Molecular mechanism of a parasite kinesin motor and implications for its inhibition

Plasmodium parasites cause malaria and are responsible annually for hundreds of thousands of deaths. They have a complex life cycle in which distinct stages are transmitted between, and reproduce in, human and mosquito hosts. In the light of emerging resistance to current therapies, components of the parasite replicative machinery are potentially important targets for anti-parasite drugs. Members of the superfamily of kinesin motors play important roles in the microtubule-based replicative spindle machinery, and kinesin-5 motors are established anti-mitotic targets in other disease contexts. We therefore studied kinesin-5 from Plasmodium falciparum (PfK5) and characterised the biochemical properties and structure of the PfK5 motor domain. We found that the PfK5 motor domain is an ATPase with microtubule plus-end directed motility. We used cryo-EM to determine the motor’s microtubule-bound structure in no nucleotide and AMPPNP-bound states. Despite significant sequence divergence in this motor, these structures reveal that this parasite motor exhibits classical kinesin mechanochemistry. This includes ATP-induced neck-linker docking to the motor domain, which is consistent with the motor’s plus-ended directed motility. Crucially, we also observed that a large insertion in loop5 of the PfK5 motor domain creates a dramatically different chemical environment in the well characterised human kinesin-5 drug-binding site. Our data thereby reveal the possibility for selective inhibition of PfK5 and can be used to inform future exploration of Plasmodium kinesins as anti-parasite targets.


INTRODUCTION
Malaria is a massive disease burden world-wide, with an estimated 219 million cases in 2017, a year which also saw the first increase in cases for nearly two decades 1 . With resistance to current frontline therapeutics rapidly rising [2][3][4] , new drug targets are urgently needed. Malaria is caused by Plasmodium parasites, which are unicellular eukaryotes belonging to the Apicomplexa phylum. Malaria parasites have a complex life-cycle involving distinct stages that are transmitted between, and reproduce in, human and mosquito hosts 5 . The cytoskeleton plays an important role throughout the parasite life cycle, and the microtubule (MT) based spindle machinery is involved in the many rounds of mitotic and meiotic replication required for parasite proliferation. Anti-mitotics are well-established as drugs in a variety of settings, notably human cancer 6 -thus, components of the malaria replicative machinery are attractive anti-parasite targets. However, given the obligate intracellular nature of malaria parasites, any therapeutic target must be sufficiently divergent to be selectively disrupted compared to host homologues.
Members of the kinesin superfamily are such potential targets. Kinesins are motor proteins that bind to MTs and convert the energy of ATP binding and hydrolysis into MT-based mechanical work. Different kinesin families have specialised functions, such as translocation of cargo along MTs, regulation of MT polymer dynamics, and organisation of higher-order MT structures like mitotic and meiotic spindles 7,8 . MTs are built from heterodimers of the highly conserved α-and β-tubulin and, whereas there is approximately 95% sequence conservation of tubulins between Plasmodium sp. and Homo sapiens, sequence conservation within kinesin families is much lower, typically 40-50 %. This raises two important questions: do distantly related members of kinesin families diverge in their molecular properties, and could such sequence divergence allow selective inhibitors to be developed?
The kinesin-5 family are involved in cell division in many organisms, have long been investigated as an anti-mitotic therapeutic target for human cancer 9 , and have also been considered as a target for anti-fungals 10 . Kinesin-5 family members are found in most eukaryotes, including Plasmodium sp., and the family is predicted to have been established in the last eukaryotic common ancestor 11 . Kinesin-5s from several species form a tetrameric bipolar structure, with two opposing pairs of motor domains that can organise MT arrays such as those found in spindles [12][13][14] . Several classes of Homo sapiens kinesin-5 (HsK5) inhibitors have been characterised that block motor ATPase activity and bind to allosteric sites in the motor domain. The best studied of these allosteric binding pockets is defined by kinesin-5-specific sequences in a key structural region of the motor domain, loop5 15 . Loop5 is critical for the correct operation of human kinesin-5, since its deletion or mutation disrupts the motor's mechanochemical cycle [16][17][18] . Furthermore, Drosophila kinesin-5 is resistant to the HsK5 inhibitor STLC, but can be sensitised by replacement of loop5 with the cognate HsK5 sequence 19 . The loop5-defined allosteric site thus has proven promise in mediating selective inhibition of kinesin-5 family members from different species.
To investigate the idea that kinesin-5 from Plasmodium falciparum (PfK5) -the deadliest form of human malaria -could be a selective anti-malarial therapeutic target, we characterised the biochemical properties and MT-bound structure of the PfK5 motor domain. We show that the PfK5 motor domain is an ATPase with MT plus-end directed motility, as demonstrated in MT gliding experiments. MT-bound structures of the PfK5 motor domain, determined using cryo-electron microscopy (cryo-EM), reveal classical kinesin mechanochemistry despite the significant sequence divergence of this motor. This includes ATP-induced neck-linker docking to the motor domain, which is consistent with PfK5 motor domain plus-ended directed motility. Finally, we show that a large insertion in loop5 of the PfK5 motor domain creates a dramatically different chemical environment in the well characterised loop5 drug-binding site, revealing the possibility for selective inhibition of PfK5.

