expressing the DD -rtTA from the pEF1 α 108 promoter. We then measured the fluorescence level in individual cells by flow cytometry in 109 the presence or absence of Shield1 and Doxycycline, as shown in Supplementary Fig. 1B, 1C. 110 Figure 2 – The CRISPRaTOR, an antithetic integral controller based on the CRISPR-antiCRIPSR system: A) Schematic representation of the CRISPRaTOR. The DD-rtTA drives the expression of the Fluc from the pTRE3G promoter, and, only in the Closed Loop configuration, also of the antiCRISPR AcrIIA4-N7. In the Open Loop configuration, the plasmid encoding for the anti -CRISPR protein is substituted by an empty plasmid. B) Experimental validation of robust perfect adaptation to changes in protein degradation. Relative Fluc luminescence is computed as the Fluc luminescence first normalized to the Renilla luminescence at the indicated concentration of the Shield1 molecule , and then divided by its value at 1000 nM of Shield1. Doxycycline is kept constant at 25ng/ml. The green pointed arrows indicate a significant difference in relative luminescence versus the value indicated by the green blunted arrow. Transfected plasmids wi th molar ratios are schematically represented as colored circles with numbers indicating relative molar ratios. C) Experimental exploration of the parameter space in which RPA is achieved. The heatmaps report the relative Fluc luminescence at the indicated concentration of Shield1 and Doxycycline for the CRISPRaTOR in Closed Loop and Open Loop configurations. D) Experimental validation of robust perfect adaptation to changes in plasmid ratios. Relative Fluc luminescence values are computed as the normalized Fluc luminescence at the indicated molar ratios divided by its value at the 1:1 molar ratio. Transfected plasmids with molar ratios are schematically represented as colored circles with numbers indicating relative molar ratios, while the X indicates the plasmid whose ratio is being changed. The concentration of doxycycline is kep t constant at 25ng/ml. Shield1 is kept constant at 1000nM to stabilize DD -rtTA. The pointed arrows indicate a significant difference in relative luminescence versus the value indicated by the blunted arrow of the same color (blue or green). VPR-dCas9-N8: nuclease-deficient Cas9 fused to the transactivation domain VPR and to synthetic coiled-coil N8; AcrIIA4-N7: Anti-CRISPR protein fused to the synthetic coiled-coil N7; DD-rtTA: reverse tetracycline TransActivator 3G fused to the FKBP derived Destabilization Domain (DD) whose degradation is modulated by the small molecule Shield1.; TRE3G: Tetracycline Responsive Element promoter 3G. gRNA_B: gude RNA with sequence B; FLuc: firefly Luciferase. n=4 biological replicates. A minimum of n=3 when one of the measurements was identified as an outlier (Grubbs’ test, alpha=0.2). Statistics analysis has been conducted through a two -way ANOVA test. * P ≤ 0.05 ** P ≤ 0.01 *** P ≤ 0.001 **** P ≤ 0.0001. .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint As Shield1 stabilizes the DD-rtTA protein, while Doxycycline allows it to bind to the pTRE3G 111 promoter, then full expression of mCherry is achievable only in the presence of both drugs, as 112 confirmed by our experimental results. 113 Having confirmed that DD-rtTA acts as a bona-fidae transcriptional activator whose level can 114 be modulated by Shield1, we tested the ability of the CRISPRaTOR in Figure 2A in keeping the 115 expression of the Fluc from the pTRE3G promoter stable in the face of changes in Shield1 116 concentrations, starting at a saturating dose of 1000 nM, which stabilizes the DD -tagged 117 protein. As a negative control, we implemented an Open Loop circuit where AcrII A4-N7 is 118 absent and substituted by an empty plasmid . Since the CRISPRaTOR is encoded on five 119 different plasmids as shown in Supplementary Fig. 1 D, we performed the experiment in 120 transient