Optimized Vivid-derived Magnets photodimerizers for subcellular optogenetics

Light-inducible dimerization protein modules enable precise temporal and spatial control of biological processes in non-invasive fashion. Among them, Magnets are small modules engineered from the Neurospora crassa photoreceptor Vivid by orthogonalizing the homodimerization interface into complementary heterodimers. Both Magnets components, which are well-tolerated as protein fusion partners, are photoreceptors requiring simultaneous photoactivation to interact, enabling high spatiotemporal confinement of dimerization with a single-excitation wavelength. However, Magnets require concatemerization for efficient responses and cell preincubation at 28°C to be functional. Here we overcome these limitations by engineering an optimized Magnets pair requiring neither concatemerization nor low temperature preincubation. We validated these “enhanced” Magnets (eMags) by using them to rapidly and reversibly recruit proteins to subcellular organelles, to induce organelle contacts, and to reconstitute OSBP-VAP ER-Golgi tethering implicated in phosphatidylinositol-4-phosphate transport and metabolism. eMags represent a very effective tool to optogenetically manipulate physiological processes over whole cells or in small subcellular volumes.


INTRODUCTION
Macromolecular interactions between and amongst proteins and organelles mediate a considerable amount of biochemical signaling processes. A principal method of testing the physiological significance of such interactions is to drive their association with a user-supplied stimulus such as light or drugs. Typically, two different components, each fused to a specific protein, come together ("heterodimerize") to reconstitute a given protein-protein interaction following addition of a small molecule (DeRose et al., 2013;Putyrski and Schultz, 2012;Spencer et al., 1993) or upon light illumination (Losi et al., 2018;Rost et al., 2017). Light offers much greater spatial and temporal resolution than drugs, and as such, optogenetic dimerizers are generally used to probe phenomena at cellular and subcellular scales. At the organism scale, light is much less invasive but suffers from penetration issues.
One popular photodimerizer pair is "Magnets", engineered from the Neurospora crassa Vivid photoreceptor, which comprises an N-terminal Ncap domain responsible for homodimerization and a C-terminal light-oxygen-voltage-sensing (LOV) domain (Kawano et al., 2015). Magnets employ the ubiquitous cofactor flavin adenine dinucleotide (FAD) as the light-sensing moiety.
The Magnets pair was engineered from the Vivid homodimer by introducing complementary charges, giving rise to nMag (negative Magnet) and pMag (positive Magnet). The two Magnets components are quite small (150 aa) for photodimerizers, exhibit relatively fast association and dissociation kinetics, and function when fused to a broad range of proteins, including peripheral and intrinsic membrane proteins (Benedetti et al., 2018;Kawano et al., 2015Kawano et al., , 2016. Furthermore, heterodimerization of Magnets requires light-dependent activation of both components, rather than just one. This property results in low levels of background activity and allows induction of dimer formation with single-wavelength excitation in small cytoplasmic volumes (Benedetti et al., 2018).
However, the Magnets system has two prominent shortcomings. First, the low thermodynamic stability of the Magnets components precludes their proper expression and folding at 37oC. Thus, they cannot be used in mammals. When used in cultured mammalian cells they require a preincubation at low temperature (28oC) for 12 hours to allow expression and folding. Second, as the Magnets components heterodimerize with low efficiency, robust activation requires concatemerization (Furuya et al., 2017;Kawano et al., 2015), which may affect trafficking, motility and function of target proteins, create vector payload constraints, and give rise to recombination and/or silencing of the sequence repeats.
Here, we overcome these limitations of the Magnets by structure-guided protein engineering and validation by cellular assays. The resulting reagents, "enhanced Magnets" (eMags), have greater thermal stability and dimerization efficiency, as well as faster association and dissociation kinetics. We confirmed their effectiveness in a variety of applications including protein recruitment to different organelles, the generation/expansion of organelle contact sites, and the rapid and reversible reconstitution of VAP-dependent inter-organelle tethers that have key regulatory functions in lipid transport.

