Selective targeting of non-centrosomal AURKA functions through use of a novel targeted protein degradation tool

Targeted protein degradation tools are becoming a new therapeutic modality, allowing small molecule ligands to be reformulated as heterobifunctional molecules (referred to as ‘PROTACs’, for PROteolysis Targeting Chimeras) that recruit a ubiquitin ligase to the target of interest, leading to ubiquitination of the target and its destruction via the ubiquitin-proteasome system. A number of PROTACs against targets of clinical interest have been described, but detailed descriptions of the cell biology modulated by PROTACs are missing from the literature. Here we describe the functional characterization of a PROTAC derived from AURKA inhibitor MLN8237 (alisertib). We demonstrate efficient and specific destruction of both endogenous and overexpressed AURKA by Cereblon-directed PROTACs. At the subcellular level, we find differential targeting of AURKA on the mitotic spindle compared to centrosomes. The phenotypic consequences of PROTAC treatment are therefore distinct from those mediated by alisertib, and in mitotic cells differentially regulate the centrosome- and chromatin-based microtubule spindle assembly pathways. In interphase cells we find that PROTAC-mediated clearance of non-centrosomal AURKA, and not PROTAC-mediated inhibition of its activity, efficiently modulates the cytoplasmic role played by AURKA in mitochondrial dynamics, whilst the centrosomal pool is refractory to PROTAC-mediated clearance. Our results point to differential accessibility of subcellular pools of substrate, governed by substrate conformation or localization in compartments more or less accessible to PROTAC action, a phenomenon not previously described for this new class of drugs.


Introduction
The advent of targeted protein degradation tools that exploit the endogenous protein degradation machinery to eliminate disease proteins from the cell has started a revolution in therapeutic strategy and drug design 1 . One novel way to target proteins for degradation is through PROteolysis Targeting Chimeras (PROTACs), consisting of a chimeric molecule that binds at one end to a protein target, and at the other to a ubiquitin ligase (E3), most commonly the Cereblon (CRBN) substrate recognition protein together with the CUL4A E3 ligase complex, or to the von Hippel Lindau (VHL) protein in association with the CUL2 complex 2,3 . This PROTAC-mediated ternary complex formation between functional E3 and target protein facilitates ubiquitin transfer 4 , leading to ubiquitination of the target and its proteolysis at the 26S proteasome.
This new paradigm of 'event-driven' pharmacology (in contrast to the use of traditional 'occupancy-based' drugs) holds great hope for the development of catalytic drugs able to work at lower doses and with higher specificity than the ligands from which they are derived. Moreover, the altered pharmacodynamics of substrate destruction versus inhibition raises the possibility of repurposing small molecule ligands (including those that have failed clinical trials as inhibitors of their targets) into PROTACs. However, although a number of publications document the success of novel PROTACs in eliminating their cellular targets, there has been little impact of this technology so far in the field of cell biology. PROTACs have clear potential as a method for investigating cellular protein functions on a timescale more favourable than siRNA-mediated interference and in a way that does not depend on effective inhibition of an enzymatic function. In this study, we investigate the properties of a novel PROTAC tool based on a known small molecule inhibitor of the mitotic kinase Aurora A (AURKA), MLN8237 (also known as alisertib) [5][6][7] .
AURKA is a well-studied regulator of mitosis, playing critical roles in centrosome maturation, mitotic timing, microtubule nucleation and spindle assembly 8,9 . Distinct populations of AURKA are either recruited to centrosomes by CEP192, or on spindle microtubules (MTs) via the MT-associated protein TPX2. These separate populations can be independently perturbed through disruption of either interaction [10][11][12] . AURKA activity at centrosomes contributes to mitotic entry. Activation of AURKA is thought to occur either through auto-phosphorylation in the T-loop (at T287/288), a process promoted by CEP192 oligomerization at the centrosomes, or through interaction with a number of known binding partners that act to stabilize the 'DFG-In" conformation to favour kinase activity independently of T-loop phosphorylation [13][14][15] . The best-known of these interactors is TPX2. At nuclear envelope breakdown (NEB), TPX2 is released by importin-α, under the influence of the RanGTP gradient around the mitotic chromosomes, to bind and activate AURKA. In vitro tests show that binding by TPX2 and T-loop phosphorylation independently activate AURKA approximately 100-fold 16,17 . These separable intracellular AURKA activities (defined by pT288 at the centrosomes and TPX2 binding around chromatin) contribute to distinct pathways of MT nucleation that act together to achieve mitotic spindle assembly. Critical targets of AURKA in both pathways are NEDD1 and TACC3. Recruitment and phosphorylation of NEDD1 allows recruitment of the γ-TURC nucleating complex whilst phosphorylation of TACC3 promotes assembly of a pTACC3-AURKA-clathrin complex proposed to stabilize parallel MTs in the spindle.
