A photo-SAR study of photoswitchable azobenzene tubulin-inhibiting antimitotics identifying a general method for near-quantitative photocontrol

Azobenzene analogues of the tubulin polymerisation inhibitor combretastatin A4 (PSTs) were previously developed to optically control microtubule dynamics in living systems, with subsecond response time and single-cell spatial precision, by reversible in situ photoswitching of their bioactivity with near-UV/visible light. First-generation PSTs were sufficiently potent and photoswitchable for use in live cells and embryos. However, the link between their seconds-scale and hours-scale bioactivity remained untested. Furthermore, the scope for modifications to tune their photo-structure–activity-relationship or expand their function was unknown. Here, we used large-field-of-view, long-term tandem photoswitching/microscopy to reveal the temporal onset of cytostatic effects. We then synthesised a panel of novel PSTs exploring structural variations that tune photoresponse wavelengths and lipophilicity, identifying promising blue-shifted analogues that are better-compatible with GFP/YFP imaging. Taken together, these results can guide new design and applications for photoswitchable microtubule inhibitors. We also identified tolerated sites for linkers to attach functional cargos; and we tested fluorophores, aiming at RET isomerisation or reporter probes. Instead we found that these antennas greatly enhance long-wavelength single-photon photoisomerisation, by an as-yet un-explored mechanism, that can now drive general progress towards near-quantitative long-wavelength photoswitching of photopharmaceuticals in living systems, with minimal molecular redesign and broad scope.


MT biology needs spatiotemporally precise reagents
The protein tubulin is dynamically polymerised to form microtubules (MTs), major cellular scaffolds whose structure and remodelling underlie diverse processes from intracellular transport and cellular motility to the function of the mitotic spindle during proliferation. 1 Small molecule drugs that interfere with MT structure and dynamics are indispensable as tools for research in and around cytoskeletal biology. 2 These include taxanes, vinca alkaloids, 3,4 and inhibitors binding at the colchicine site, e.g., colchicine, combretastatin A-4 (CA4, Fig. 1a), and their closely isosteric analogues 5 (for more detail, see ESI Note 1 †).However, it is near-impossible to use these drugs to study particular roles of MTs in specic cell regions, cell populations, or tissues, at precisely dened times, since their activity cannot be spatially or temporally directed. 6eveloping spatiotemporally-targetable MT modulators has thus been an important goal, and optical control strategies have been prioritised since light can be applied with high spatiotemporal precision even in vivo. 7,8Although a few optogenetic MT-modulating tools were recently created, [9][10][11] most research has focused on chemical tools.3][14][15][16][17] However, irreversible approaches cannot overcome the diffusion of the active drug, so their spatiotemporal resolution is limited.

Photoswitchable MT reagents
Spatially and temporally patterning bioactivity instead demands reversible switching in situ in live cells, over many off 4 on cycles.Azobenzene-based "PSTs" that are structural analogues of CA4 (Fig. 2a) are the most widely-applied photopharmaceutical MT tools.PSTs can be reversibly photoswitched between bio-inactive trans and >100-fold more potent cis isomers, 18 with full photostability, by low intensity near-UV and visible light: a combination of features that out-performs almost 19 all other photoresponsive MT inhibitors. 20In cell biology, PSTs can photocontrol MT architecture, dynamics, and many MT-dependent processes with excellent spatiotemporal precision. 21,22Streu, 23 Hartman, 24 Rastogi and Brittain 25 reported complementary aspects of PST biochemistry; and we applied soluble in vivo-compatible prodrugs PST-1P and PST-2S (Fig. 2a) to study complex phenomena in cultured cells 26,27 and in developing embryos. 28,29eds for improving MT reagents Despite these applications, 21 there is much room to improve the utility of PSTs as tool compounds.For this, the chemical space for PSTs must rst be characterised and understood.Though structure-activity-relationships (SAR) for combretastatin-like stilbenes are well known, 30 cis-PST potency is affected by unique factors that do not apply to the isosteric stilbenes (e.g., PSTs have >20-fold lower potency than expected, for reasons that are still unclear), and PST cis % trans switching adds new dimensions of structure-property relationships that must be balanced for performance.Therefore, we aimed to explore SAR and photophysical tuning of bioactive PST derivatives in a "photo-SAR" or PSAR study, in three areas: (i) classical SAR, to nd the restraints on cis-PST structure needed to keep their potency useful within their solubility range; (ii) chemical biology, to nd what modications for developing multifunctional tool compounds are tolerated; (iii) photochemistry, to modify their practical performance (e.g., incremental absorption band shiing through substitutions, or larger modications that test the limits of what photostationary states can be accessed in bioactive derivatives).Rastogi and Brittain 25 have made the only other systematic study of PST derivatives, though mostly this was limited to varying meta-substituents on the B ring: a separate scope of structural explorations that complements the PSAR developed here.
Finally, even the link between rapid-onset MT-inhibiting effects seen in live cell imaging assays (<seconds), and the long-term antimitotic effects that are the only cellular readout assessed in most photopharmacology studies (since easily and cheaply done in parallel throughput), has not been convincingly explored for this compound class (or indeed for most other photopharmaceuticals).However, it is necessary to conrm that the transition from short-term to long-term effects is indeed mechanistically linked, before one can be condent that cheap long-term assays can guide reagent development for the painstaking short-term imaging and photocontrol studies where reagents are ultimately supposed to perform.We therefore resolved to begin by monitoring this transitional link in time.