Optimization of the Magnets heterodimer interface
Optimal photo-heterodimerizer performance convolves together several parameters: i) Efficient, fast interaction of the two different components upon light stimulus, ii) little or no formation of homodimerswhich would compete with productive heterodimer complexes, iii) low background before light stimulus; and ideally, iv) fast heterodimer dissociation following light offset. The existing Magnets systems, especially the Fast1 and Fast2 variants with fast dissociation kinetics (Kawano et al., 2015), have weak dimerization efficiency and thus perform poorly on the first criterion, necessitating the use of concatemers (usually 3 copies) of either or both monomers to achieve acceptable reconstitution in a number of settings (Benedetti et al., 2018;Furuya et al., 2017;Kawano et al., 2015).
A pair with greater dimerization efficiency would be desirable, ideally allowing single copies of the complementary Magnets to suffice. With the goal of engineering such a pair, we first established a robust screen for reconstitution of Magnets dimerization using light-dependent accumulation of a protein at the outer mitochondrial membrane (Benedetti et al., 2018) (Fig.   1A), which is readily visible and quantifiable. The nMagHigh1 monomer, tagged with the green fluorescent protein EGFP, was used as bait on the outer mitochondrial membrane by fusion to the transmembrane C-terminal helix from OMP25 ("nMag-EGFP-Mito") (Supp. Fig. 1A and Supp. Table 1). The pMagFast2 monomer, tagged with the red fluorescent protein TagRFP-T (Shaner et al., 2008), was used as the cytoplasmic prey ("pMag- TagRFP Table 2). We co-expressed both constructs in HeLa cells by co-transfection, grew cells at 28°C for 24 hours, and tested light-dependent prey capture and release by the bait (Fig. 1B).
Importantly, excitation light for TagRFP-T, as well as that for mCherry and the infrared fluorescent protein iRFP (Shcherbakova and Verkhusha, 2013), is well outside the action spectrum of LOV domain proteins (400-500 nm light excitation) (Losi et al., 2018); EGFP excitation light is coincident with Magnets activation and is thus used sparingly in these experiments.
Next, we began the process of Magnets redesign by optimizing the placement of chargecomplementing amino acids in the Vivid dimer interface, using the crystal structure of the lightactivated dimer (PDB ID 3RH8) (Vaidya et al., 2011) (Supp. Fig. 3A-C) as a guide, and mitochondrial recruitment as the testbed. The original Magnets pair was built upon the mutations Ile52 and Met55 to Arg (positive Magnet) and Ile52 to Asp and Met55 to Gly (negative Magnet) within the Ncap domain (See Supp. Fig. 3A), which mediates dimerization. To achieve more efficient dimerization, we first sought to optimize charge placement at the interface. Substitution of Asp52 to Glu in nMag-Asp52Glu to modify the position of the negative charges somewhat disrupted heterodimerization, consistent with Kawano et al., 2015(Kawano et al., 2015. We next tried to introduce two negative charges into nMag, at the same two sites where positive charges had been introduced into pMag. nMag-Gly55Glu completely inhibited heterodimerization, whereas nMag-Gly55Asp somewhat improved it (Supp. Fig. 3D). Adding a third positive charge to pMag at position 48 also completely disrupted heterodimerization. In the end, we left the charges alone and instead sought to improve heterodimer interface packing and helical preference with nMag-Gly55Ala, which indeed improved both heterodimerization efficiency and association kineticsmore so than nMag-Gly55Asp. In fact, the nMag-Gly55Ala mutation alone sufficiently improved mitochondrial recruitment after preincubation at 28°C so that it functioned well as a monomer (Supp. Fig. 3D).