AURKA undergoes targeted proteolysis in every cell cycle as a substrate of the Anaphase-Promoting Complex (APC/C) ubiquitin ligase at mitotic exit 18,19 . However, AURKA is detectable in interphase cells and has been attributed a number of non-mitotic roles including ciliation control, cell cycle regulation of MYCNdependent transcription, DNA damage pathways and mitochondrial regulation [20][21][22][23] . Overall, there is a growing interest in the roles played by AURKA outside of mitosis and their contribution to its cancerpromoting activity. AURKA has long been a postulated therapeutic target due to its well-documented overexpression in cancer, although the role it plays in oncogenesis is far from clear. Recent structural and conformational studies have led to improved understanding of its mode of activation and the realization that multiple active forms may persist through interphase that depend on different binding partners. Recent work from our lab has shown that un-degraded AURKA retains activity after mitosis 24 . Therefore, a PROTAC tool able to eliminate AURKA protein could be an important cell biology tool as well as a potential therapeutic strategy. Here we test characteristics of PROTAC activity directed against AURKA and investigate the cell biology that accompanies targeted protein degradation of this critical cellular target.
For assaying live cell degradation of AURKA-Venus TO in G2 arrested cells, 1.5 x 10 4 RPE1 AURKA TO cells were seeded per well in 8-well slides (Ibidi GmbH) and treated for 16 hr with 10 µM RO3306 (Tocris Bioscience) prior to PROTAC treatment.
For assaying degradation by immunoblot of cell extracts, 2 x 10 5 AURKA-Venus KI cells were seeded in 6well plates prior to 16

Cell transfection
Cells were transfected with 1µg of plasmids using electroporation with Neon Transfection System (Thermo Fisher Scientific) using the following parameters: pulse voltage 150 V, pulse width 10 ms, and 2 pulses total on the transfection device according to the manufacturer's protocol. AURKA and AURKB plasmids used were expressed with C-terminal Venus tags in pVenus-N1 vector. D32-66, S51D, S155R and D67 versions of AURKA were generated by PCR mutagenesis, with cloning maps available on request.

Immunofluorescence
Cells were seeded at 2 x 10 4 onto glass coverslips and then fixed with cold 100% methanol (−20°C), permeabilized and blocked with 3% bovine serum albumin (BSA) and 0.1% Triton X-100 in PBS (blocking buffer) for 15 min at room temperature. Cells were washed 3 times in PBS with 0.1% Triton X-100 for 5 min each prior to 1 hour incubation with primary antibodies diluted in blocking buffer at room temperature in a humidity chamber. Slides were then washed 3 times again in PBS with 0.1% Triton X-100 for 5 min each before incubation with secondary antibodies diluted in blocking buffer for 45 min at room temperature in a humidity chamber. DNA was stained with Hoechst-33342 (1µg/mL) and coverslips were mounted with Prolong Gold antifade reagent.

Microscopy
All images were acquired on automated epifluorescence imaging platforms based on Olympus IX81 or IX83 inverted microscopes (Olympus Life Science, Southend-on-Sea, UK) with LED illumination source and motorized stage. Time-lapse was carried out using cells seeded on Ibidi 8-well slides, and imaged at 37°C in L-15 medium/ 10% FBS using a 40X NA1.3 OIL objective. Epifluorescent stacks of fixed cells after processing by IF were acquired using 60X NA 1.0 OIL objective with 200 nm step. Image acquisition was controlled by Micro-Manager 26 and images exported as tiff files.

Image analysis, quantifications and statistical analyses
Images were analysed using FIJI 27 , measuring net green intensity (Ti) of cell after background subtraction at T0 and T200 mins. Picked cells which remained in prometaphase for the duration of the 200 mins. % degradation measured as (T0I -T200I)/ T0I. Linescans were carried out using the BAR package in FIJI.