transfection in Hek293T cells by transfecting equimolar ratios of all the plasmids 121 except for the one encoding the guide -RNA, whose molar concentration was doubled . As 122 reported in Fig. 2B and Supplementary Fig. 2A, for a fixed concentration of Doxycycline of 25 123 ng/ml, the Fluc luminescence in the Open Loop circuit decreases proportionally with the 124 Shield1 concentration. This is expected, as the DD -rtTA protein stability decreases with 125 decreasing concentration of Shield 1 while its mRNA level is unchanged, thus causing an 126 overall decrease in DD -rtTA protein level at equilibrium, and hence of the Fluc expression 127 downstream of the pTRE3G promoter. On the contrary, in the case of the CRISPRaTOR (Fig. 2B 128 - blue), the luminescence level remains constant, thus demonstrating Robust Perfect 129 Adaptation; this can be explained by the fact that as DD-rtTA protein stability decreases, this 130 transiently decreases the DD-rtTA protein level, which in turn decreases expression of 131 antiCRISPR, and frees up more VPR-dCas9-N8 transactivator and consequently increased 132 transcription of the DD-rtTA mRNA, thus re-establishing the correct level of DD-rtTA protein 133 and hence of Fluc expression. To further explore the experimental conditions in which Robust 134 Perfect Adaptation is maintained, we repeated the same experiment, but fixing Doxycycline 135 concentrations at either 50 ng/ml, 75 ng/ml, or 100 ng/ml , and changing Shield1 136 concentrations from 300nM to 1000nM. The results are summarised in Figure 2C and 137 Supplementary Fig 2 B-E. The overall results demonstrate that only the closed loop 138 CRISPRaTOR is able to maintain Fluc expression stable across a wide range of experimental 139 conditions. 140 To further investigate the ability of the CRISPRaTOR in maintaining the FLuc expression stable 141 in the context of transient transfection, we changed the amount of the plasmid encoding DD-142 rtTA relative two the other two plasmids encoding VPR -dCas9-N8 and AcrIIA4-N7, to assess 143 the ability of the CRISPRaTOR to counteract plasmid copy number variations. As shown in Fig. 144 2E and Supplementary Fig. 2F, increasing or decreasing the relative amount of the DD -rtTA 145 up to 50% had a strong effect on the open loop circuit, on the contrary the CRISPRaTOR was 146 able to maintain the relative luminescence constant for most of the plasmid concentrations. 147 Taken together these results demonstrate that the CRISPRaTOR implements an effective AIC 148 motif conferring Robust Perfect Adaptation. 149 150 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint Cells with stable integration of the CRISPRaTOR 151 152 Upon successful characterization of the biomolecular circuit through transient transfection, 153 we aimed at testing the CRISPRaTOR in a more physiologically relevant condition that could 154 be encountered in biotechnological and therapeutic applications. Hence, we generated a cell 155 line with stable genomic integration of the CRISP RaTOR to verify whether Robust Perfect 156 Adaptation would hold also in this setting. 157 To facilitate integration, we reduced the size of the CRISPRaTOR while maintaining the ability 158 to assess its robustness, by modifying it as reported in Figure 3A. Specifically, we chose to 159 directly monitor the DD-rtTA protein expression by fusing the EGFP fluorescent protein to its 160 C-terminus. Additionally, we employed a P2A sequence to co-express the Puromycin 161 resistance protein from the same construct for subsequent selection of stably integrated cells. 