Thermostabilization of the Magnets proteins
Having improved the system to allow single-copy use at 28°C, we next sought to improve the temperature stability of the proteins to allow experiments at 37°C. As before, recruitment to the mitochondrial membrane in HeLa cells was used as the cellular assay: nMagHigh1-Gly55Ala-EGFP-OMP25 and pMagFast2-TagRFP-T were co-expressed on the outer mitochondrial membrane and in the cytoplasm, respectively, of HeLa cells by co-transfection. Identical amounts of DNA, in the same plasmid ratio, were used, to allow side-by-side quantification of expression level, background association in the dark, heterodimerization efficiency, and kinetics of association and dissociation. Cells were preincubated at 28°C, 33°C, 35°C, or 37°C for 12-24 hours and then imaged at 37°C to quantify mitochondrial accumulation. We made and tested a number of mutants (Supp. Fig. 3A, Supp. and Ser99Asn (all from thermophilic homologues) each improved dimerization efficiency at 28°C, and the latter allowed it at 33°C. Thr69Leu is in the interface and improves hydrophobic interactions (Supp. Fig. 5A,B), Met179Ile is in the hydrophobic core and improves packing (Supp. Fig. 5C,D), and Ser99Asn is surface-exposed and optimizes hydrogen bonding and secondary-structure preference (Supp. Fig. 5E,F). Combining these three mutations substantially increased dimerization at both 28°C and 33°C, and all further variants were tested on top of this combination. Mutations of Asn133 to lysine or phenylalanine (the latter from thermophiles) both enhanced dimerization at 33°C, with Asn133Phe facilitating it at 35°C, but with slower dissociation kinetics. The additional Tyr94Glu mutation (from thermophiles, improves helical preference) permitted weak dimerization at 37°C with dissociation kinetics comparable to the original Magnets molecules. The adjacent mutations Asn100Arg/Ala101His (from thermophiles, improves helical preference) allowed stronger 37°C dimerization. Finally, Tyr126Phe (from thermophiles, improves helical preference) and Arg136Lys (from thermophiles, improves helical preference, improves electrostatics with FAD cofactor; Supp. Fig. 5G,H) further increased dimerization efficiency.
We selected a pair of variants, eMags, with these nine mutations (Thr69Leu, Tyr94Glu, Ser99Asn, Asn100Arg, Ala101His, Tyr126Phe, Asn133Phe, Arg136Lys, and Met179Ile) added to nMagHigh1-Gly55Ala and pMagFast2. eMags supports dimerization upon growth at 37°C without preincubation at a lower temperature, while the original Magnets variants were completely nonfunctional after these growth conditions (Fig. 1C,D and Supp. Fig. 6). eMags show greater dimerization efficiency (~4-5x), as judged by greater prey accumulation on mitochondria (p=0.0004, Kruskal-Wallis and Dunn's multiple comparison post hoc tests; Fig.   1D) and faster association and dissociation kinetics (τON = 3.6  0.3 s, τOFF = 23.1  0.6 s) than original Magnets in cells preincubated at 28°C (τON = 7.6  0.3 s, τOFF = 32.0  1.3 s; p < 0.0001 for both τON and τOFF, unpaired Student's t-test; Fig. 1C). Omission of the Tyr126Phe mutation in eMags produced eMagsF, with similar but slightly lower dimerization efficiency as eMags, but significantly faster association and dissociation kinetics (τON = 2.8  0.3 s, τOFF = 14.0  0.6 s; p < 0.0001 for both τON and τOFF, unpaired t-test; Fig. 1C). A 3x prey concatemer (i.e. nMagHigh1-EGFP-OMP25 and pMagFast2(3x)-TagRFP-T)still requiring preincubation at 28°Cis needed to bring the prey recruitment of original Magnets in line with that of monomeric eMags and eMagsF (Fig. 1D). This concatemerized original Magnets also suffers from slower dissociation kinetics (τON = 5.6  0.5 s, τOFF = 45.9  1.4 s; p = < 0.0001 for both τON and τOFF, unpaired t-test; Fig. 1C,D). We refer to nMagHigh1-Gly55Ala and pMagFast2 with these nine mutations as eMagA (Acidic heterodimerization interface) and eMagB (Basic heterodimerization interface), respectively. eMags enables rapid, local and reversible control of protein recruitment to subcellular compartments We then sought to establish performance of the new eMags constructs in a variety of experimental contexts. In the first, we used eMags to conditionally recruit cytosolic proteins to intracellular organelles other than mitochondria. For the endoplasmic reticulum (ER), we selected the N-terminal transmembrane domain of cytochrome P450 (Szczesna-Skorupa and Kemper, 2000), which displays on the cytoplasmic face of the ER, as bait (fused to EGFP). Coexpression of this construct, ER-EGFP-eMagA, with eMagB-TagRFP-T (prey) in COS7 cells showed large, rapid, reversible accumulation of prey to the ER upon whole-cell illumination ( Fig. 2A, Supp. Fig. 2B) (See Supp. Fig. 1A,B, Supp. Tables 1,2, Methods for a complete list and detailed information on bait and prey constructs used in these experiments). With focal illumination, robust prey accumulation occurred only in the irradiated ER region (Fig. 2B), in spite of the known rapid diffusion of proteins within the ER network (Nehls et al., 2000).