Quantified data analyses were plotted using GraphPad 6.01 (San Diego, CA, USA). Results were analyzed with ANOVA, Student's t-test or Mann Whitney U test (non-parametric) as indicated in figure legends.

AURKA and AURKB biochemical assays
AURKA and AURKB biochemical assays were performed as part of the ThermoFisher SelectScreen kinase profiling service. These assays were in the Z'-Lyte activity format and used full-length purified protein and ATP at Km (10 µM for AURKA, 75 µM for AURKB).

Results
We set out to investigate the action of AURKA-directed targeted protein degradation tools (PROTACs) against AURKA in single cell time-lapse assays using cell lines that we have previously described 29  We found that CRBN-based PROTAC compounds were able to elicit destruction of both AURKA-Venus KI and AURKA-Venus TO in time-lapse movies of mitotic arrested cells ( Figure 1A-D). Compound D, a Cereblon driven PROTAC, reduced AURKA levels in a dose dependent manner ( Figure 1B, C), with an EC50 in the 100 nM range (Figure 1B). At a dose of 1µM, Compound D caused loss of AURKA-Venus with t1/2 approximately 2 hours ( Figure 1C). The activity of the PROTAC against AURKA-Venus KI in time-lapse assays ( Figure 1D), or against endogenous AURKA in extracts from mitotic arrested HeLa cells ( Figure 1E, F), appeared to correlate with linker length, suggesting that topological constraints limit the efficacy of PROTAC action. The VHL-based PROTACs tested were inactive in all but one dose ( Figure 1D). Taken together, the most efficient PROTAC tested in these initial experiments was Compound D, which we named AURKA-PROTAC-D (PROTAC-D). We tested the correlation between linker length and efficacy of the PROTAC by creating a new compound with extra-long linker, Compound DX ( Table 1). As predicted, Compound DX reduced AURKA levels more efficiently than PROTAC-D ( Figure 1G). We tested the specificity of PROTAC action of CRBN-directed compounds in further experiments ( Figure 1H, I) demonstrating that neither MLN8237 nor the CRBN ligand (pomalidomide) on its own affected AURKA levels ( Figure 1H). In addition, the action of Compound D was blocked by competition with excess pomalidomide ( Figure 1I), supporting that recruitment of Cereblon E3 holo-complex was necessary for AURKA level reduction.
While analysing these experiments, we noticed that AURKA-Venus KI cells arrested in mitosis with STLC were more likely to exit mitosis after treatment with PROTAC than after treatment with DMSO. As we wanted to be able to separate PROTAC treatment effects caused by target degradation from any residual inhibitory effects caused by just on-target or off-target engagement, we used Compound A (Cpd A) as a negative control in this and subsequent experiments. Cpd A is a MLN8237-VHL molecule with linker length consistent with PROTAC-D that showed no PROTAC activity against AURKA ( Figure 1D). Cpd A had a small and non-significant effect in promoting mitotic slippage (consistent with a weak inhibition of AURKA activity) compared to PROTAC-D ( Figure 2A). Since AURKA is itself a substrate of mitotic exit degradation under control of the APC/C FZR1 33 , failure of the SAC, leading to activation of the APC/C, would be predicted to result in degradation of AURKA independently of PROTAC-mediated ubiquitination. Therefore, in the single cell mitotic degradation assays shown in Figure 1, we quantified only cells that remained arrested in mitosis for the duration of the assay. However, we also carried out experiments to test directly whether mitotic degradation pathways were involved in PROTAC-D-driven disappearance of AURKA-Venus, using a combination of drugs (APCin, proTAME) that inhibits the activity of the APC/C ubiquitin ligase 34 .
Degradation of AURKA-Venus in response to PROTAC-D was not prevented by inhibition of APC/C ( Figure   2B, C) and was therefore independent of mitotic exit.
The effect of PROTAC-D in promoting mitotic exit could potentially be explained by a number of studies showing a role for AURKA in the SAC [35][36][37] , but could also occur through 'mitotic slippage', should there be any non-specific targeting of Cyclin B1 by PROTAC-D in the presence of an active SAC 38 . We tested this possibility using a RPE1-cyclin B1-Venus KI line 39 . Degradation of Cyclin B1-Venus and escape of cells from SAC-induced arrest were both strongly promoted by PROTAC-D ( Figure 2D, E), and weakly by Cpd A.