PfK5ΔL6-MD is a slow ATPase
To characterise PfK5 mechanochemistry, we first wanted to measure its MT-stimulated ATPase activity. The PfK5 motor domain (PfK5MD, amino acids 1-493) contains a 105 amino acid asparagine and lysine rich insertion in loop6 that is characteristic of malaria proteins 20 (Fig 1A, left), but which is poorly conserved (16-30% sequence identity) among Plasmodium kinesin-5s. To enable our study, we engineered loop6 out of our construct, an approach previously taken by another group 21 . We refer to this construct as PfK5ΔL6-MD, and it was purified to 99 % purity (Fig 1A, right). In the absence of MTs, PfK5ΔL6-MD exhibited a low ATP hydrolysis rate (Fig 1B), but addition of MTs stimulated PfK5ΔL6-MD ATPase activity ( Fig 1C). Changes in pH or ionic strength showed no or minimal impact respectively on PfK5ΔL6-MD ATPase rate (S1A, B Fig). From these data, the motor Kcat and Km of MTs (KMT) for PfK5ΔL6-MD was calculated. The Km of ATP (KATP) was also determined ( Fig 1D). PfK5ΔL6-MD has a MT-stimulated ATP hydrolysis rate of 0.13 ATP s -1 , which is slow compared to kinesin-5 from S. pombe 22 , S. cerevisiae 23 , and H. sapiens 24 (with rates of 1.2, 0.5, and 2.9 ATP s -1 respectively). However, PfK5ΔL6-MD has similar KMT (5.4 M) and KATP (9.5 M) values compared to previously characterised kinesin-5s 23,24 . Thus, despite substantial sequence divergence, and although it has the lowest ATPase rate observed to date for the family, PfK5ΔL6-MD exhibits overall similar ATPase properties compared to other kinesin-5s.