Optogenetic regulation of inter-organellar contacts
In another set of applications, we validated the efficiency of eMags to induce organelle contacts ( Fig. 3A, Supp. Fig. 1D). Conditional induction or expansion of such contacts may help elucidate the contribution of inter-organelle contacts and signaling to a variety of biochemical pathways.
We first designed a light-inducible ER-lysosome tethering system. Using the targeting sequences above (Fig. 2), ER-mCherry-eMagA and Lys-eMagB-iRFP were co-transfected into COS7 cells. Before blue light activation, ER-lysosome overlap, as detected by mCherry and iRFP overlap, was minimal (Fig. 3B); during 1 min. irradiation, colocalization rapidly increased by ~50% (τON =7.5  0.8 s, N=14 cells, 3 independent experiments), most likely through expansion of pre-existing contacts or by stabilization and expansion of new contacts. Following light offset, ER-lysosome colocalization declined quickly to baseline (τOFF = 35.9  1.7 s; Fig.   3B). The longer time courses of organelle association-dissociation (tens of seconds), relative to cytoplasmic protein recruitment (seconds), is consistent with a combination of slower mobility of organelles than free protein and the processive assembly and disassembly of membrane contacts.
Using a similar targeting strategy, ER-mCherry-eMagA and eMagB-iRFP-Mito were used to drive ER-mitochondrial association (Fig. 3C). In HeLa cells, used for these experiments, ER and mitochondria form a closely interacting network even in control conditions. Upon 2 min. irradiation, however, overlap increased by ~20%, with kinetics (τON = 28.0  1.9 s, τOFF = 49.1  2.5 s, N=14 cells, 3 independent experiments; Fig. 3C) on the order of that seen for ER-lysosome contacts.
Finally, for mitochondrion-lysosome manipulation, we used eMagA-mCherry-Mito and Lys-eMagB-iRFP. In HeLa cells, baseline colocalization was quite low (Fig. 3D); such contacts are typically transient and involve small contact area (Wong et al., 2018). Upon activation, increased associations between lysosomes and mitochondria were observed, revealing contact expansion (τON = 40.1  2.6 s, τOFF = 58.4  2.6 s, N=17 cells, 3 independent experiments). In some cases, movement of lysosomes away from mitochondria resulted in the elongation of tubules from mitochondria, and even in their fission (Fig. 3E), indicating strong association.

Control of the PI4P Golgi pool by reconstitution of VAP (Opto-VAP)
In a final application, we tested eMags for acute manipulation of intracellular PI4P via reconstitution of an ER-transGolgi network (TGN) tether. Key components of this tether are the ER protein VAMP-associated protein (VAP) and Oxysterol-binding protein 1 (OSBP1). OSBP1, which binds VAP (via an FFAT motif) and membranes of the TGN (via a PI4P-binding PH domain), also contains an ORD domain (OSBP-related domain) that promotes exchange of TGN PI4P for ER cholesterol (Murphy and Levine, 2016). Following shuttling to the ER, PI4P is degraded by the phosphatidylinositide phosphatase Sac1 (Mesmin et al., 2013;Saint-Jean et al., 2011;Zewe et al., 2018). This model of ER-Golgi PI4P transport is supported by biochemical, pharmacological, and genetic studies (Dong et al., 2016;Mesmin et al., 2013;Strating et al., 2015). We sought to use the eMags tools to offer direct optogenetic control over this PI4Pcholesterol exchange through regulation of VAP-OSBP1 binding interactions.