However, in contrast to AURKA-Venus KI degradation ( Figure 2B), Cyclin B1-Venus KI degradation measured upon PROTAC-D treatment was sensitive to APC/C inhibition ( Figure 2F). These results allowed us to conclude that Cyclin B1-Venus degradation in the presence of PROTAC-D is the result of weakened SAC and that Cyclin B1 is not targeted directly by PROTAC-D.
In further experiments to test that degradation of AURKA in response to PROTAC-D was independent of the well-characterized APC/C-dependent pathway, we used versions of AURKA known to be resistant to APC/C-mediated degradation. AURKA possesses an atypical APC/C degron motif, the so-called A-box, in its N-terminal disordered region 40 . The A-box function appears to be negatively regulated through phosphorylation on Ser51, since phospho-mimic substitution at this site blocks mitotic degradation of AURKA 33,41,42 . We found, using single cell time-lapse degradation assays, that an A-box deleted (Δ32-66) version of AURKA-Venus stably expressed in an RPE-FRT/TO line was efficiently degraded in response to PROTAC-D ( Figure 2G). We concluded that PROTAC-mediated degradation of AURKA does not require the substrate motif essential for its canonical degradation, either for ubiquitination, or at any downstream step in substrate processing at the 26S proteasome. We additionally tested the S51D version of AURKA-Venus alongside the WT protein in time-lapse degradation assays, after transient electroporation into U2OS cells. We found not only that both WT and 'non-degradable' S51D were sensitive to PROTAC-D in mitotic cells ( Figure 2H), but that they were also sensitive in interphase cells ( Figure 2I), as further confirmation that PROTAC-mediated processing of AURKA for destruction is independent of cell cycledependent pathways. We note that the measured rate of degradation is lower in interphase cells than mitotic cells, most likely because degradation is masked by ongoing synthesis (observed as accumulation of the protein in DMSO-treated control cells). We concluded that degradation of AURKA measured in response to PROTAC-D treatment is a direct consequence of PROTAC-D-mediated targeting. Next, we asked whether target destruction mediated by PROTAC-D was specific for AURKA. Since the PROTAC target ligand MLN8237 has a degree of selectivity for AURKA over its cellular paralogue AURKB, but is not completely specific (it inhibits AURKB activity at doses of ≥ 50 nM 5 , and the reported selectivity ratio AURKA-TPX2(1-43): AURKB-INCENP(783-918) is approximately 5-fold 6 ), we might expect to find some degradation of AURKB in response to a PROTAC carrying the MLN8237 warhead. Furthermore, considering that within the mitotic cell AURKA resides in multi-protein complexes governing its localization and function, we hypothesized that the 'ectopic' recruitment of ubiquitination machinery by PROTAC-D might lead to ubiquitination and destruction of AURKA binding partners. Therefore, we examined if PROTAC-D caused reduction in cellular levels of AURKB, or of two well-known interacting partners of AURKA, TPX2 and TACC3 (Figure 3). Surprisingly, we found that treatment with PROTAC-D caused very little degradation of AURKB-Venus in an inducible U2OS cell line 43 (Figure 3A), or of endogenous AURKB in HeLa cells ( Figure 3B, C). We also found no degradation of endogenous TPX2 in mitotically-enriched HeLa cells after 3 hours of treatment with PROTAC-D ( Figure 3B, C). TPX2 and TACC3 levels were unchanged in cells treated for up to 12 hr, when endogenous AURKA was no longer detectable ( Figure 3D). Therefore PROTAC-D-mediated destruction is highly specific for AURKA. The resistance of AURKA binding partners to PROTAC-D treatment suggests that the ubiquitination step is highly specific for the AURKA moiety of mitotic complexes, or alternatively, that only unbound AURKA is targetable by PROTAC-D.
Given the unexpected resistance of AURKB to AURKA PROTAC action, we compared in vitro kinase inhibition activities for AURKA and AURKB of PROTACs -D and -DX, Cpd A, and their warhead MLN8237.
We found that both of the PROTACs had greater selectivity for AURKA over AURKB than MLN8237 in kinase inhibition assays (fold selectivity of PROTAC-D = 21.6, PROTAC-DX = 23.7, MLN8237 = 8.3) ( Figure 3E), explaining the lack of AURKB degradation seen in Figures 3A-C). The increased selectivity for AURKA suggests that the increased size and/or complexity of the PROTAC creates new steric parameters influencing target discrimination, and is consistent with published findings from others that the requirement for ternary complex formation in PROTAC action can build a further layer of specificity into drug action 44,45 .
