A monogenic and fast-responding Light-Inducible Cre recombinase as a novel optogenetic switch

Optogenetics enables genome manipulations with high spatiotemporal resolution, opening exciting possibilities for fundamental and applied biological research. Here, we report the development of LiCre, a novel light-inducible Cre recombinase. LiCre is made of a single flavin-containing protein comprising the asLOV2 photoreceptor domain of Avena sativa fused to a Cre variant carrying destabilizing mutations in its N-terminal and C-terminal domains. LiCre can be activated within minutes of illumination with blue light, without the need of additional chemicals. When compared to existing photoactivatable Cre recombinases based on two split units, LiCre displayed faster and stronger activation by light as well as a lower residual activity in the dark. LiCre was efficient both in yeast, where it allowed us to control the production of β-carotene with light, and in human cells. Given its simplicity and performances, LiCre is particularly suited for fundamental and biomedical research, as well as for controlling industrial bioprocesses.

transgene that is only expressed in the cells to be mutated. The location and orientation of 72 LoxP sites can be chosen so that recombination generates either a deletion, an inversion or a 73 translocation. Similar systems were developed based on other recombinases/recognition 74 targets, such as Flp/FRT 3 or Dre-rox 4 . To control the timing of recombination, several 75 systems were made inducible. Tight control was obtained using recombinases that are inactive 76 unless a chemical ligand is provided to the cells. For example, the widely-used Cre-ERT 77 chimeric protein can be activated by 4-hydroxy-tamoxifen 5 . Other inducible systems rely on 78 chemical-induced dimerization of two halves of the recombinase. For example, the FKBP-79 FRB split Cre system consists of two inactive proteins that can assemble in the presence of 80 rapamycin to form a functional recombinase complex 6 . Similar systems were reported that 81 rendered dimerization of the split Cre fragments dependent on phytohormones 7 . Although 82 powerful, these systems present some caveats: ligands are not always neutral to cells and can 83 efficient expression of their two different coding sequences, as previously reported 24 . An 118 inducible system based on a single protein may avoid this limitation. Its implementation by 119 transgenesis would also be simpler, especially in vertebrates. 120 121 We report here the development of LiCre, a novel Light-Inducible Cre recombinase 122 that is made of a single flavin-containing protein. LiCre can be activated within minutes of 123 illumination with blue light, without the need of additional chemicals, and it shows extremely 124 low background activity in absence of stimulation as well as high induced activity. Using the 125 production of carotenoids by yeast as a case example, we show that LiCre and blue light can 126 be combined to control metabolic switches that are relevant to the problem of metabolic 127 burden in bioprocesses. We also report that LiCre can be used efficiently in human cells, 128 making it suitable for biomedical research. Since LiCre offers cheap and precise 129 spatiotemporal control of a genetic switch, it is amenable to numerous biotechnological 130 applications, even at industrial scales. 131

RESULTS 132 133
The stabilizing N-ter and C-ter α-helices of the Cre recombinase are critical for its 134 activity 135 136 A variety of optogenetic tools have been successfully developped based on LOV 137 domain proteins, which possess α-helices that change conformation in response to light 25 . We 138 reasoned that fusing a LOV domain to a helical domain of Cre that is critical for its function 139 could generate a single protein with light-dependent recombinase activity. We searched for 140 candidate α-helices by inspecting the structure of the four Cre units complexed with two LoxP 141 DNA targets 26,27 ( Fig. 1a-b). Each subunit folds in two domains that bind to DNA as a clamp. terminal domains (Fig. 1a) and contacts involving αN lock the four carboxy-terminal domains 146 in a cyclic manner (Fig. 1b). These helices were therefore good candidates for manipulating 147 Cre activity. We focused on αA and αN because their location at protein extremities was 148 convenient to design chimeric fusions. 149 150 We tested the functional importance of helices αA and αN by gradually eroding them. 151 We evaluated the corresponding mutants by expressing them in yeast cells where an active 152 Cre can excise a repressive DNA element flanked by LoxP sites, and thereby switch ON the 153 expression of a Green Fluorescent Protein (GFP) (Fig. 1c). After inducing the expression of 154 Cre mutants with galactose, we counted by flow cytometry the proportion of cells that 155 expressed GFP and we used this measure to compare recombinase activities of the different 156 mutants (Fig. 1d). As a control, we observed that the wild-type Cre protein activated GFP 157 expression in all cells under these conditions. Mutants lacking the last 2 or the last 3 carboxy-158 terminal residues displayed full activity. In contrast, mutants lacking 4 or more of the C-ter 159 residues were totally inactive. This was consistent with a previous observation that deletion of 160 the last 12 residues completely suppressed activity 28 . Our series of mutants showed that helix 161 αN is needed for activity and that its residue D341 is crucial. The role of this aspartic acid is 162 most likely to stabilize the complex: the tetramer structure indicates salt bridges between 163 D341 and residue R139 of the adjacent unit (Fig. 1e). Interestingly, E340 might have a similar 164 role by interacting with R192, although this residue was not essential for activity. 165 Biomolecular simulations using a simplistic force-field model showed that the free-energy 166 barrier for displacing the αN helix was much lower if E340 and D341 were replaced by 167 alanines (Fig. 1f). Consistent with this prediction, we observed that a double mutant E340A 168 D341A lost ~10% of activity (Fig. 1g). This mild (but reproducible) reduction of activity 169 suggested that the double mutation E340A D341A led to a fragilized version of Cre where 170 multimerization was suboptimal. 171