162 We then encoded the CRISPRaTOR on only two plasmids as reported in Figure 3C, D. To 163 prevent promoter crosstalk and independent expression of each cistron21, we cloned the two 164 expression cassettes in each plasmid in reverse orientation. To genomically integrate the two 165 plasmids, we opted for DNA transposons, specifically, the PiggyBac system22, and the Sleeping 166 Beauty system23. We also decided to genomically integrate a constitutively expressed nuclear 167 mCherry protein to facilitate the identification of cell nuclei for imaging and fluorescent 168 experiments. For this integration, we generated a third plasmid, as shown in Figure 3E, in 169 which we cloned an H2B-tagged mCherry protein under the control of the weak PGK promoter 170 and inserted it into a vector containing Tol2 repeats for the mT2TP transposase 24. Our 171 integration strategy involved positive selection of integrated cells through Fluorescence 172 Activated Cell Sorting and antibiotic resistance, as illustrated in Figure 3B. We first s tably 173 integrated the nuclear mCherry protein by transient transfection in Hek293T cells of the two 174 plasmids in Figure 2E, one encoding for the cargo to be integrated (the nuclear mCherry) and 175 the other for the mT2TP transposase. We then sorted for the red-positive cells three times to 176 obtain a uniform cell population ( Figure 3B and Supplementary Fig. 3A). The next step 177 involved transient transfecting of the red-positive cells with the two plasmids encoding for the 178 CRISPRaTOR together with additional two plasmids encoding for the Sleeping Beauty and the 179 PiggyBac transposases as shown in Figure 3B, C, D. Antibiotic selection by puromycin followed 180 by sorting of green fluorescent cells was then used to select for cells stably integrating the 181 CRISPRaTOR (Fig. 3B and Supplementary Fig. 3B). Indeed, as DD-rtTA-EGFP and Puromycin 182 resistance are expressed in the same plasmid under the control of the gRNA -inducible 183 7B_pMin promoter, while the transactivator VPR-dCas9-N8 together with the cognate guide 184 RNA (gRNA_B) is expressed on the other plasmid, only cells that have been transfected with 185 both plasmids will be positively selected by puromycin and will pass cell sorting for the green 186 fluorescence. 187 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint 188 The resulting cell line can be used to test both the closed-loop CRISPRaTOR against the Open 189 Loop (OL) configuration. Indeed, growing the cells in the absence of Doxycycline “opens the 190 loop” as the antiCRISPR is not expressed, whereas growing cells in the presence of Doxycycline 191 “closes the loop”. After constructing the stable cell line, we first assessed the functionality of 192 the CRISPRaTOR in response to saturating concentrations of Shield1 and Doxycycline. As 193 illustrated in Supp. Figure 3C , in the absence of both Doxycycline and Shield1, the cells 194 displayed no green fluorescence. Upon addition of Shield1, the population exhibited maximal 195 green fluorescent intensity. Conversely, in the presence of both Shield1 and Doxycycline, cells 196 reached an intermediate level of green fluorescence intensity. This is to be expected, since in 197 Figure 3 – Development of a cell line with stable integration of the CRISPARATOR: A) Schematic representation of the CRISPARATOR genomically integrated in Hek293T cells. The destabilization domain (DD) is fused at the N -terminal of the rtTA protein, whereas at its C -terminal, the fluorescent protein EGFP and the Puromycin resistance protein separated by the self-cleaving 2A peptide are found. The circuit exhibits two distinct operational states: the Closed Loop and the Open Loop. The Closed Loop configuration is achieved upon the addition of Doxycycline, when the AcrIIA4 protein is expressed. Conversely, in the absence of Doxycycline, the system transitions to the Open Loop state, where the AcrIIA4 protein cannot be expressed. B) Schematic representation of the strategy to genomically integrate the CRISPRaTOR by means of transposases. The H2B_mCherry is integrated with the Tol2 transposase through multiple rounds of cell sorting post transfection. Once a stable mCherry-expressing cell line