Comparing IC50 values for inhibition of in vitro kinase activity of PROTAC-D and -DX versus MLN8237, we found that inhibition of AURKA kinase activity by the PROTAC molecules is weaker than that mediated by MLN8237 (5-10 fold). Interestingly PROTAC-DX, which has stronger PROTAC activity in comparison to PROTAC-D ( Figure 1G), does not have higher activity in this assay ( Figure 3E). This finding is in line with the idea that the efficiency of PROTAC activity is not only impacted by binding affinity to the target or E3 ligase, but also related to efficiency of ternary complex formation between E3 and target protein 46 .
Having validated PROTAC-D as an effective and specific tool for depletion of cellular AURKA, we investigated how PROTAC-mediated AURKA destruction would compare to enzymatic inhibition as a method for down-regulating AURKA functions in mitotic cells. We fixed cell populations synchronized for passage through mitosis and treated for 4 hr with parallel doses of PROTAC-D or MLN8237, or with DMSO as a negative control, and stained them by immunofluorescence (IF) for the presence of AURKA, markers of AURKA activity and tubulin, in order to assess the phenotypic consequences of drug treatment ( Figure   4). We looked first at AURKA staining and found that cells treated with PROTAC-D displayed a marked loss of the pool of AURKA associated with the spindle (seen in DMSO-treated controls). However, AURKA was preserved at the centrosomes ( Figure 4A). By contrast, treatment with MLN8237 abrogated almost all AURKA localization to centrosomes, consistent with the known role of AURKA activity in centrosome maturation that includes recruitment of AURKA to the pericentriolar material (PCM) 47 . This finding suggested that the centrosome-associated pool of AURKA seen in PROTAC-D-treated mitotic cells would be unexpectedly fully active (that is, neither degraded nor inhibited by PROTAC treatment). We tested this idea by measuring levels of pSer83-LATS2 as a well-known centrosomal marker of AURKA activity, finding that this marker was entirely resistant to PROTAC-D treatment (at doses sufficient to deplete most of the cellular pool of AURKA), whilst responding in dose-dependent fashion to MLN8237 ( Figure 4B,C).
Given the >5-fold difference in enzyme inhibition of PROTAC-D and MLN8237 ( Figure 3E) and the likelihood that the intracellular dose of PROTAC-D is limited by its size 48 Figure 4D and as previously described 5 ). We were surprised to find that PROTAC-D-treated cells showed none of these defects ( Figure 4D, E). Even at the highest dose tested (250 nM), we did not see the orientation defects characteristic of low dose inhibition of AURKA activity 5,49 . Instead, we observed that the mitotic spindles were shorter in length after PROTAC treatment. Distribution of the centrosomal pLATS2 staining shown in Figure 4B confirms that the pole-to-pole distance of correctly oriented bipolar spindles is reduced ( Figure 4F).
Our finding of 'short spindles' was reminiscent of the previously reported finding that specific perturbation of AURKA binding to TPX2 controls spindle length independently of any effect on assembly 10 which can occur under the influence of the centrosomal AURKA pool. Therefore, we hypothesized that PROTAC-D had selectively depleted the TPX2-associated pool of AURKA to eliminate the chromosome-centred MT nucleation pathway whilst leaving the centrosomal pathway untouched. We decided to test this idea using a modified cell synchronisation assay that would better allow us to compare the roles of kinase inhibition and target degradation in mitotic cells independent of their different effects on AURKA-dependent centrosome maturation. We pre-synchronised cells at metaphase by release of cells arrested for 24 hours in Thymidine into APCin/proTAME for 6 hours. We then treated metaphase-arrested cells with different doses of MLN8237, PROTAC-D and Cpd A for 3 hours before fixing them for IF analysis. We reasoned that use of these drugs would reveal phenotypes resulting from degradation of AURKA, distinguishing them from those arising purely out of kinase inhibition: Degradation-dependent effects would be sensitive to PROTAC-D whilst insensitive to Cpd A treatment (with sensitivity to MLN8237 depending on kinase-dependence of the phenotype studied); kinase inhibition phenotypes would show greatest sensitivity to MLN8237, whilst PROTAC-D and Cpd A would show weak or no effect, but importantly, they would produce the same effect.