The overall design strategy was to replace endogenous VAP with a split version, which could be reconstituted by eMag dimerization and would then associate with OSBP1 to drive transport.
Unlike the earlier examples, this necessitated careful consideration of the domain architectures of VAP and OSBP1, to best ensure that 1) split-VAP would not reconstitute in the absence of light activation and 2) that the eMagA and eMagB fusions would not interfere with either VAP reconstitution or OSBP1 interaction. VAP is an integral membrane protein composed of a cytosolic major sperm protein (MSP) domain (which binds FFAT motif-containing proteins), a coiled-coil domain and a C-terminal membrane anchor (Kaiser et al., 2005;Kim et al., 2010) ( Fig. 4A and Supp. Fig. 1E). Two distinct VAP genes exist in the vertebrate genome: VAPA and VAPB, which can form either homomers or heteromers with one another. OSBP1 has an Nterminal PH domain that preferentially binds PI4P (Mesmin et al., 2013;Murphy and Levine, 2016;Venditti et al., 2019), an internal FFAT motif, and a C-terminal ORD domain which binds in a competitive way PI4P and cholesterol.
Given this domain structure, we opted to convert VAPB into a cytosolic version through deletion of the C-terminal transmembrane helix (leaving VAPB(1-218)); we retained the MSP and coiledcoil domains as both may contribute to VAP dimerization (Kim et al., 2010) (Fig. 4B). We fused TagRFP-T to the N-terminus of this cytosolic fragment, and eMagB to its C-terminus (TagRFP-T-VAPB(1-218)-eMagB; Fig. 4B, Supp. Fig. 1E and Table S2). We then used ER-eMagA-EGFP to recruit VAPB(1-218) to the ER upon blue light irradiation, where it could interact with OSBP1.
We refer to this pair of constructs as "Opto-VAP".
We first tested the efficiency of Opto-VAP by transfecting both components into HeLa cells and imaging them by confocal microscopy. The prey protein (TagRFP-T-VAPB(1-218)-eMagB) was imaged throughout the experiment, while ER-eMagA-EGFP was imaged only during optogenetic activation. Before blue light irradiation, the prey protein was homogeneously distributed throughout the cytosol, with focal accumulation around the Golgi (Fig. 4C). We interpret this observation as reflecting interaction of VAPB with endogenous OSBP1, which is abundant in the Golgi, where it binds the PI4P-rich TGN membranes via its PH domain (Mesmin et al., 2013).
The cytosolic VAPB(1-218) prey, with its MSP domain, could compete with endogenous VAP for binding to the FFAT motif of OSBP1 (Fig. 4A,B). A robust presence of PI4P in the TGN under To confirm that the observed PI4P transfer was indeed mediated by OSBP and Opto-VAP, cells were preincubated for 30 min with 10 μM itraconazole (ITZ), an antifungal and anticancer agent that occludes the lipid-transport domain of OSBP and thus blocks its lipid trafficking properties (Strating et al., 2015). After ITZ treatment, no change was detected in the accumulation of the PI4P probe (iRFP-P4C) at the Golgi (graph in Fig. 4C and Supp. Fig. 7, 8A), despite the efficient recruitment of TagRFP-T-VAPB(1-218)-eMagB to the ER membrane (N=16 cells, 2 independent experiments).