is obtained, const ructs codifying for the CRISPRaTOR are transfected along with their transposases. Puromycin selection occurs only when both plasmids encoding for the CRISPARATOR are present. Sorting for mCherry and EGFP are subsequently performed as quality control. C) Plasmid codifying for the VPR -dCas9_N8, under the control of the strong, constitutive CMV promoter and the guide-RNA gRNA_B under the control of the U6 promoter. The SleepingBeauty transposase is on a second plasmid under the control of the CMV promoter and recognizes the SB sequences D) Plasmid codifying for the AcrIIA4-N7, under the control of the Doxycycline-inducible TRE3G promoter, and for the DD-rtTA-EGFP-2A-Puromycin, under the control of 7B_pMin promoter. On the second plasmid, t he PiggyBac transposase, under the control of the CMV promoter, recognizes the PB sequence. E) Plasmid codifying for H2B_mCherry, under the control of the weak, constitutive promoter PGK, and a second plasmid codifying for the transposase mT2TP, under the control of the CAGGS promoter. The mT2TP transposase recognizes the Tol2 sequences. .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint this condition the rtTA protein binds to the pTRE3G promoter and drives expression of the 198 antiCRISPR AcrIIA4. This protein then sequesters part of the VPR -dCas9 thus reducing 199 transcription from the 7B_pMin promoter of the DD -rtTA-EGFP construct leading to a 200 decrease in green fluorescence. 201 Next, we aimed at assessing the ability of the CRISPRaTOR in achieving RPA in the stable cell 202 line, that is in maintaining the level of DD-rtTA-EGFP protein stable against changes in Shield1 203 concentration. To this end, we employed a High Content Screening platform comprising an 204 Opera Phenix Imaging to quantify green fluorescence across different combinations of 205 Doxycycline and Shield1 concentrations (Figure 4A). As summarised in Figure 4B, we tested 206 the CRISPRaTOR at six fixed concentrations of Doxycycline from 100 ng/ml to 1000 ng/ml 207 (Closed Loop, CL) . For each fixed concentration of doxycycline, we tested 12 different 208 concentrations of Shield 1 ranging from 0 nM to 1000 nM. As a negative control, we tested 209 the cells in the absence of Doxycycline , thus preventing DD-rtTA-EGFP from binding to the 210 Figure 4 – Robust Perfect Adaptation in stable cell lines: A) Schematic representation of the workflow of the High Content Screening platform used to detect and quantify EGFP fluorescence to track rtTA levels in CRISPRaTOR-containing Hek293T cells. Cells were plated in a 96-well plate and, 18 hours after seeding, tr eated with different concentrations of Shield1 and Doxycycline. Forty -eight hours after treatment, cells were imaged using the Opera Phenix. B) Quantification of the green fluorescent signal from confocal images in Hek293T cells integrated with the CRISPRaTOR for decreasing concentrations of Shield1 and for the indicated fixed concentration of Doxycycline. In the absence of Doxycycline, the CRISPRaTOR is in the Open Loop configuration. Fluorescence values are relative to the value measured at 1000 nM Shield1. C) Representative fluorescence images at confocal microscope of Hek293T cells integrated with the CRISPRaTOR following the indicated treatments. D) Relative fluorescence measured as in (C) but represented as a bar plot for only two conditions: Doxycycle 100 ng/ml (blue) and without Doxycycline (green). The pointed arrows indicate a significant difference in relative luminescence versus the value indicated by the blunted arrow of the same color (blue or green). For imaging quantification experiments, n=4 biological replicates coming from 4 96 well plates. Statistics analysis has been conducted through a two-way ANOVA test. * P ≤ 0.05 ** P ≤ 0.01 *** P ≤ 0.001 **** P ≤ 0.0001. .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint pTRE3G promoter and expressing AcrIIA4 to close the feedback loop (Open Loop, OL). Results 211 in Figure 4B show that in the Open Loop configuration, decreasing Shield1 concentration 212