172
We also tested the functional importance of α-helix A, either in a normal context 173 where the C-terminal part of Cre was intact or where it carried the destabilizing E340A 174 D341A mutation (Fig. 1g). Deletion of residues 2-37, which entirely ablated helix A, 175 eliminated enzymatic activity (Fig. 1g). Very interestingly, the effect of shorter deletions 176 depended on the C-terminal context. When the C-terminus was wild-type, removing residues 177 2-21 (immediately upstream of helix A) had no effect and removing residues 2-28 (partial 178 truncation of αA) decreased the activity by ~10%. When the C-terminus contained the E340A 179 D341A mutation, deletions 2-21 and 2-28 were much more severe, reducing the activity by 180 12% and 80%, respectively. This revealed genetic interactions between the extremities of the 181 protein, which is fully consistent with a cooperative role of helices αA and αN in stabilizing 182 an active tetramer complex. From these observations, we considered that photo-control of Cre 183 activity might be possible by fusing αA and αN helices to LOV domain photoreceptors. from Neurospora crassa 12,29,30 . The resulting chimeric protein, which contained the full-190 length Cre connected to VVD via four amino-acids, did not display light-dependent 191 recombinase activity ( Supplementary Fig. S1). Our next strategy was based on a modified 192 version of the asLOV2 domain from Avena sativa which had been optimized by Guntas et 193 al. 31 . These authors used it to build an optogenetic dimerizer by fusing its Jα C-ter helix to the 194 bacterial SsrA peptide. Instead, we fused Jα to the αA amino-terminal helix of Cre. Using the 195 same GFP reporter system as described above for detecting in-vivo recombination in yeast, we 196 built a panel of constructs with various fusion positions and we directly quantified their 197 activity with and without blue-light illumination. All fusions displayed reduced activity in 198 both dark and light conditions as compared to wild-type Cre. Three constructs -199 corresponding to fusions of asLOV2 to residues 19, 27 and 32 of Cre, respectively -displayed 200 higher activity after light stimulation. We recovered the corresponding plasmids from yeast, 201 amplified them in bacteria to verify their sequence and re-transformed them in yeast which 202 confirmed the differential activity between dark and light conditions for all three constructs 203 (Fig. 2a). Fusion at position 32 (named LOV2_Cre32) displayed the highest induction by 204 light, with activity increasing from 15% in dark condition to 50% after 30 minutes of 205 illumination. Although this induction was significant, a 15% activity of the non-induced form 206 remained too high for most applications. We therefore sought to reduce this residual activity, 207 which we did in two ways. 208 209 First, we randomized the residues located at the junction between asLOV2 and Cre. 210 We used degenerate primers and in-vivo recombination (see methods) to mutagenize 211 LOV2_Cre32 at these positions and we directly tested the activity of about 90 random clones. 212 Five of them showed evidence of low residual activity in the dark and we characterized them 213 further by sequencing and re-transformation. For all five clones, residual activity was indeed 214 reduced as compared to LOV2_Cre32, with the strongest reduction being achieved by an 215 isoleucine insertion at the junction position (Fig. 2b iv). However, this improvement was also 216 accompanied by a weaker induced activity and a larger variability between independent 217 assays. 218 219 As a complementary approach to reduce residual activity, we took advantage of the 220 above-described genetic interaction between N-ter truncations and C-ter mutations targeting 221 residues 340 and 341. We built another series of constructs where asLOV2 fusions to αA 222 helix were combined with the A340A341 double mutation. This approach yielded one 223 construct (LOV2_CreAA20), corresponding to fusion at position 20, which displayed a 224 residual activity that was undistinguishable from the negative control, and a highly-225 reproducible induced activity of ~25% (Fig. 2c) We placed LiCre under the expression of the P MET17 promoter and we tested various 231 illumination intensities and durations on cells that were cultured to stationary phase in 232 absence of methionine (full expression). Activity was very low without illumination and 233 increased with both the intensity and duration of light stimulation (Fig. 3a). The minimal 234 intensity required for stimulation was comprised between 0.057 and 1.815 mW/cm 2 . The 235 highest activity (~65% of switched cells) was obtained with 90 minutes illumination at 36.3 236 mW/cm 2 . Extending illumination to 180 minutes did not further increase the fraction of 237 switched cells. Remarkably, we observed that 2 minutes of illumination was enough to switch 238 5% of cells, and 5 minutes illumination generated 10% of switched cells (Fig. 3b). 239

240