We next tested the Opto-VAP system in gene-edited HeLa cells lacking both VAP genes (VAP double-KO cells). It was reported that in these cells the Golgi complex is partially disrupted, with formation of PI4P-enriched hybrid Golgi-endosome structures (Dong et al., 2016), a finding that we have confirmed in cells kept in the dark (Fig. 4D-bottom). Blue light activation led to rapid recruitment of TagRFP-T-VAPB(1-218)-eMagB to the ER (Supp. Fig. 8B), whose reticular appearance was less obvious in these cells (Fig. 4D- Golgi-endosome structures was observed (Fig. 4D), indicating PI4P loss. Thus, Opto-VAP is able to fully restore the activity of the deleted VAPA and VAPB genes in recruiting OSBP1 to perform PI4P-cholesterol exchange. After blue-light interruption, both Opto-VAP localization and PI4P levels reversed to baseline (τOFF = 93.7  5.0 s) (Fig. 4D) (N=20 cells, 4 independent experiments). As before, ITZ completely inhibited PI4P transport but had no effect on Opto-VAP recruitment (N=16 cells, 3 independent experiments) ( Fig. 4D and Supp. Fig. 8B). The time courses of Opto-VAP recruitment and recovery, and of PI4P loss and recovery, are similar between the wild-type and double-KO cells, suggesting that Opto-VAP assembly and function are largely independent of endogenous levels of VAPA and VAPB.
As a final verification of the necessity of the ORD domain in the observed PI4P transport, we constructed TagRFP-T-eMagB-PHOSBP, with the PH domain of OSBP1 but not the ORD domain ( Fig. 4A, Supp. Fig. 1E, Supp. Fig. 9A and Table S2). In both wild-type or VAP-DKO HeLa cells, blue-light activation induced rapid prey recruitment to the ER, but with no accompanying changes in iRFP-P4C fluorescence (Supp. Fig. 9B,C; n=16 cells for HeLa, n=17 for VAP-DKO, 2 independent experiments). Thus, the ORD domain is critical for PI4P transport, with the PH domain alone having no effect.

CONCLUDING REMARKS
In this work, we have both engineered a dramatically improved photodimerizer pair and used it in a set of experiments elucidating details of organellar interactions and cellular lipid metabolism and transport. In a previous study (Benedetti et al., 2018), we had compared multiple optogenetic dimerizer reagents and found that the Magnets system, based on orthogonalization of the Vivid LOV domain homodimer (Kawano et al., 2015), offers major advantages over other systems in several different assays. Magnets have rapid association and dissociation kinetics and require both monomers to undergo blue-light activation to permit dimerization. These properties make the background activation of Magnets low, so that they are well-suited to optogenetic modulation of small volumes and sub-cellular organelles. However, the existing Magnets tools have two critical disadvantages, which preclude their wider adoption: 1) their weak dimerization efficiency necessitates the use of concatemers, which can perturb target proteins and slow kinetics, and 2) the low thermodynamic stability means that expression and maturation must occur at reduced temperatures, complicating cell culture experiments and ruling out mammalian in vivo work entirely.
To overcome these limitations, we established a robust cell-culture screen that captures dimerization efficiency, association and dissociation kinetics, and folding and maturation. This screen allowed us to identify variants encompassing mutations across the whole protein with particular focus on the dimer interface. Mutations were selected based on sequence alignments with thermophilic fungal Vivid domains and structure-guided design. After several rounds of mutagenesis and screening, we selected final "enhanced Magnets" (eMag) variants with nine mutations over the starting scaffolds. The eMag reagents showed greater dimerization efficiency allowing use as monomers instead of concatemers, full function after their folding and maturation at 37oC, and faster association and dissociation kinetics than the original Magnets.
We thoroughly validated the eMag constructs in a range of cellular assays involving protein recruitment to different membranes, inter-organellar association, and bilayer lipid metabolism and trafficking. The success of the engineering effort validates the design strategy and shows that many mutations from thermophilic fungi grafted well to the scaffold of the Vivid photoreceptor of Neurospora crassa, a mesophilic fungus. These mutations improved packing, hydrogen bonding, and secondary structure preference. These improved optogenetic dimerizers will be broadly applicable and useful for applications across diverse fields.

B. Localized and global recruitment of a soluble prey to an ER-targeted bait in a HeLa cell.
Localized activation was achieved by illuminating the cell within a 3 µm x 3 µm ROI with 200 ms blue-light pulses at 0.5 Hz for 60 seconds. The cell was then allowed to recover in the absence of blue light for 2 min prior to global illumination. Scale bar: 5 m.