PfK5ΔL6-MD generates slow MT gliding
To determine the motile properties of PfK5ΔL6-MD, we used a MT gliding assay. PfK5ΔL6-MD was expressed with a C-terminal SNAP-tag (PfK5ΔL6-MD-SNAP), purified to 92 % purity (Fig 2A), covalently labelled with biotin, and attached to a neutravidin coated surface. The velocity of fluorescently-labelled MTs driven by PfK5ΔL6-MD-SNAP activity was measured. PfK5ΔL6-MD-driven MTs moved at an average velocity of 5.4 nm/s (95 % confidence interval = 5-5.9) (Fig 2A, S1C Fig). This is slow compared to 23-92 nm/s reported for HsK5-MD 22,25,26 . However, this slow MT gliding corresponds with the slow rate of ATP hydrolysis observed in the ATPase assay. Inclusion of polarity-marked MTs in the assay further showed that PfK5ΔL6-MD is a plus-end directed motor (Fig 2B).  (Table 1). (E) Frequency distribution of PfK5ΔL6-MD-SNAP MT dwell times, in different nucleotide states. Number of experimental replicates = 3, however frequency distributions are calculated from pooled experimental data (number of technical replicates is shown in Table 1). Number of biological replicates = 2. The fit for one-phase exponential decay models is shown, with corresponding decay constant (koff), and the mean dwell time (τ) of PfK5ΔL6-MD-SNAP binding events, which is equal to the half-life of the model. (F) Mean MT association (kMTlanding) as a function of MT dissociation (koff), plotted with 95 % confidence intervals.

PfK5ΔL6-MD single molecule interactions with MTs
To better understand the slow ATPase and MT gliding activity we observed for PfK5ΔL6-MD, we analysed the interactions of single molecules of fluorescently labelled PfK5ΔL6-MD-SNAP with MTs. We made single molecule measurements in different nucleotide conditions, to investigate how MT affinity changes with nucleotide state (Fig 2C, S1D Fig, and Table 1), using the non-hydrolysable ATP analogue AMPPNP to mimic the ATP bound state. From these data, we calculated the MT association rate, or kMTlanding (Fig 2D), and MT dissociation rate, or k off of PfK5ΔL6-MD ( Fig 2E). This demonstrated that in saturating ADP conditions, PfK5ΔL6-MD-SNAP had a comparatively low kMTlanding and a high koff, indicating that PfK5ΔL6-MD-SNAP has low MT affinity when bound to ADP (Fig 2F). In the absence of nucleotide or in saturating AMPPNP conditions, PfK5ΔL6-MD-SNAP had a comparatively high kMTlanding and a low koff, showing that when no nucleotide is present, or in its ATP-bound state, PfK5ΔL6-MD-SNAP has high MT affinity. In saturating ATP conditions, the average MT dwell time of PfK5ΔL6-MD-SNAP is 0.12 s ( Fig  2E) -similar to the ATPase rate of PfK5ΔL6-MD (0.13 ATP s -1 ). This indicates that PfK5ΔL6-MD ATPase activity is rate limited by one of the MT-bound stages of the ATPase cycle. Given that a similar dwell time to the ATPase rate was also observed in ADP saturating conditions (0.14 s), it is therefore possible that PfK5ΔL6-MD is rate limited by MT release when bound to ADP. Taken together, single molecule measurements of PfK5ΔL6-MD-SNAP MT binding support observations of slow ATPase and MT gliding activity.

MT-bound PfK5ΔL6-MD structure determination using cryo-EM
To gain molecular insight into the behaviour of PfK5ΔL6-MD, its interaction with MTs and its sensitivity to nucleotide binding, we visualised MT-bound PfK5ΔL6-MD in different nucleotide states using cryo-EM (     (Table 2). These provide a detailed picture of how PfK5ΔL6-MD interacts with both α-and β-tubulin. They also show that this divergent parasite motor has a canonical kinesin fold: it is built from a central β-sheet, sandwiched between three α-helices on each side, and flanked by a small βsheet (β-lobe1), and a β-hairpin (β-lobe2) (Fig 3F,G). Using these models, we analysed conformational differences between the no nucleotide and AMPPNP states.