We compared these performances with those of two previous systems that were both 241 based on light-dependent complementation of a split Cre enzyme. We constructed plasmids 242 coding for proteins CreN59-nMag and pMag-CreC60 described in Kawano et al. 21 and 243 transformed them in our yeast reporter strain. Similarly, we constructed and tested plasmids 244 coding for the proteins CRY2 L348F -CreN and CIB1-CreC described in Taslimi et al. 20 . All four 245 coding sequences were placed under the control of the yeast P MET17 promoter. We analyzed 246 the resulting strains as above after adapting light to match the intensity recommended by the 247 authors (1.815 mW/cm 2 for nMag/pMag and 5.45 mW/cm 2 for CRY2 L348F /CIB1). As shown 248 in Fig. 3c, we validated the photoactivation of nMag/pMag split Cre in yeast, where activity 249 increased about 4-fold following 90 minutes of illumination, but we were not able to observe 250 photoactivation of the CRY2 L348F /CIB1 split Cre system (Fig. 3d). In addition, the 251 photoactivation of nMag/pMag split Cre was not as fast as the one of LiCre, since 30 minutes 252 of illumination was needed to observe a significant increase of activity. This observation is 253 consistent with the fact that dimerization of split Cre, which is not required for LiCre, limits 254 the rate of formation of an active recombination synapse. Another difference was that, unlike 255 LiCre, nMag/pMag split Cre displayed a mild but significant background activity in absence 256 of illumination (~6% of switched cells) (Fig. 3c). Altogether, these results show that, at least 257 in the yeast cellular context, LiCre outperforms these two other systems in terms of 258 efficiency, rapidity and residual background activity. 259 260 To demonstrate the control of a biological activity by light, we built a reporter where 261 Cre-mediated excision enabled the expression of the HIS3 gene necessary for growth in 262 absence of histidine. We cultured cells carrying this construct and expressing LiCre and we 263 spotted them at various densities on two HISselective plates. One plate was illuminated 264 during 90 minutes while the other one was kept in the dark and both plates were then 265 incubated for growth. After three days, colonies were abundant on the plate that had been 266 illuminated and very rare on the control plate (Fig. 3e). LiCre can therefore be used to trigger 267 cell growth with light. 268

269
We then sought to observe the switch in individual cells. To do so, we replaced GFP 270 by mCherry in our reporter system, so that the excitation wavelength of the reporter did not 271 overlap with stimulation of LiCre. We expressed and stimulated LiCre (90min at 3.63 272 mW/cm 2 ) in cells carrying this reporter and subsequently imaged them over time. As 273 expected, we observed the progressive apparition of mCherry signal in a fraction of cells ( the internal region, these sites become proximal and PCR amplification is efficient (Fig. 3h). 283 We mixed known amounts of edited and non-edited genomic DNA and performed real-time 284 qPCR to build a standard curve that could be used to infer the proportion of edited DNA from 285 qPCR signals. After this calibration, we applied this qPCR assay on genomic DNA extracted 286 from cells collected immediately after different durations of illumination at moderate intensity 287 (3.63 mW/cm2). Results were in full agreement with GFP-based quantifications (Fig. 3i). 288 Excision of the target DNA occurred in a significant fraction of cells after only 2 minutes of 289 illumination, and we estimated that excision occurred in about 30% and 40% of cells after 20 290 and 40 minutes of illumination, respectively. To determine if DNA excision continued to 291 occur after switching off the light, we re-incubated half of the cells for 90 minutes in the dark 292 prior to harvest and genomic DNA extraction. The estimated frequency of DNA excision was 293 strikingly similar to the one measured immediately after illumination (Fig. 3j). We conclude 294 that the reversal of activated LiCre to its inactive state is very rapid in the dark (within 295 minutes). 296

297
The qPCR assay also allowed us to compare the efficiency of light-induced 298 recombination between cell populations in exponential growth or in stationary phase. This 299 revealed that LiCre photoactivation was about 4-fold more efficient in non-dividing cells (Fig.  300 3k). Although the reasons for this difference remain to be determined, this increase of LiCre 301 photoactivation at stationary phase makes it particularly suitable for bioproduction 302 applications, where metabolic switching is often desired after the growth phase (see 303 discussion We built a structural model of LiCre to conceptualize its mode of activation (Fig. 4a). 308 We based this model on i) the available structure of the Cre tetramer complexed with its target 309 DNA 27 , ii) the available structure of asLOV2 in its dark state 31  only one of the two reporters (fluorescence in only one of the channels), ruling out the 334 possibility that monomer activation is solely limiting (Fig. 4b). However, the probability that 335 a reporter had switched depended on whether the other reporter had also switched. For 336 example, the proportion of green cells in the whole population (marginal probability to switch 337 the green reporter) was ~20%, but the proportion of green cells in the subpopulation of red 338 cells (conditional probability) was over 30%. Similarly, red cells were more frequent in the 339 subpopulation of green cells than in the whole population (Fig. 4b). These observations ruled 340 out the possibility that formation of a functional LiCre:DNA synaptic complex was solely 341 limiting. We conclude that neither monomer activation nor synapse formation is the sole rate-342 LiCre provides a light-switch for carotenoid production 345 346 LiCre offers a way to change the activities of cells without adding any chemical to 347 their environment. This potentially makes it an interesting tool to address the limitations of 348 metabolic burden in industrial bioproduction (see discussion). We therefore tested the 349 possibility to use LiCre to control the production of a commercial compound with light. 350

351