Time-dependency of photopharmaceutical effects
Like many photopharmaceuticals, PSTs undergo Z / E relaxation over time.In short-term assays where MT dynamics are monitored directly, this is usually irrelevant since high-intensity bidirectional isomerisations are used to jump between photostationary states (PSSs).However, in long-term assays (>24 hours) where indirect, downstream effects such as cell death are monitored, illuminations are oen applied by low intensity light pulses every 1-5 minutes, aiming to build up and maintain a PSS throughout the experiment.As far as we know, no studies have explored the time-dependency of these readouts, so we were motivated to nd out (1) how much E / Z illumination time is needed for cellular responses to be detectable via the usual long-term readouts (combines progressive E / Z isomerisation with cell inhibition); and (2) how much time is needed during dark phases for cells to resume normal behaviour (combines Z / E relaxation with biological recovery).Such experiments would be highly time-consuming in endpoint assays (needs a great many endpoints to trace evolution over time), or in typical microscopy (small elds of view mean only few cells can be tracked in parallel).However, reconstruction-free lensless microscopy (RFLM) is ideally suited to track large numbers of cells non-invasively in parallel in such longitudinal assays.RFLM setups can be very compact (10 × 10 × 15 cm) and can simply be placed in an ordinary cell culture incubator to image and track thousands of cells per condition, 31 over many conditions in parallel (here, eight).We modied an RFLM setup to co-apply 380 nm light at typical long-term lowintensity settings, during dened phases, for in situ E / Z photoswitching.
With short photoactivation periods (20 min, 1 h) no effects on proliferation were detected.However, aer ca. 3 h of photoactivation, antimitotic effects were notable, since growth rates (rate of change of conuency over time) became strongly dependent on presence or absence of PST (Fig. 1, red brackets).This reassured us that the PSAR of PSTs would be reliably assessed in typical long-term assays (24-48 h), even though we expected them to have differences in their efficiencies of bulk photoisomerisation and rates of Z / E relaxation.
Photo-structure-activity-relationship (PSAR) study PSAR panel design.The SAR of stilbene-type combretastatin analogues, and the X-ray structure of the combretastatintubulin complex, suggest that one A-ring ortho position, and a neighbouring ortho and meta position on the B-ring, should be the most tolerant of substitutions: 30,32 but this PSAR has not been tested for azobenzene analogues.We synthesised a panel of 29 novel azobenzenes and azoheteroarenes (PST-6-32, Fig. 2) to test these structural variations, primarily by diazonium couplings (see ESI †).Their photophysical properties, including E and Z isomer absorption spectra, PSS compositions, wavelengths giving the highest proportion of Z isomer, reversibility of photoswitching, and Z / E relaxation half-lives t 1/2 in aqueous media, were determined in vitro 21,33 (key compounds in Fig. 2c, others in Table S1 † and Fig. S1-S3 †).
PSAR panel cell screening.The compounds were screened for cytotoxicity in the HeLa cervical cancer cell line under dark vs. 390 nm-illuminated conditions (Fig. 3 and S9 †). 21Our  S1. † priority was not to maximise Z-isomer potency but rather to identify the structural tolerance for photoswitchably bioactive PSTs, that are potent as the Z isomer (illuminated) yet have low toxicity as the E isomer (dark): which is needed for photoswitchable tools.ESI Note 1 † gives a full PSAR discussion.In brief however, for photoswitchability of bioactivity, the B ring tolerates small low-polarity groups at the meta-position (active bromo 8 but lower-potency alkoxy 9/10, inactive carboxylate 12); ortho-substituents are disfavoured although still have activity (6 vs. 7); and if small, double substitution is allowed (diuoro 16).Matching this, larger rings replacing the phenyl system that were reported for stilbenes [34][35][36][37] were not tolerated in the azobenzenes (19-21); so their alluring red-shied photoresponse (Fig. 2c) could not be harnessed.The A ring also tolerates small low-polarity ortho-substituents (22, 23); but neither isosteric nor smaller meta groups are tolerated if they are polar (24-26; Fig. 3b).
Within this SAR range, multiple tolerated modications can be made simultaneously.For example, PST-27 was designed for blue-shied photoresponse (less electron-rich), by replacing three methoxy groups with methyl/ethyls; this extra hydrophobicity required a B ring meta hydroxyl for solubility but became our most potently photoswitchably bioactive compound.This led us to prepare its highly water-soluble phosphate ester prodrug PST-27P for more reliable use in biology.We conrmed its strong wavelength-dependency of cytotoxicity over the entire range 360-535 nm (Fig. 3c and d), supporting our premise that PST analogues' toxicities are determined by their in situ photoconversion to the bioactive cis isomer at PSS under the wavelength applied.Being blue-shied, PST-27P had twice as good orthogonality to GFP/YFP excitation wavelengths (6-to-8 fold lower activity than max, Fig. 3d) as compared to known PST-1 (3-to-4 fold), which may make it more useful in many practical settings. 21Unfortunately, the yet more blue-shied PST-31 lost too much potency to be a useful reagent; but we expect other strategies for further blue-shiing could be successful (e.g. a 3,4,5-triethyl A ring) in future work.