AMPPNP binding causes PfK5ΔL6-MD nucleotide binding site closure
The PfK5ΔL6-MD nucleotide binding site (NBS) is located away from the MT surface, and despite the overall low sequence conservation of PfK5ΔL6-MD compared to HsK5 (S4 Fig), is composed of three loops containing conserved sequence motifs. These are the P-loopwhich interacts with the α-and β-phosphates of bound nucleotide -loop9 and loop11, which contain the switch-I and switch-II motifs respectively 33 .
In the absence of bound nucleotide, density for all three NBS loops is visible in PfK5ΔL6-MD, although density for the P-loop in the no nucleotide state is poorly defined (Fig 4A). In addition, density corresponding to the C-terminal end of loop11, which is approximately 19 Å from the NBS, and contains a two residue Plasmodium-conserved insertion, is not visible.
In the no nucleotide state, density corresponding to loop9 and the P-loop are well separated, while density is observed connecting loop9 and 11 ( Fig 4A). These configurations create an 'open' nucleotide binding site primed for ATP binding formed by a cavity between these three loops. In contrast, in the AMPPNP state, there is clear density corresponding to the bound nucleotide in the PfK5ΔL6-MD NBS ( Fig 4A, S4 Fig). In addition, the nucleotide is surrounded by loop9 and 11, which have closed around the bound nucleotide to form an NBS that supports ATP hydrolysis ( Fig 4A) 34 . Superimposing the PfK5ΔL6-MD no nucleotide and AMPPNP state models by alignment on αβ-tubulin allows visualisation of the structural response of the PfK5ΔL6-MD NBS to AMPPNP binding ( Fig 4B). Even while the resolution is not sufficient to determine the exact conformation of each of these loops, it is very clear that in the presence of AMPPNP, all three NBS loops move, with loop9 and the P-loop coming closer together, thereby burying the nucleotide. In summary, AMPPNP binding to PfK5ΔL6-MD causes a conformational rearrangement that forms a closed, catalytically competent NBS.

AMPPNP binding causes PfK5ΔL6-MD subdomain rearrangement
What are the consequences of these NBS rearrangements on the structure of PfK5ΔL6-MD?
The structure of PfK5ΔL6-MD can be subdivided into three distinct subdomains 35 , which are predicted to move with respect to each other during the motor's MT-based ATPase cycle ( Fig 4C). The tubulin-binding subdomain (Fig 4C, purple hues) consists of MT binding elements in which helixα4 binds a shallow cavity at the intra-tubulin dimer interface. In addition, helixα5 and β-lobe2 contact β-tubulin. The remaining two subdomains, the P-loop subdomain (Fig 4C, blue hues) and switch-I/II subdomain (Fig 4C, green hues), contain approximately half of the central β-sheet each, along with adjacent secondary structure elements. The NBS is located at the junction of these subdomains ( Fig 4C).
We measured the relative rotation of each helix (S3 Table) in the transition from no nucleotide to AMPPNP states. This reveals the rearrangement of the P-loop and switch-I/II subdomains around a static MT binding domain (Fig 4D). The P-loop subdomain pivots such that helixα0 moves towards the MT surface, while helixα6 and the majority of the subdomain moves away from the MT surface. The switch-I/II subdomain rotates such that its constituent secondary structure elements move towards β-tubulin.

AMPPNP binding causes PfK5ΔL6-MD neck linker docking
What are the consequences for these subdomain rearrangements for the functional output of PfK5ΔL6-MD? Approximately 18 Å away from the MT surface, and 27 Å away from the NBS, is the neck linker. The neck linker is a C-terminal peptide extending from helixα6, which links the motor domain to the kinesin stalk, and which is relatively conserved in the kinesin superfamily (S3C Fig). In the PfK5ΔL6-MD no nucleotide reconstruction, no density is observed extending from helixα6, indicating that the neck linker is disordered in this nucleotide state (Fig 5A). In the AMPPNP state however, there is clear density for the neck linker at the C-terminus of helixα6 that extends along the motor domain in the direction of the MT plus-end ( Fig 5B). In addition, density corresponding to the PfK5ΔL6-MD N-terminus is also visualised in the AMPPNP state, consistent with formation of backbone interactions between this region and the neck linker, to form short β-strands known as the cover neck bundle (CNB) 36  Neck linker docking is enabled by the above-described rotation of the P-loop subdomain, which moves the N-terminus and the central β-sheet away from helixα5/loop13 in the static MT binding subdomain (Fig 4D, Fig 5)