in a proportional decrease in green fluorescence. We confirmed that this effect was 213 specific for the green fluorescence, as the mCherry fluorescence remained constant, as shown 214 in Supplementary Figure 4A. On the contrary, in the presence of Doxycycline, the CRISPRaTOR 215 can maintain green fluorescence intensity unchanged over a larger range of Shield1 216 concentrations. In Figure 4D, the relative green fluorescence values in Figure 4B are reported 217 as a bar plot for only two conditions: a fixed Doxycycline concentration of 100 ng/ml (blue 218 bars), or without Doxycycline (green bars) . It can be observed that in the presence of 219 Doxycycline, even when Shield1 concentration is reduced by 10-fold from 1000 nM to 100 nM, 220 green fluorescence is unchanged. This is, however, not the case in the absence of doxycycline 221 (Open Loop) , where fluorescence is reduced by about 20%. Moreover, even a 100 -fold 222 reduction in Shield 1 (10 nM) results in less than 20% reduction in fluorescence in the presence 223 of doxycycline (Closed Loop), but in more than a 40% reduction in the absence of doxycycline 224 (Open Loop). As a control, we also measured the mCherry fluorescence in Supplementary 225 Figure 4B. We also report raw data of EGFP and mCherry fluorescence ( Supplementary Fig. 226 4C, D). 227 In this work, we present the CRISPRaTOR, a protein-protein-based biomolecular circuit for 228 precise control of gene expression in mammalian cells. We demonstrate that the CRISPRaTOR 229 exhibits Robust Perfect Adaptation, as it can maintain expression of a protein of interest in the 230 face of changes in protein degradation and in plasmid copy number. Importantly, we also show 231 that RPA holds not only in transient transfection, but also in cells genomically integrated with 232 the CRISPRaTOR, thus paving the way for practical applications in biotechnology that require 233 stable cell lines. 234 Recently, two other implementations of AIC controllers in mammalian cells have been 235 reported, one based on sense-antisense RNA13 and the other on protein split-inteins14. In the 236 first implementation, an antisense RNA is produced under the control of a promoter activated 237 by a specific transcription factor (tTA), which then binds to the tTA mRNA, creating a negative 238 feedback loop. Despite being very versatile, this implementation has some drawbacks as it can 239 trigger the Integrated Stress Response in the cell because of the formation of double-stranded 240 RNA. Moreover, it can be applied only when the mRNA to be controlled is stable13. The second 241 implementation uses engineered proteins with split inteins for the key sequestration reaction 242 by exploiting a protein splicing reaction. The splicing deactivates the inteins but preserves the 243 functions of the proteins involved , hence also allow ing the implementation of more 244 sophisticated Proportional-Integral (PI) controller s. The split-inteins implementation of the 245 AIC however is more difficult to adapt to the control of endogenous proteins, as the 246 engineering of proteins with inteins requires careful consideration of intein selection, 247 insertion sites, and its effect on protein folding . In b oth implementations of the AIC , the 248 authors demonstrated Robust Perfect Adaptation (RPA) only by transient transfection. 249 The CRISPRaTOR is a complementary implementation of the AIC in mammalian cells. It 250 exploits the CRISPR -antiCRISPR sequestration reaction that can be easily generalised to 251 control any endogenous transcription factor by: (1) designing the proper guide-RNA to direct 252 VPR-dCas9 to the endogenous promoter, and (2) engineering a synthetic promoter responsive 253 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint to the endogenous transcription factor to drive the expression of the anti-CRISPR. Both steps 254 can be routinely performed with a very high success. One of the limitations of the current 255 implementation is the large genomic size (about 15Kb) that may hinder the delivery in primary 256 cells or tissues where viral vectors are necessary. 