Conrmation of cellular mechanism of action
We next tested if the new PSTs' cis isomers retain tubulin inhibition as their primary mechanism of action (as for Z-PST-1-5). 21Indeed, PST-27P gave light-specic disruption of MT network organisation at its cell viability EC 50 , and gross MT depolymerisation above it (confocal imaging, Fig. 4a), as did other well-performing compounds (PST-8 and PST-16, Fig. 4b).PST-27 was progressed to further mechanistic tests, where it strongly inhibited microtubule assembly in vitro under 390 nm illumination, almost without perturbing polymerisation in the dark (Fig. 4c).Cellular readouts downstream from MT inhibition were consistent with this mechanism of action: PST-27P gave light-specic accumulation of a sub-G 1 (dying) population (Fig. 4d) consistent with the viability assay, accompanying a distinctive light-specic G 2 /M phase arrest expected for cellular inhibition of tubulin dynamics (Fig. 4e and f).Finally, we obtained an X-ray crystal structure of the metastable Z-PST-27 bound to tubulin, conrming the same binding site and pose as the reference stilbene CA4 (Fig. 4g; PDB: 9F8G).Thus, we concluded that the Z-PST binding pose, SAR, and mechanism of action, are validated as matching known colchicine-site inhibitors. 30scovery of long-wavelength assisted photoswitching Fluorophore conjugate design.We had seen so far that larger groups are best tolerated in ortho at the A or B rings, and we aimed to exploit this to create photoswitchably tubulin-binding PST conjugates with functional cargos.Here, we tested uorophore conjugates that might use resonant energy transfer (RET) to drive E $ Z isomerisation (ESI Notes 2-3 †), or else might be light-dependent MT imaging agents (e.g., E: distributed, Z: MT-bound).In brief, we attached uorophores via linkers to low steric demand ether and anilide groups at the ortho position of the A ring.We aimed for uorophores with minimal absorption at 390 nm (Fig. S5a †), so that the conjugates might reach similarly Z-rich PSSs under UV illumination as the parent azobenzenes (later conrmed: Fig. 5b) and so might be bioactive against tubulin.
We chose a nitrobenzoxadiazole uorophore (NBD, ca.450/ 550 nm ex/em) whose <510 nm emission tail overlaps with the absorption of E & Z-PSTs (Fig. S1 and S2 †), for conjugate MR110, expecting to drive some isomerisation by RET.For conjugate MR69, 18 we instead chose an environment-independent rhodamine (RhB, ca.530/560 nm ex/em; as a secondary amide to prevent forming a nonuorescent spirolactam; Fig. 5a) where we expected that under excitation at e.g.550 nm, its tiny or ∼zero emission overlap with azobenzene Z or E isomer absorption (<540 nm) ought to make RET isomerisation either Z / E selective but slow, or else impossible (Fig. S5b †).