Following fixation, we stained cells for AURKA and its interactors CEP192 and TPX2 ( Figure 5). Similar levels of CEP192 at centrosomes after the different treatments confirmed that centrosome maturation had occurred in a large fraction of the cellular pool of metaphase cells ( Figure 5A). Quantification of AURKA levels in these cells showed the total cellular pool of AURKA reduced more than three-fold after PROTAC-D treatment ( Figure 5B). Measured AURKA levels were also somewhat lower (by about 30%) after treatment with Cpd A or with low (25 nM) or high (250 nM) doses of MLN8237. Since we have found that these treatments do not affect endogenous AURKA levels, nor AURKA-Venus levels in intact cells, we assumed that the reduced AURKA levels seen in IF reflected loss of AURKA in the fixation step, that could be a consequence of reduced interaction with the mitotic spindle. Indeed, MLN8237 and TPX2 may compete with each other for AURKA binding 50 (see Discussion). We measured pole-pole distances in this experiment and found them reduced by PROTAC-D treatment. AURKA inhibition with 25 nM MLN8237 also gave rise to short spindles, whereas Cpd A had no effect on spindle length ( Figure 5C). We concluded that Cpd A and PROTAC-D both bind too weakly to AURKA to significantly inhibit its activity, and that the short spindle phenotype seen after PROTAC-D treatment depends on destruction of AURKA by PROTAC-D.
Consistent with this conclusion, we observed that PROTAC-D alone of the drug treatments removed both cytoplasmic and spindle pools of AURKA ( Figure 5D). Kinase inhibition mediated by 25 nM MLN8237 or 250 nM Cpd A caused some loss of signal from the spindle, but also increased cytoplasmic levels of AURKA ( Figure 5D, E). Comparison of AURKA pixel intensities in fixed areas on the centrosome or neighbouring spindle confirmed that depletion of the spindle signal was greater than at the centrosome ( Figure 5F). We concluded from our data that PROTAC-D preferentially targets the pool of AURKA that associates with TPX2 to govern mitotic spindle length. Moreover, because kinase inhibition assays indicate that Cpd A and PROTAC-D bind AURKA with equivalent affinity (Figure 3E), we concluded that the short spindle phenotype seen after PROTAC-D -but not Cpd A -treatment ( Figure 5C) depends on destruction of AURKA protein.
We investigated further why PROTAC-D treatment led to selective depletion of the spindle-associated pool of AURKA. This could result from conformation-dependent targeting by the PROTAC, with the preferred target being either the TPX2-bound pool, or a free pool of AURKA (provided this turns over faster with the TPX2-bound pool than the centrosomal pool). Alternatively, the AURKA pool at the centrosomes might be 'protected' from PROTAC action (for example if either the PROTAC or CRBN E3 ligase were excluded from the PCM matrix). We investigated this question further by measuring the efficacy of PROTAC-D treatment in living cells under conditions where AURKA-Venus localization is perturbed.
We first compared the responses to PROTAC-D treatment of different versions of AURKA-Venus: AURKA-S155R, a version of AURKA showing strongly reduced interaction with TPX2 51 , and two N-terminally truncated versions of AURKA (D67, D127) that show increased localization to the nucleus in RO3306- Figure S2). We found that S155R showed a similar pattern of targeting by PROTAC-D to the wild-type version, suggesting that interaction with TPX2 would not influence targeting. Moreover sensitivity to PROTAC-D of both WT and S155R versions of AURKA-Venus was highly variable in interphase cells, suggesting that factors other than TPX2 regulate the sensitivity of AURKA-Venus to PROTAC-D. By contrast, we observed that degradation of D67 was enhanced and less variable.

arrested interphase cells (Supplementary
D67 was also more strongly localized to the nucleus. When we measured the nuclear versus cytoplasmic degradation of wild-type protein, we found that nuclear AURKA-Venus was more efficiently depleted than the cytoplasmic pool upon PROTAC-D treatment (Supplementary Figure S3). The AURKAD127 version was also strongly nuclear (and absent from centrosomes), but appeared to be a hyper-stable version of the protein, showing a tendency to accumulate in cells over the time-course of the experiment in absence of PROTAC treatment, and low responsiveness to PROTAC (Supplementary Figure S3). From these data we concluded that nuclear localization appears to favour targeting of AURKA by PROTAC-D, but that the unstructured N-terminal domain, or the lysine residues within it, are required for turnover of AURKA (both for proteostasis in unperturbed cells as well as response to PROTAC-D).