257 Finally, we generated a cell line stably integrated with the CRISPRaTOR that exhibits RPA. This 258 cell line can be used to express any protein of interest under the control of the pTRE3G 259 promoter guaranteeing its robust expression over time. 260 261 AUTHOR INFORMATION 262 Corresponding Authors 263 *E-mail: [email protected] 264 ORCID 265 Diego di Bernardo: 0000-0002-1911-7407 266 267 Author contributions 268 A.M. and D.d.B. designed the research; A.M. designed, built, and experimentally validated 269 circuits, carried out experiments and performed data analysis; V.F. and F.R. helped to perform 270 experiments. A.M. and D.d.B wrote the paper. All authors contributed to review and editing; 271 D.d.B. supervised the project and secured funding. 272 Notes 273 The authors declare no competing financial interest. 274 ACKNOWLEDGMENTS 275 We thank Luigi Ferrante from the Fluorescence Activated Cell Sorting facility for his help in the 276 construction of the stable cell line. We also thank Sandro Montefusco and Antonella Capuozzo 277 from the High Content Screening facility run by Prof. Diego Medina, for their support in high 278 content imaging and fluorescence quantification experiments. This work was supported by 279 Fondazione Telethon, by University of Naples Federico II and Compagnia di San Paolo - 280 Programme STAR Plus . A.M. was supported by a fellowship from the European School of 281 Moleculer Medicine (SEMM). 282 283 284 285 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint

286 Plasmid Construction Most of the plasmids were constructed using the Golden -Gate based 287 EMMA assembly kit 25, following authors’ protocol 26, and Gibson assembly method 27. All the 288 sequences regarding N7 and N8 coiled coils17, used to modify the CRISPR-antiCRISPR system, 289 AcrIIA414 and FKBP-derived DD28, gRNA inducible constructs with 7 binding sites for gRNA B18, 290 gRNAs sequences18, VPR-dCas929, PiggyBac22, SleepingBeauty23 and Tol224, have all been taken 291 from published papers. The Tet-On®3G system, comprising the TRE3G promoter and the rtTA 292 protein, has been acquired from Takara Bio. 293 294 Cell culture and transfection The H ek293T cell line (ATCC) was cultured in DMEM Gluta - 295 max (Gibco) supplemented with 10% Tet -Free Fetal Bovine Serum (Euroclone) and 1% 296 Penicillin-Streptomycin (Euroclone). Cells have been kept at 37°C in a 5% CO2 environment. 297 For luciferase experiments, 2x104 Hek293T cells per well were seeded in CoStar White 96-well 298 plates (Corning) to perform standard transfection, while 4,5x10 4 were seeded when 299 performing reverse transfection. For Flow Cytometry assay the same numbers of cells have 300 been seeded in 96-well cell culture plates (Corning). After 18 hours of seedling, for standard 301 transfection, or immediately after seedling, for reverse transfection, cells have been 302 transfected using a home-made solution of PEI (MW 25000, Polysciences, stock concentration 303 0,324 mg/ml, pH 7.5) using 250ng of DNA per well. 304 Luciferase assay To normalize reporter values to transfection efficiency, 10ng of pRL -TK 305 (encoding for Renilla Luciferase) have been used for each experiment. The cells were collected 306 48 hours after transfection and lysed with 5X Passive Lysis Buffer (Biotin) diluted in water. 