Carotenoids are pigments that can be used as vitamin A precursors, anti-oxydants or 352 coloring agents, making them valuable for the food, agriculture and cosmetics industries 32 . 353 Commercial carotenoids are generally produced by chemical synthesis or extraction from 354 vegetables, but alternative productions based on microbial fermentations offer remarkable 355 advantages, including the use of low-cost substrates and therefore a high potential for 356 financial gains. Bioproduction of carotenoids from microbes has therefore received an 357 increasing interest. It can be based on microorganisms that naturally produce carotenoids 32 . It 358 is also possible to introduce recombinant biosynthesis pathways in host microorganisms, 359 which offers the advantage of a well-known physiology of the host and of optimizations by 360 genetic engineering. For these reasons, strategies were previously developed to produce 361 carotenoids in the yeast S. cerevisiae. Expressing three enzymes (crtE, crtI and crtYB) from 362 Xanthophyllomyces dendrorhous enabled S. cerevisiae to efficiently convert farnesyl 363 pyrophosphate (FPP) into β-carotene 33 . FPP is naturally produced by S. cerevisiae from 364 Acetyl-CoA and serves as an intermediate metabolite, in particular for the production of 365 ergosterol which is essential for cellular viability (Fig. 5). Thus, and as for any bioproduction 366 consuming a cellular resource, this design is associated with a trade-off: redirecting FPP to β-367 carotene limits its availability for ergosterol biosynthesis and therefore impairs growth; and 368 its consumption by the host cell can limit the flux towards the recombinant pathway. A 369 promising way to deal with this trade-off would be to favor the flux towards ergosterol during 370 biomass expansion and, after enough producer cells are obtained, to switch the demand in 371 FPP towards β-carotene. We therefore explored if LiCre could offer this possibility. 372 373 First, we tested if LiCre could allow us to switch ON the exogenous production of 374 carotenoids with light. If so, one could use it to trigger production at the desired time of a 375 bioprocess. We constructed a S. cerevisiae strain expressing only two of the three enzymes 376 required for β-carotene production. Expression of the third enzyme, a bifunctional phytoene 377 synthase and lycopene cyclase, was blocked by the presence of a floxed terminator upstream 378 of the coding sequence of the crtYB gene (Fig. 5b). Excision of this terminator should restore 379 a fully-functional biosynthetic pathway. As expected, this strain formed white colonies on 380 agar plates, but it formed orange colonies after transformation with an expression plasmid 381 coding for Cre, indicating that β-carotene production was triggered (Fig. 5c). To test the 382 possible triggering by light, we transformed this strain with a plasmid encoding LiCre and 383 selected several transformants, which we cultured and exposed -or not -to blue light before 384 spotting them on agar plates. The illuminated cultures became orange while the non-385 illuminated ones remained white. Plating a dilution of the illuminated cell suspension yielded 386 a majority of orange colonies, indicating that LiCre triggered crtYB expression and β-carotene 387 production in a high proportion of plated cells (Fig. 5c). We quantified bioproduction by 388 dosing total carotenoids in cultures that had been illuminated or not. This revealed that 72 389 hours after the light switch the intracellular concentration of carotenoids had jumped from 390 background levels to nearly 200µg/g (Fig. 5d). Thus, LiCre allowed us to switch ON the 391 production of carotenoids by yeast using blue light. 392

393
We then tested if LiCre could allow us to switch OFF with light the endogenous 394 ergosterol pathway that competes with carotenoid production for FPP consumption. The first 395 step of this pathway is catalysed by the Erg9p squalene synthase. Given the importance of 396 FPP availability for the production of various compounds, strategies have been reported to 397 control the activity of this enzyme during bioprocesses, especially in order to reduce it after 398 biomass expansion 34-36 . These strategies were not based on light but derived from 399 transcriptional switches that naturally occur upon addition of inhibitors or when specific 400 nutrients are exhausted from the culture medium. To test if LiCre could offer a way to switch 401 ERG9 activity with light, we modified the ERG9 chromosomal locus and replaced the coding 402 sequence by a synthetic construct comprising a floxed sequence coding for Erg9p and 403 containing a transcriptional terminator, followed by a sequence coding for the catalytic 404 domain of the 3-hydroxy3-methylglutaryl coenzyme A reductase (tHMG1) (Fig. 5e). This 405 design prepares ERG9 for a Cre-mediated switch: before recombination, Erg9p is normally 406 expressed; after recombination, ERG9 is deleted and the tHMG1 sequence is expressed to 407 foster the mevalonate pathway. Given that ERG9 is essential for yeast viability in absence of 408 ergosterol supplementation 37 , occurrence of the switch can be evaluated by measuring the 409 fraction of viable yeast cells prior and after the induction of recombination. When doing so, 410 we observed that expression of Cre completely abolished viability, regardless of illumination. 411 In contrast, cultures expressing LiCre were highly susceptible to light: they were fully viable 412 in absence of illumination and lost ~23% of viable cells after light exposure (Fig. 5f) organisms. Therefore, we tested its efficiency in human cells. For this, we constructed a 420 lentiviral vector derived from the simian immunodeficiency virus (SIV) and encoding a 421 human-optimized version of LiCre with a nuclear localization signal fused to its N-terminus. 