Given this experimental limitation in using truncated versions of AURKA-Venus, we turned instead to an experiment designed to investigate how targeting of the wild-type protein is affected by its subcellular localization. Since we had found endogenous AURKA to be more strongly depleted at the spindle than on centrosomes, we treated RPE1-AURKA-Venus KI cells with siRNA against CEP192 (CEP192i) to displace AURKA from the centrosomes 11 . This relocalization of AURKA is readily observed in cells arrested in G2 phase when AURKA-Venus KI expression is high ( Figure 6A-C). We tested AURKA-Venus degradation in response to PROTACs under these conditions, and measured slightly increased degradation of AURKA-Venus by immunoblot ( Figure 6D, E). We then tested the effect of CEP192i on mitotic cells, when a much larger pool of AURKA is normally recruited to centrosomes. In STLC-arrested cells we found that AURKA-Venus KI delocalized from centrosomes after CEP192i ( Figure 6F, G) was more readily degraded in response to PROTAC-D ( Figure 6H). We concluded that PROTAC-D is unable to bring about efficient degradation of centrosome-localized AURKA-Venus.
Given the differential targeting of AURKA pools by PROTACs -D and -DX, we tested whether PROTAC-DX could act to down-regulate the cytoplasmic pool of AURKA that regulates the mitochondrial network during interphase 20,52 . Our recent work has found that excess AURKA present in FZR1 KO cells fragments the mitochondrial network 24 . We treated FZR1 KO cells with PROTAC-DX and Cpd A and found that PROTAC-DX, but not Cpd A, rescues mitochondrial morphology (Figure 7). Therefore PROTAC-DX is able to prevent interphase activity of AURKA in a manner that depends on destruction of the protein, since Cpd A, which has similar activity as an inhibitor of AURKA but is unable to bring about its degradation, is unable to suppress AURKA activity at the same dose. We conclude that PROTAC-mediated clearance is more efficient than kinase inhibition in downregulating cytoplasmic AURKA activity.

Discussion
We have described a small molecule that acts as a specific degrader of AURKA to clear endogenous, exogenous GFP-tagged, or overexpressed protein from the cell. Amongst the molecules we tested, successful degraders were CRBN-specific. Although we have not formally excluded the possibilities that VHL is insufficiently active in U2OS cells to generate degradation-competent ubiquitin conjugates in response to CpdA, or that the four linker constructs we tested all occluded ternary complex formation between AURKA and VHL, our observation that the same four linker constructs were all able to support PROTAC activity directed to CRBN are in line with the published finding that protein-protein interaction surfaces of CRBN are more favourable to stable ternary complex formation than the equivalent surfaces of VHL 46 .
We found that the activity of CRBN-directed molecules correlated with linker length but was independent of the affinity of the compound for its AURKA target. Therefore, it is likely that our longer linkers promote the assembly of productive ternary complexes by bringing together AURKA and CRBN in an orientation that allows the E3 complex to ubiquitinate AURKA at appropriate lysine residues. The physical properties of the linker are critical parameters in PROTAC activity, and further optimisation of PROTAC-DX could include different linker patterns to alter linker flexibility, as well as lengths.
Compounds showing PROTAC activity against AURKA were several-fold less potent than their MLN8237 warhead in inhibiting AURKA activity in vitro, consistent with reduced affinity for their target. Indeed, we found no evidence for direct inhibition of cellular AURKA functions by PROTAC-D (that is, all functions we examined were insensitive to the control compound, Cpd A, which shows equivalent activity against AURKA in kinase inhibition assays but has no PROTAC activity). Therefore we concluded that binding of PROTAC-D is weak enough, and/or the molecule present at sufficiently low intracellular levels, to achieve targeted degradation of AURKA in absence of any significant inhibition of AURKA kinase activity, and without exhibiting the hook effect characteristic of heterobifunctional ligands 53 .