307 Firefly Luciferase and Renilla Luciferase expression were measured using the Dual Luciferase 308 Assay (Promega) on a Glomax Explorer plate reader (Promega). Firefly Luciferase Arbitrary 309 Units (Luciferase [A.U.]) were calculated by normalizing each sample’s Firefly Luciferase 310 activity to the constitutive Renilla activity detected in the same sample. Relative Luciferase 311 [A.U.] values have been obtained by dividing each sample’s Luciferase [A.U.] by the Luciferase 312 [A.U.] of the sample treated with the maximum dose of Shield1 used, 1000 nM. For the RPA 313 experiments, a Firefly Luciferase with two destabilization sequences has been used (Luc2CP , 314 Promega). 315 Cell Sorting For integration of the vector containing H2B-mCherry, Hek293 were seeded in a 316 6 well plate and transfected with the vector to integrate and the mT2TP transposase with a 317 2:1 molar ratio. Following transfection and expansion, cells underwent three rounds of sorting 318 for mCherry-positive cells, using a BD FACS Aria III Cell Sorting System (Becton Dickinson). For 319 integration of the CRISPRaTOR, H2B -mCherry containing cells were seeded in a 6 well plate 320 and transfected with the CRISPRaTOR-codifying vectors and their respective transposases in a 321 3:1 molar ratio. Following 2 weeks of Puromycin selection (1,5 µg/ml), the mCherry -EGFP 322 double positive cell were selected through cells sorting, using a BD FACS Aria III Cell Sorting 323 System (Becton Dickinson). 324 325 Flow cytometry Cells were collected 48 hours after treatment, washed and resuspended with 326 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint PBS (Euroclone). Flow Cytometry analysis was carried out using an Accuri C6 + (BD 327 Biosciences), analyzing 10000 cells for each sample. A 488 -nm laser with a 670 nm LP filter 328 was used to excite and detect mCherry fluorescence, while a 488-nm laser with a 533/30 nm 329 filter was used to excite and detect EGFP fluorescence. 330 331 Drug treatment Cells were treated with Doxycycline (Clontech) and/or Shield1 332 (MedCehmExpress) immediately before transfection. For High Content Screening and Flow 333 Cytometry experiments, cells were treated 18 hours after seedling. Doxycycline was dissolved 334 in H2O, while Shield1 was dissolved in DMSO. For transient transfection experiments, drugs’ 335 concentrations are referred to the medium volume before adding the transfection mix. 336 337 High Content Screening and Fluorescence Quantification For High Content Screening 338 experiments, 1x104 (Hek293T) cells per well were seeded in PhenoPlate 96-well plates (Perkin 339 Elmer) and, the day after, treated with Doxycycline and Shield1. Forty -eight hours after 340 treatment, cells were fixed in 4% PFA and imaged with the Opera Phenix High Content 341 Screening System (Perkin Elmer), acquiring at least 10 images per well. Cells’ nuclei have been 342 identified through mCherry fluorescence. Fluorescent signals (mCherry and EGFP) were 343 quantified using a custom script developed on Signals Image Artists (Perkin Elmer). 344 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint

345 1. Khammash, M. H. Perfect adaptation in biology. Cell Syst 12, 509–521 (2021). 346 2. El-Samad, H., Goff, J. P . & Khammash, M. Calcium homeostasis and parturient 347 hypocalcemia: An integral feedback perspective. J Theor Biol 214, 17–29 (2002). 348 3. Alon, U., Surette, M. G., Barkai, N. & Leibler, S. Robustness in bacterial chemotaxis. 349 Nature 397, 168–171 (1999). 350 4. Barkai, N. & Leibler, S. Robustness in simple biochemical networks. Nature 387, 913–351 917 (1997). 352 5. Becskel, A. & Serrano, L. Engineering stability in gene networks by autoregulation. 353 Nature (2000) doi:10.1038/35014651. 354 6. Araujo, R. P . & Liotta, L. A. The topological requirements for robust perfect adaptation 355 in networks of any size. Nat Commun 9, 1757 (2018). 356 7. Ma, W., Trusina, A., El-Samad, H., Lim, W. A. & Tang, C. Defining Network Topologies 357 that Can Achieve Biochemical Adaptation. Cell 138, 760–773 (2009). 