422 To quantify the efficiency of this vector, we also constructed a stable reporter cell line where 423 expression of a membrane-located mCherry fluorescent protein could be switched ON by 424 Cre/Lox recombination. We obtained this line by Flp-mediated insertion of a single copy of 425 the reporter construct into the genome of Flp-In™ 293 cells (Fig. 6a, see methods). Our assay 426 consisted of producing LiCre-encoding lentiviral particle, depositing them on reporter cells 427 for 24h, illuminating the infected cultures with blue light and, 28 hours later, observing cells 428 by fluorescence microscopy. As shown in Fig. 6b, mCherry expression was not detected in 429 non-infected reporter cells. In cultures that were infected but not illuminated, a few positive 430 cells were observed. In contrast, infected cultures that had been exposed to blue light 431 level; with a self-cleaving peptide, cleavage of the precursor protein can be incomplete, 500 generating uncleaved products with unknown activity. This was the case for nMag/pMag split 501 Cre in mammalian cells, where a non-cleaved form at ~72 kDa was reported and where 502 targeted modifications of the cleavage sequence increased both the abundance of this non-503 cleaved form and the non-induced activity of the system 45 . The second benefit of LiCre being 504 a single protein is to avoid problems of suboptimal stoichiometry between the two protein 505 units, which was reported as a possible issue for CRY2/CIB1 split Cre 24 . A third benefit is to 506 avoid possible intra-molecular recombination between the homologous parts of the two 507 coding sequences. Although not demonstrated, this undesired possibility was suspected for 508 nMag/pMag split Cre because its two dimerizers derive from the same sequence 45 . The other 509 advantages of LiCre are its performances. In the present study, we used a yeast-based assay to 510 compare LiCre with split Cre systems. Unexpectedly, although we used the improved version 511 of the CRY2/CIB1 split Cre containing the CRY2-L348F mutation 20 , it did not generate 512 photo-inducible recombination in our assay. This is unlikely due to specificities of the who reported that the induced activity of the CRY2-L348F/CIB1 system was low and highly 518 variable. In contrast, we validated the efficiency of nMag/pMag split Cre and so did other 519 independent laboratories 7,46,4745 . LiCre, however, displayed weaker residual activity than 520 nMag/pMag split Cre in the dark. Reducing non-induced activity is essential for many 521 applications where recombination is irreversible. Very recently, the nMag/pMag split Cre 522 system was expressed in mice as a transgene -dubbed PA-Cre3.0 -which comprised the 523 promoter sequence of the chicken beta actin gene (CAG) and synonymous modifications of 524 the original self-cleaving coding sequence. The authors reported that this strategy abolished 525 residual activity, and they attributed this improvement to a reduction of the expression level 526 of the transgene 45 . It will therefore be interesting to introduce LiCre in mice with a similar 527 expression system and compare it to PA-Cre3.0. Importantly, LiCre also displayed higher 528 induced activity and a faster response to light as compared to nMag/pMag split Cre. This 529 strong response probably results from its simplicity, since the activation of a single protein 530 involves fewer steps than the activation of two units that must then dimerize to become 531 functional. In conclusion, LiCre is simpler and more efficient than previously-existing photo-532 activatable recombinases. With their capability to convert low-cost substrates into valuable chemicals, cultured 537 cells have become essential actors of industrial production. However, although metabolic 538 pathways can be rewired in favor of the desired end-product, the yields of bioprocesses have 539 remained limited by a challenging and universal phenomenon called metabolic burden. This 540 effect corresponds to the natural trade-off between the fitness of host cells and their efficiency 541 at producing exogenous compounds 48 . Loss of cellular fitness is sometimes due to viability 542 issues -e.g. if the end-product is toxic to the producing cells -and sometimes simply to the 543 fact that resources are allocated to the exogenous pathway rather than to the cellular needs. 544 Reciprocally, satisfying the cellular demands can compromise the efficiency of exogenous 545 pathways. In the case of carotenoids production by yeast, metabolic burden was shown to be 546 substantial 49 (µ max reduced by ~12%). This growth defect presumably involves competition 547 for FPP, which is consumed to produce carotenoids but which is also crucially needed by 548 cells to synthesize ergosterol, a major constituent of their membranes 50 . 549 HO locus. This way, we integrated it in a leu2 his3 strain, which could then switch from 631 LEU+ his-to leu-HIS+ after Cre-mediated recombination (Fig. 4e). To construct a GFP-632 based reporter, we ordered the synthesis of sequence LEULoxGreen (Supplementary Text S1) 633 from GeneCust who cloned the corresponding NheI-SacI fragment into pGY262 to obtain 634 pGY407. We generated strain GY984 by crossing BY4726 with FYC2-6B. We transformed 635 GY984 with the 4-Kb NotI insert of pGY407 and obtained strain GY1752. To remove the 636 ade2 marker, we crossed GY1752 with FYC2-6A and obtained strain GY1761. Plasmid 637 pGY537 targeting integration at the LYS2 locus was obtained by cloning the BamHI-EcoRI 638 fragment of pGY407 into the BamHI, EcoRI sites of pIS385. Plasmid pGY472 was produced 639 by GeneCust who synthesized sequence LEULoxmCherry (Supplementary Text S1) and 640 cloned the corresponding AgeI-EcoRI insert into the AgeI,EcoRI sites of pGY407. We 641 generated GY983 by crossing BY4725 with FYC2-6A. We obtained GY2033 by 642 transformation of FYC2-6B with a 4-Kb NotI fragment of pGY472. We obtained GY2207 by 643 transformation of GY983 with the same 4-Kb NotI fragment of pGY472. To generate 644 GY2206, we linearized pGY537 with NruI digestion, transformed in strain GY855 and 645 selected a LEU+ Lys-colony (pop-in), which we re-streaked on 5-FoA plates for vector 646 excision by counter-selection of URA3 (pop-out) 65 . Strain GY2214 was a diploid that we 647 obtained by mating GY2206 with GY2207. 