We observed that clearance of AURKA from the cell is less efficient than that mediated by its cognate E3, APC/C-FZR1 (t1/2 ~ 100 min vs t1/2 ~ 45 min 18 ). We speculate that even with further optimization, it seems unlikely that any PROTAC would eliminate AURKA faster than its cognate pathway, since the rate-limiting step for degradation of many ubiquitinated substrates is not recruitment to the 26S proteasome, but determinants of processing that are partly substrate-specific (such as unfolding of substrate at the proteasome) and partly determined by the configuration of ubiquitin chains 54 . Indeed, a recent study found that the presence of unstructured regions determines the PROTAC mediated degradation of VHL-directed substrates 55 .
Interestingly, we tested versions of AURKA known to be resistant to APC/C-mediated degradation through mutation or removal of the essential N-terminal degron, and found them to be degraded in the presence of PROTAC-D as efficiently as wild-type AURKA. Therefore the position and topology of ubiquitin chains assembled on AURKA by CRBN and APC/C-FZR1 are likely to be very different.
We also observed that different cellular pools of AURKA substrate were differentially targeted by PROTAC treatment, since in mitotic cells the spindle-associated fraction of AURKA was eliminated whilst the centrosome fraction was preserved. Since the centrosomal pool of AURKA retained its activity, spindle assembly was buffered against the loss of the chromatin-associated TPX2-activated AURKA pool and the observed phenotype of PROTAC-D treatment in mitotic cells is therefore shortened spindles, consistent with a previous study of cells engineered to express a non-AURKA-binding version of TPX2 10 . Similarly, in interphase cells we observed that PROTAC-D treatment efficiently cleared the non-centrosomal pool of AURKA, but that centrosomal AURKA was preserved. Delocalization of the centrosomal pool through siRNA-mediated depletion of CEP192 promoted clearance of the total cellular pool of AURKA-Venus by PROTAC-D. Since centrosomal AURKA is efficiently inhibited by MLN8237, we would expect it to be accessible to bind MLN8237-derived PROTAC molecules. One explanation for its inaccessibility to PROTAC-D action could be that PROTAC-D at the centrosome fails to recruit CRBN or another component of the E3 complex required for ubiquitination of its target. Alternatively, there may be deubiquitinase enzymes active at the centrosomes that act to stabilize ubiquitinated AURKA.
Our results point to differential accessibility of subcellular pools of substrate, governed by substrate conformation or localization in compartments more or less accessible to PROTAC action, a phenomenon that has not previously been described for PROTAC agents acting via CRBN. Our finding of localized response to PROTAC-D is in contrast with treatment by the AURKA inhibitor alisertib, which promotes a clear dose-dependent depletion of pLATS2, a marker of AURKA activity at the centrosomes.
Given the complex conformational and spatial regulation of AURKA 15,17,56 we tested for conformationspecific targeting of the kinase using different versions of AURKA-Venus. The conformational dynamics of AURKA are strongly constrained through interaction with TPX2 50,57,58 , which favours the so-called 'DFG-In' active confirmation and it has been suggested that different functional pools possess distinct conformational properties that will modulate interactions with inhibitors 50 . In this study, alisertib/MLN8237 was shown to be a 'Type 1' inhibitor that promotes the inactive DFG-Out state, and TPX2 to oppose DFG-Out inducers, exhibiting negative cooperativity on binding with MLN8237 50 . In the DFG-Out state, the active site is more open (i.e. more accessible to MLN8237 or PROTAC-D). We expected therefore that a version of AURKA impaired in TPX2 binding (S155R 51 ) should be more strongly degraded in response to PROTAC-D in presence of TPX2 (consistent with a model whereby PROTAC-D can more easily access AURKA in the DFG-Out conformation). However, we did not find a significant difference in response to PROTAC-D between S155R and wild-type AURKA. Instead, our data were consistent with the idea that the nuclear pool of AURKA is more efficiently degraded than the cytoplasmic pool. Further experiments are required to establish whether these effects are related to the conformation of the target, or to the presence or activity of components of the UPS machinery engaged by PROTAC-D.
AURKA is of strong interest as a therapeutic target for various cancers, but despite extensive testing in clinical trials, alisertib has yet to reach the clinic. Our study is the first to describe a drug that shows specificity for different subcellular pools of AURKA, raising the possibility of developing PROTACs to finetune the activity of AURKA (and other targets that have shown disappointing clinical results) to produce cellular phenotypes that are potentially more desirable in pharmacological or therapeutic contexts. For example, alisertib-derived PROTACs could be used to target cytoplasmic functions of AURKA without inducing mitotic errors that are consequence of inhibiting AURKA function at the centrosome.