Conjugates give assisted photoswitching.Our expectations were dramatically overturned, in that MR69 in particular was very efficiently isomerised to high-E-PSSs by exciting the RhB uorophore (540-590 nm; Fig. S4 †), although the azobenzene alone does not respond to these wavelengths (cf.Fig. 5d: PST-30 at $560 nm).In brief, key features of the conjugates include: (1) they have fast, efficient, E / Z photoswitching under UV light, as usual for azobenzenes (385 nm; Fig. 5b-d).( 2) Exciting the uorophore motif gives exceptionally efficient switching (Fig. 5c).MR69 Z / E photoswitching above 550 nm approaches its PSS even more photon-efficiently than does its parent azobenzene under UV light (rough switching rates for >550 nm switching of MR69 and 400 nm switching of PST-30 are the same; Fig. 5d).( 3) Conjugate switching by exciting the uorophore can be exceptionally complete: MR69 reaches 94 ± 3% E isomer at PSS 554 nm (typical azobenzenes reach ca.80% E with best-chosen cyan/green light wavelengths).
Assisted switching has promise to solve general problems in photopharmacology.The conjugates' performance is exciting.Their switching at long wavelengths can be more complete, and (more importantly) orders of magnitude more photon-efficient, than direct Z / E photoswitching of the parent azobenzene at its typical wavelengths (Fig. 5d: 20 to 400-fold faster switching of MR69 at 540-560 nm, than green light switching of PST-30 at 520-540 nm): yet, direct E / Z photoswitching is not much impaired.RhB conjugate Z / E switching operated efficiently even to ∼600 nm, which has much better penetration into tissue than wavelengths azobenzenes typically harness.
As the uorophore motifs were capturing the energy used for this switching, we named this effect "assisted" isomerisation.Excitingly, this can address general challenges that have troubled photopharmacology: e.g., how to achieve efficient isomerisation at long wavelengths, with a simple design for predictable photoresponse (here, at whatever wavelengths the uorophore motif absorbs), without requiring drastic photoswitch redesign that blocks many substituent positions.For a brief discussion of the assisted switching design, rationale, and outcomes, see ESI Notes 2-5 †; but since the mechanisms behind this effect turned out to be more complex than simple RET, we report their elucidation in two separate papers. 39,40Because azobenzenes are popular quenchers for uorescent probes, we think it likely that such assisted switching will have been ongoing before: but it seems that it was not necessarily measured and reported as such; and as far as we are aware, such conjugates have never before been trialled for photoswitching in biology under single photon excitation at wavelengths longer than those at which azobenzenes usually absorb.Thus, we briey examined their biological performance.
As expected, installing linkers to give PST-29/30 lowered potency (ether PST-30 was still somewhat photoswitchably bioactive; Fig. 6a).Unfortunately, the cellular localisation of conjugates MR69 and MR110 was dominated by their uorophores' intrinsic distribution: delocalised lipophilic cation MR69 to mitochondria (Fig. 6b), 41 and hydrophobic NBD MR110 in lipid vesicles (Fig. 6c): 41 thus, unsurprisingly, they did not stain MTs even aer UV illumination.To test their targetbinding, cell-free tubulin polymerisation assays were run at high 21 concentrations, but did not show inhibition (Fig. 6d and  e).We thus halted investigations, realising that substantial tuning would be needed for cellularly-useful assisted-switching tubulin photopharmaceuticals, regarding both the assisting chromophore and the photoswitchable ligand (discussion at Conclusions and in ESI Note 4 †).