648 649 Yeast expression plasmids. Mutations E340A D341A were introduced by GeneCust 650 by site-directed mutagenesis of pSH63, yielding plasmid pGY372. We generated the N-651 terΔ21 mutant of Cre by PCR amplification of the P GAL1 promoter of pSH63 using primer 652 1L80 (forward) and mutagenic primer 1L71 (reverse), digestion of pSH63 by AgeI and co-653 transformation of this truncated plasmid and amplicon in a trp1Δ63 yeast strain for 654 homologous recombination and plasmid rescue. We combined the N-terΔ21 and the C-ter 655 E340A D341A mutations similarly, but with pGY372 instead of pSH63. We generated N-656 terΔ28 and N-terΔ37 mutants, combined or not with C-ter E340A D341A mutations, by the 657 same procedure where we changed 1L71 by mutagenic primers 1L72 and 1L73, respectively. 658 To generate a Cre-VVD fusion, we designed sequence CreCVII (Supplementary Text 659 S1) where the Cre sequence from GENBANK AAG34515.1 was fused to the VVD-660 M135IM165I sequence from Zoltowski et al. 29 via four additional residues (GGSG). We 661 ordered its synthesis from GeneCust, and we co-transformed it in yeast with pSH63 662 (previously digested by NdeI and SalI) for homologous recombination and plasmid rescue. 663 This generated pGY286. We then noticed an unfortunate error in AAG34515.1, which reads a 664 threonine instead of an asparagine at position 327. We cured this mutation from pGY286 by 665 site-directed mutagenesis using primers 1J47 and 1J48, which generated pGY339 which 666 codes for Cre-VVD described in Supplementary Fig. S1. We constructed mutant C-terΔ14 of 667 Cre by site-directed mutagenesis of pGY286 using primers 1J49 and 1J50 which 668 simultaneously cured the N327T mutation and introduced an early stop codon. Mutants C-669 terΔ2, C-terΔ4, C-terΔ6, C-terΔ8, C-terΔ10, C-terΔ12 of Cre were constructed by GeneCust 670 who introduced early stop codons in pGY339 by site-directed mutagenesis. 671

672
To test LOV2_Cre fusions, we first designed sequence EcoRI-LovCre_chimJa-BstBI 673 (Supplementary Text S1) corresponding to the fusion of asLOV2 with Cre via an artificial α-674 helix. This helix was partly identical to the Jα helix of asLOV2 and partly identical to the αA 675 helix of Cre. This sequence was synthesized and cloned in the EcoRI and BstBI sites of 676 pSH63 by GeneCust, yielding pGY408. We then generated and directly tested a variety of 677 LOV2_Cre fusions. To do so, we digested pGY408 with BsiWI and MfeI and used this 678 fragment as a recipient vector; we amplified the Cre sequence from pSH63 using primer 679 1G42 as the reverse primer, and one of primers 1M42 to 1M53 as the forward primer (each 680 primer corresponding to a different fusion position); we co-transformed the resulting 681 amplicon and the recipient vector in strain GY1761, isolated independent transformants and 682 assayed them with the protocol of photoactivation and flow-cytometry described below. We 683 generated and tested a variety of LOV2_CreAA fusions by following the same procedure 684 where plasmid pGY372 was used as the PCR template instead of pSH63. To introduce random residues at the peptide junction of LOV2_Cre32 (Fig. 2b), we 693 first generated pGY417 using the same procedure as for the generation of pGY415 but with 694 pSH47 instead of pSH63 as the PCR template so that pGY417 has a URA3 marker instead of 695 TRP1. We then ordered primers 1N24, 1N25 and 1N26 containing degenerate sequences, we 696 used them with primer 1F14 to amplify the Cre sequence of pSH63, we co-transformed in 697 strain GY1761 the resulting amplicons together with a recipient vector made by digesting 698 plasmid pGY417 with NcoI and BsiWI, and we isolated and directly tested individual 699 transformants with the protocol of photoactivation and flow-cytometry described below. 700 Plasmids from transformants showing evidence of reduced background were rescued from 701 yeast and sequenced, yielding pGY459 to pGY464. 702

703
To replace the P GAL1 promoter of pGY416 by the P MET17 promoter, we digested it with 704 SacI and SpeI, we PCR-amplified the P MET17 promoter of plasmid pGY8 with primers 1N95 705 and 1N96, and we co-transformed the two products in yeast for homologous recombination, 706 yielding plasmid pGY466. We changed the promoter of pGY415 using exactly the same 707 procedure, yielding plasmid pGY465. We changed the promoter of pSH63 similarly, using 708 primer 1O83 instead of 1N96, yielding plasmid pGY502. 709 710 To express the nMag/pMag split Cre system in yeast, we designed sequence CreN-711 nMag-NLS-T2A-NLS-pMag-CreCpartly (Supplementary Text S1) and ordered its synthesis 712 from GeneCust. The corresponding BglII fragment was co-transformed in yeast for 713 homologous recombination with pGY465 previously digested with BamHI (to remove 714 asLOV2 and part of Cre), yielding plasmid pGY488 that contained the full system. We then 715 derived two plasmids from pGY488, each one containing one half of the split system under 716 the control of the Met17 promoter. We obtained the first plasmid (pGY491, carrying the 717 TRP1 selection marker) by digestion of pGY488 with SfoI and SacII and co-transformation of 718 the resulting recipient vector with a PCR product amplified from pGY465 using primers 719 1O80 and 1O82. We obtained the second plasmid (pGY501, carrying the URA3 selection 720 marker) in two steps. We first removed the pMag-CreC part of pGY488 by digestion with 721 NdeI and SacII followed by Klenow fill-in and religation. We then changed the selection 722 marker by digestion with PfoI and KpnI and co-transformation in yeast with a PCR product 723 amplified from pSH47 with primers 1O77 and 1O89. 724

725