C. Recruitment of a soluble prey to lysosomes in a DIV14 primary hippocampal neuron. The left two fields show colocalization of the lysosomally anchored bait with the lysosomal marker Lamp1-iRFP. Recruitment of the prey to a single lysosome, or to all lysosomes, was achieved by local and global illumination, respectively. Following localized illumination delivered as in (B), the cell was allowed to recover in the absence of blue light for 1 min, and then globally illuminated. Scale bar: 5 m.
D. Schematic representation of the strategy and constructs used to induce PI(4,5)P2 depletion at the plasma membrane via the eMagF-dependent recruitment of an inositol 5-phosphatase. iRFP-PHPLCδ is a PI(4,5)P2 probe.   during, and after Opto-VAP activation, with or without ITZ treatment (N=20 and 16, respectively; from 3 independent experiments).

TABLES AND TABLES LEGENDS
Supp.
nMagHigh1-EGFP-Mito was generated through the PCR amplification of the nMagHigh1-EGFP coding sequence from nMagHigh1-EGFP-CAAX, and inserted into a pGFP-OMP25 (Nemoto and De Camilli, 1999) vector at NheI and XhoI sites. pMagFast2(1x)-TagRFP-T was generated through the PCR amplification of the third unit of pMagFast2(3x) and TagRFP-T in pMagFast2(3x)-TagRFP-T (Benedetti et al., 2018) and inserted in the same vector at HindIII and XbaI site. In order to recreate an optimal Kozak sequence Met and Gly were added before the initial His, at the N-term of pMagFast2 in this construct. All nMagHigh1 and pMagFast2 mutants eMagAF-EGFP-PM and eMagA-EGFP-PM were generated replacing nMagHigh1 in nMagHigh1-EGFP-CAAX with the engineered variants at HindIII and XbaI sites. mCherry-eMagBF -5ptaseOCRL was synthetized by digesting mCherry-pMagFast2(3x)-5ptaseOCRL (Benedetti et al., 2018) with NotI and PvuI, and then ligated with eMagBF amplified from eMagBF-TagRFP-T. iRFP-PHPLC plasmids were previously described (Idevall-Hagren et al., 2012 Light-dependent induction of contacts between ER and lysosomes was achieve transfecting COS7 cells with ER-mCherry-eMagA and Lys-eMagB-iRFP at a 2:1 ratio in OptiMEM-I (1:4 DNA: lipofectamine ratio). ER-mitochondria contacts were elicited in HeLa cells transfected with ER-mCherry-eMagA and eMagB-iRFP-Mito at a 1:1 ratio in OptiMEM-I (1:4 DNA: lipofectamine ratio). Mitochondria-lysosome contacts were evoked in HeLa cells transfected with eMagA-mCherry-Mito and Lys-eMagB-iRFP. Cells were incubated with the transfection mix for 1 hour. Subsequently, the serum-free medium was replaced by complete DMEM with no phenol red, and imaging was performed in the same medium between 16 and 28 hours after transfection.

Confocal microscopy
All optogenetic experiments, with the exception of the experiments with Opto-VAP and its controls and the light-dependent induction of inter-organellar contacts, were performed using the Improvision

Image Analysis and Statistics
Association and dissociation rates for each dimerization system were calculated from changes in prey fluorescence inside a cytosolic ROI before, during, and after the photoactivation and Statistical analyses were carried out in GraphPad Prim 8.2.1 (Graph Pad Software).

Kinetics analysis
We found that the apparent kinetics of the Magnets variants reported in this study fit well to an exponential decay model. We used the curve-fitting tool (cftool) in MATLAB to determine the kinetic rate constants, τON and τOFF, by fitting the curve to the following equation: [ ] = 0 + ∆ − − 0 Where = ∘ , 0 is time at which the light is turned on or off (for on-or off-kinetics, respectively), S0 is S at time t0, and ∆ = 0 − (∞). During the fitting process, each point is given a weight proportional to 1 . . . 2 . The parameters of the fit can be found in Supplementary  Table 4. For all the datasets acquired in this work, the R2's obtained for exponential fits are always larger than 0.86 with a median of 0.98.