Conclusions
Motivation PSTs were previously used in biology studies to disrupt MT polymerisation dynamics with spatiotemporal patterning, and so to apply spatiotemporally-dened gradients of antimitotic, anti-migratory, and cytotoxic effects in diverse cellular and in vivo models.Still, to unlock higher performance in cytoskeleton photopharmacology, designs with drastically improved photoswitching completeness and efficiency, higher potency, and/or that feature cargos allowing more elaborate chemical biology applications, are needed.

Photo-SAR
This photo-SAR study identied tolerances for modications to the PST scaffold which can now be used to adapt them for these needs: e.g., installing response to specic wavelengths, or avoiding it (wavelength orthogonality, e.g., blue-shied PST-27); or tuning spontaneous relaxation speeds and polarity (Fig. 2); which can be performed while keeping the Z-specic tubulinbinding cellular mechanism of action (Fig. 3 and 4).This will support developing higher-performance PST reagents for microtubule studies; and we believe that even more blue-shied (more GFP-orthogonal) reagents will prove to be the most valuable of those reagents.

Assisted switching with colchicinoids
We also found positions that may be suitable for attaching payloads, and used these to create uorescent conjugates.We found these offered powerful solutions to the longstanding practical challenges of redshiing and high-completion isomerisation, which have hindered photopharmacology from in vivo application.The elucidation of the assisted switching mechanism is being reported elsewhere. 39,40We think it likely that assisted switching will not actually prove impactful within the PST series: because PSTs are stoichiometric binders for a high-expression target (tubulin: ca. 10 mM in the cytosol 42 ), so tubulin-inhibitory activity requires rather high concentrations of conjugates in the cytosol.Since most uorophores are larger than the (moderate-potency) PST pharmacophore, we suspect that inevitable losses of potency upon payload attachment will be complicated by uorophore biodistribution effects such that cellular applications as photoswitchably-binding colchicinoids are blocked (ESI Note 4 †).

Assisted switching for other reagents and targets
However, we see strong scope for applying assisted switching elsewhere, e.g. with (a) inherently high-potency scaffolds, since these may better tolerate uorophore attachment; (b) reagents where incomplete Z / E switch-off has limited performance so far; (c) reagents being developed for use in deeper tissues, or in other optically dense or scattering systems including in materials, where long-wavelength response can be a great advantage.Although we will report such applications from our work in due course, 39,40 we believe that the discovery of this assisted switching effect will have signicantly broader impact than our photopharmacology-oriented investigations: and we look forward to its adoption by the chemical community.

Fig. 3 PST
Fig. 3 PST cytotoxicity can be optically controlled, with up to 20-fold enhancement from dark to lit conditions (MTT viability assay; HeLa cervical cancer cells; 46 h incubation).(a) Cell viability for selected PSTs.Data as mean ± SD.(b) EC 50 values for all compounds (full data in Fig. S9; † *PST-31 irradiated at 370 nm and PST-32 at 425 nm).(c and d) Wavelength-dependent cell viability with PST-27P.

Fig. 5
Fig. 5 (a) Conjugates MR69 and MR110 have (b) similar absorption as the spectral sum of their components, but (c and d) undergo dramatically efficient and near-complete isomerisation when irradiated in the absorption region of their fluorophore motif (here, 520-600 nm).((b): "add." is the sum of isolated spectra, "mix" is the spectrum of a mixed solution.c: time-course of absorption at 370 nm (high: mostly-E; low: mostly-Z) during orthogonal illuminations alternating between 380 nm (deep purple, E / Z) and the indicated wavelengths (coloured as appropriate); grey/ red boxes indicate the unique spectral region for assisted photoswitching with RhB.(c and d): monochromator light source with 5 nm FWHM used for switching.(d): "photo-rates" are normalised photoisomerisation conversion per unit light intensity (details at Fig. S4 †).).
CytoSwitch number 03EFIBY145 to O. T.-S.); the Swiss National Science Foundation (31003A_166608 to M. O. S.).M. G. acknowledges an LMUExcellent fellowship.We thank the LMU imaging platform CALM for microscopy access and Jörg Standfuss (PSI Villigen) for input and support for structural biology.O. T.-S.thanks Dirk Trauner for committed support and discussions during the progress of this research.