To express the CRY2 L348F /CIB1 split Cre system in yeast, we designed sequences 726 CIB1CreCter and CRY2CreNter and ordered their synthesis from GeneCust, obtaining 727 plasmids pGY526 and pGY527, respectively. To obtain pGY531, we extracted the synthetic 728 insert of pGY527 by digestion with BglII and we co-transformed it in yeast with the NdeI-729 BamHI fragment of pGY466 for homologous recombination. To obtain pGY532, we 730 extracted the synthetic insert of pGY526 by digestion with BglII and we co-transformed it in 731 yeast with the SacI-BamHI fragment of pSH47 for homologous recombination. 732

733
To build a switchable strain for carotene production, we modified EUROSCARF 734 strain Y41388 by integrating a LoxP-KlLEU2-T ADH1 -LoxP cassette immediately upstream the 735 CrtYB coding sequence of the chromosomally-integrated expression cassette described by 736 Verwaal et al. 33 . This insertion was obtained by transforming Y41388 with a 6.6Kb BstBI 737 fragment from plasmid pGY559 and selecting a Leu+ transformant, yielding strain GY2247. 738 To obtain pGY559, we first deleted the crtE and crtI genes from YEplac195-YB_E_I 33 by 739 MluI digestion and religation. We then linearized the resulting plasmid with SpeI and co-740 transformed it for recombination in a leu2Δ yeast strain with a PCR amplicon obtained with 741 primers 1P74 and 1P75 and template pGY407. After Leu+ selection, the plasmid was 742 recovered from yeast, amplified in bacteria and verified by restriction digestion and 743 sequencing. 744

745
We used CRISPR/Cas9 to build a switchable strain for squalene synthase. We cloned 746 the synthetic sequence gERG9 (Supplementary Text S1) in the BamHI-NheI sites of the 747 pML104 plasmid 66 so that the resulting plasmid (pGY553) coded for a gRNA sequence 748 targeting ERG9. This plasmid was transformed in GY2226 together with a repair-template 749 corresponding to a 4.2-Kb EcoRI fragment of pGY547 that contained LoxP-synERG9-T ADH1 -750 LoxP with homologous flanking sequences. The resulting strain was then crossed with 751 Y41388 to obtain GY2236. 752 753 Yeast culture media. We used synthetic (S) media made of 6.7 g/L Difco Yeast 754 Nitrogen Base without Amino Acids and 2 g/L of a powder which was previously prepared by 755 mixing the following amino-acids and nucleotides: 1 g of Adenine, 2 g of Uracil, 2 g of 756 Alanine, 2 g of Arginine, 2 g of Aspartate, 2 g of Asparagine, 2 g of Cysteine, 2 g of 757 Glutamate, 2 g of Glutamine, 2 g of Glycine, 2 g of Histidine, 2 g of Isoleucine, 4 g of 758 Leucine, 2 g of Lysine, 2 g of Methionine, 2 g of Phenylalanine, 2 g of Proline, 2 g of Serine, 759 2 g of Threonine, 2 g of Tryptophane, 2 g of Tyrosine and 2 g of Valine. For growth in 760 glucose condition, the medium (SD) also contained 20 g/L of D-glucose. For growth in 761 galactose condition (induction of P GAL1 promoter), we added 2% final (20 g/L) raffinose and 762 2% final (20 g/L) galactose (SGalRaff medium). Media were adjusted to pH=5.8 by addition 763 of NaOH 1N before autoclaving at 0.5 Bar. For auxotrophic selections or P MET17 induction, 764 we used media where one or more of the amino-acids or nucleotides were omitted when 765 preparing S. For example, SD-W-M was made as SD but without any tryptophane or 766 methionine in the mix powder. We acquired data for 10,000 events per sample using a FACSCalibur (BD 786 Biosciences) or a MACSQuant VYB (Miltenyi Biotech) cytometer, after adjusting the 787 concentration of cells in PBS. We analyzed raw data files in the R statistical environment 788 (www.r-project.org) using custom-made scripts based on the flowCore package 70 from 789 bioconductor (www.bioconductor.org). We gated cells automatically by computing a 790 pGY519, yielding plasmid pGY523. Fourth, we inserted the HindIII-NotI cassette of pGY521 825 into the HindIII-NotI sites of pGY519, yielding plasmid pGY524. Fifth, a HindIII-BamHI 826 fragment of pGY523 containing one terminator, and a BglII-EcoRI fragment of pGY524 827 containing another terminator were simultaneously cloned as consecutive inserts in the BglII-828 EcoRI sites of 51269. Finally, the resulting plasmid was digested with HindIII and BamHI to 829 produce a fragment that was cloned into the HindIII-BglII sites of pGY524 to produce 830 pGY525. 831 To establish stable cell lines, Flp-In™ T-REx™ 293 cells were purchased from 832 Invitrogen (ThermoFisher) and transfected with both the Flp recombinase vector (pOG44, 833 Invitrogen) and pGY525. Selection of clonal cells was first performed in medium containing 834 300 µg hygromycin (Sigma). After two weeks, we identified foci of cell clusters, which we 835 individualized by transferring them to fresh wells. One of these clones was cultured for three 836 additional weeks with high concentrations of hygromycin (up to 400 µg) to remove 837 potentially contaminating negative cells. The resulting cell line was named T4-2PURE. 838 839 Lentivirus construct and production. A synthetic sequence was ordered from 840 Genecust and cloned in the HindIII-NotI sites of pCDNA3.1 (Invitrogen™ V79020). This 841 insert contained an unrelated additional sequence that we removed by digestion with BamHI 842 and XbaI followed by blunt-ending with Klenow fill-in. The resulting plasmid (pGY561) 843 encoded LiCre optimized for mammalian codon usage, in-frame with a N-ter located SV40-844 NLS signal. This NLS-LiCre sequence was amplified from pGY561 using primers Sauci and 845 Flard (Table S3), and the resulting amplicon was cloned in the AgeI-HindIII sites of the 846 Calculation of potential mean force (PMF). We calculated the free-energy profile 862 (reported in Figure 1f) for the unbinding of the C-terminal α-helix in the tetrameric Cre-863 recombinase complex 26 (PDB Entry 1NZB) as follows. The software we used were: the 864 CHARMM-GUI server 71 to generate initial input files; CHARMM version c39b1 72 to setup 865 the structural models and subsequent umbrella sampling by