358 8. Briat, C., Gupta, A. & Khammash, M. Antithetic Integral Feedback Ensures Robust 359 Perfect Adaptation in Noisy Bimolecular Networks. Cell Syst 2, 15–26 (2016). 360 9. Aoki, S. K. et al. A universal biomolecular integral feedback controller for robust 361 perfect adaptation. Nature vol. 570 533–537 Preprint at 362 https://doi.org/10.1038/s41586-019-1321-1 (2019). 363 10. Frei, T., Chang, C. H., Filo, M., Arampatzis, A. & Khammash, M. A genetic mammalian 364 proportional-integral feedback control circuit for robust and precise gene regulation. 365 Proc Natl Acad Sci U S A 119, 1–12 (2022). 366 11. Anastassov, S., Filo, M., Chang, C. H. & Khammash, M. A cybergenetic framework for 367 engineering intein-mediated integral feedback control systems. Nat Commun (2023) 368 doi:10.1038/s41467-023-36863-9. 369 12. Williams, M. P . G. and B. R. G. The response of mammalian cells to double-stranded 370 RNA. Cytokine Growth Factor Rev. 4, 363–371 (2011). 371 13. Chavez, A. et al. Highly efficient Cas9-mediated transcriptional programming. Nat 372

12, 326–328 (2015). 373 14. Nakamura, M. et al. Anti-CRISPR-mediated control of gene editing and synthetic 374 circuits in eukaryotic cells. Nat Commun 10, 1–11 (2019). 375 15. Dong, D. et al. Structural basis of CRISPR-SpyCas9 inhibition by an anti-CRISPR 376 protein. Nature 546, 436–439 (2017). 377 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint 16. Lebar, T., Lainšček, D., Merljak, E., Aupič, J. & Jerala, R. A tunable orthogonal coiled-378 coil interaction toolbox for engineering mammalian cells. Nat Chem Biol 16, 513–519 379 (2020). 380 17. Plaper, T. et al. Coiled-coil heterodimers with increased stability for cellular regulation 381 and sensing SARS-CoV-2 spike protein-mediated cell fusion. Sci Rep 11, 1–16 (2021). 382 18. Praznik, A. et al. Regulation of protein secretion through chemical regulation of 383 endoplasmic reticulum retention signal cleavage. Nat Commun 13, 1–14 (2022). 384 19. Mallozzi, A.*, Fusco V.*, Ragazzini F.* et al. A general strategy to engineer high-385 performance mammalian Whole-Cell Biosensors, BioRxiv (2024). 386 20. Haugwitz, M., Garachtchenko, T., Nourzaie, O., Gandlur, S. & Sagawa, H. Rapid, on-387 demand protein stabilization and destabilization using the ProteoTunerTM systems. 388 Nat Methods 5, iii–iv (2008). 389 21. Johnstone, C. P . & Galloway, K. E. Supercoiling-mediated feedback rapidly couples and 390 tunes transcription. Cell Rep 41, 111492 (2022). 391 22. Zhao, S. et al. PiggyBac transposon vectors: The tools of the human gene encoding. 392 Transl Lung Cancer Res 5, 120–125 (2016). 393 23. Mátés, L. et al. Molecular evolution of a novel hyperactive Sleeping Beauty 394 transposase enables robust stable gene transfer in vertebrates. Nat Genet 41, 753–395 761 (2009). 396 24. Kawakami, K. Tol2: A versatile gene transfer vector in vertebrates. Genome Biol 8, 1–397 10 (2007). 398 25. Martella, A., Matjusaitis, M., Auxillos, J., Pollard, S. M. & Cai, Y . EMMA: An Extensible 399 Mammalian Modular Assembly Toolkit for the Rapid Design and Production of 400 Diverse Expression Vectors. ACS Synth Biol 6, 1380–1392 (2017). 401 26. Jones, S., Martella, A. & Cai, Y . EMMA Assembly Explained: A Step-by-Step Guide to 402 Assemble Synthetic Mammalian Vectors. Methods in Enzymology vol. 617 (Elsevier 403 Inc., 2019). 404 27. Gibson, D. G. et al. Enzymatic assembly of DNA molecules up to several hundred 405 kilobases. Nat Methods 6, 343–345 (2009). 406 28. Haugwitz, M., Garachtchenko, T., Nourzaie, O., Gandlur, S. & Sagawa, H. Rapid, on-407 demand protein stabilization and destabilization using the ProteoTunerTM systems. 408 Nat Methods 5, iii–iv (2008). 409 29. Karottki, K. J. la C. et al. Awakening dormant glycosyltransferases in CHO cells with 410 CRISPRa. Biotechnol Bioeng (2020) doi:10.1002/bit.27199. 411 412 .CC-BY-ND 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 March 30, 2024. ; https://doi.org/10.1101/2024.03.30.587417doi: bioRxiv preprint