molecular dynamics; WHAM, 866 version 2.0.9 (http://membrane.urmc.rochester.edu/content/wham/) to extract the PMF; and 867 VMD, version 1.9.2 73 to visualize structures. To achieve sufficient sampling by molecular 868 dynamics, we worked with a structurally reduced model system. We focused thereby only on 869 the unbinding of the C-terminal α-helix of subunit A (residues 334:340) from subunit F. 870 Residues that did not have at least one atom within 25 Å from residues 333 to 343 of subunit 871 A were removed including the DNA fragments. Residues with at least one atom within 10 Å 872 were allowed to move freely in the following simulations; the remaining residues were fixed 873 to their positions in the crystal structure. For the calculation of the double mutant A340A341 874 the corresponding residues were replaced by alanine residues. The systems were simulated 875 with the CHARMM22 force field (GBSW & CMAP parameter file) and the implicit solvation 876 model FACTS 74 with recommended settings for param22 (i.e., cutoff of 12 Å for nonbonded 877 interactions). Langevin dynamics were carried out with an integration time-step of 2 fs and a 878 friction coefficient of 4 ps -1 for non-hydrogen atoms. The temperature of the heat bath was set 879 to 310 K. The hydrogen bonds were constrained to their parameter values with SHAKE 75 . 880 The PMF was calculated for the distance between the center of mass of the α-helix 881 (residues 334:340 of subunit A) and the center of mass of its environment (all residues that 882 have at least one atom within 5 Å of this helix). Umbrella sampling 76 was performed with 13 883 independent molecular dynamics simulations where the system was restrained to different 884 values of the reaction coordinate (equally spaced from 4 to 10 Å) using a harmonic biasing 885 potential with a spring constant of 20 kcal mol -1 Å -1 (GEO/MMFP module of CHARMM). 886 Note that this module uses a pre-factor of ½ for the harmonic potential (as in the case of the 887 program WHAM). 888 For each simulation the value of the reaction coordinate was saved at every time-step 889 for 30 ns. After an equilibration phase of 5ns, we calculated for blocks of 5 ns the PMF and 890 the probability distribution function along the reaction coordinate using the weighted 891 histogram analysis method 77 . A total of 13 bins were used with lower and upper boundaries at 892 3.75 and 10.25 Å, respectively, and a convergence tolerance of 0.01 kcal/mol. Finally, we 893 determined for each bin its relative free energy ! = − ln ! where k was the Boltzmann 894 constant, T the temperature (310 K) and ! the mean value of the probability of bin i when 895 averaged over the five blocks. The error in the ! estimate was calculated with ! ! = 896 ! ! / ! where ! ! was twice the standard error of the mean of the probability. An offset 897 was applied to the final PMF so that its lowest value was located at zero. 898 899 qPCR quantification of recombinase activity. We grew ten colonies of strain 900 GY1761 carrying plasmid pGY466 overnight at 30°C in SD-L-W-M liquid cultures. The 901 following day, we used these starter cultures to inoculate 12 ml of SD-W-M medium at OD 600 902 = 0.2. When monitoring growth by optical density measurements, we observed that it was 903 fully exponential after 4 hours and until at least 8.5 hours. At 6.5 hours of growth, for each 904 culture, we dispatched 0.1 ml in 96-well plate duplicates using one column (8 wells) per 905 colony, we stored aliquots by centrifuging 1 ml of the cell suspension at 3300 g and freezing 906 the cell pellet at -20°C ('Exponential' negative control) and we re-incubated the remaining of 907 the culture at 30°C for later analysis at stationary phase. We exposed one plate (Fig. 4j  908 'Exponential' cyan samples) to blue light (PAUL apparatus, 460 nm, 3.63 mW/cm 2 intensity) 909 for 40 min while the replicate plate was kept in the dark (Fig. 4j 'Exponential' grey samples). 910 We pooled cells of the same column and stored them by centrifugation and freezing as above. 911 The following day, we collected 1 ml of each saturated, froze and stored cells as above 912 ('Stationary' negative control). We dispatched the remaining of the cultures in a series of 96-913 well plates (0.1 ml/well, two columns per colony) and we exposed these plates to blue light 914 (PAUL apparatus, 460 nm, 3.63 mW/cm 2 intensity) for the indicated time (0, 2, 5, 10, 20 or 915 40 min). For each plate, following illumination, we collected and froze cells from 6 columns 916 (Fig. 4h samples) and we reincubated the plate in the dark for 90 min before collecting and 917 freezing the remaining 6 columns (Fig. 4i, x-axis samples). For genomic DNA extraction, we 918 pooled cells from 6 wells of the same colony (1 column), we centrifuged and resuspended 919 them in 280 µl in 50 mM EDTA, we added 20 µl of a 2 mg/ml Zymolyase stock solution 920 (SEIKAGAKU, 20 U/mg) to the cell suspension and incubated it for 1h at 37°C for cell wall 921 digestion. We then processed the digested cells with the Wizard Genomic DNA Purification 922 Kit from Promega. We quantified DNA on a Nanodrop spectrophotometer and used ~100,000 923 copies of genomic DNA as template for qPCR, with primers 1P57 and 1P58 to amplify the 924 edited target and with primers 1B12, 1C22 to amplify a control HMLalpha region that we 925 used for normalization. We ran these reactions on a Rotorgene thermocycler (Qiagen). This 926 allowed us to quantify the rate of excision of the floxed region as N Lox / N Total , where N Lox 927 was the number of edited molecules and N Total the total number of DNA template molecules. 928 To estimate N Lox , we prepared mixtures of edited and non-edited genomic DNAs, at known 929 ratios of 0%, 0.5%, 1%, 5%, 10%, 50%, 70%, 90%, 100% and we applied (1P57,1P58) qPCR 930 using these mixtures as templates. This provided us with a standard curve that we then used to