MT binding interface
PfK5ΔL6-MD binds to one αβ-tubulin dimer, with helixα4 centred at the intra-dimer interface (Fig 6A). At the PfK5ΔL6-MD binding site, the sequence of the S. scrofa αβ-tubulin used in our reconstruction is identical to that of P. falciparum αβ-tubulin (S7 Fig), facilitating a more detailed investigation of this interface. To analyse which PfK5ΔL6-MD secondary structure elements interact with αβ-tubulin in the no nucleotide state, we coloured an αβtubulin surface representation according to different PfK5ΔL6-MD secondary structure elements; we also coloured PfK5ΔL6-MD according to proximity to α or β-tubulin ( Fig 6B). This analysis shows that helixα4 interacts with both α-and β-tubulin, while βlobe1, loop11 and helixα6 interact with helices 4, 5 and 12 of α-tubulin, and loop7, βlobe2 and helixα5 interact with similar secondary structure elements in β-tubulin. Much of this PfK5ΔL6-MD-MT interface is similar between the no nucleotide and AMPPNP states, (Fig 6C, S4 Table). However, subdomain re-arrangement and NBS closure in the AMPPNP state result in some changes. The largest of these occurs at the α-tubulin interface, where rotation of the P-loop subdomain decreases the interaction of helixα6 with α-tubulin, and positions βlobe1 closer to α-tubulin, increasing its interface area. The interface area of loop11 also increases in the AMPPNP state. Interestingly, βlobe1/loop2 forms an interface area with α-tubulin of 181 and 145 Å 2 in the no nucleotide and AMPPNP states respectively. Taken together, this shows that the PfK5ΔL6-MD-MT interface is similar to that observed for other kinesin-5s, although the extent to which βlobe1/loop2 interacts with α-tubulin differs between different family members 10,22,39 .

PfK5ΔL6-MD loop5 forms a unique putative drug binding site
Loop5 plays an important role in the mechanochemistry of HsK5 16,40 , and forms the drug binding pocket of HsK5-specific inhibitors 19 . It is a solvent exposed loop that creates a break in helixα2, and protrudes from the surface of the motor domain away from the MT. The loop5 sequence is well conserved between Plasmodium species (76-91 % sequence identity), and is longer compared to HsK5 (Fig 7A). In the no nucleotide state of PfK5ΔL6-MD, some very poorly defined density corresponding to loop5 can be seen at a low threshold (Fig. 7B), suggesting that this region is largely disordered in this state. In the AMPPNP state, however, clear density corresponding to loop5 can be seen, which forms two distinct regions. This density extends at an angle from helixα2, forming elongated density projecting from the motor domain between helixα1 and helixα3 ( Fig 7C). This region is at lower resolution than other parts of the reconstruction, possibly owing to intrinsic flexibility, but also possibly because of the residual resolution gradient towards the outer surface of PfK5ΔL6-MD (S2B Fig). There is no secondary structure-like density in this region of the motor, consistent with sequence-based predictions (S5 Fig) and, therefore, a model for PfK5ΔL6-MD loop5 was not be calculated.
Strikingly, however, the density corresponding to loop5 in the AMPPNP state does not protrude away from the surface of the motor but appears to cover the site between helicesα2 and 3, equivalent to the well described inhibitor binding site in HsK5. Docking of a crystal structure of HsK5 bound to the well-characterised inhibitor STLC in the PfK5ΔL6-MD density reveals the poor match between HsK5 loop5 and the PfK5ΔL6-MD loop5 density ( Fig  7D). This also suggests that, although residues outside of loop5 involved in interactions with STLC are largely conserved between HsK5 and PfK5 (Fig 7E), loop5 of PfK5ΔL6-MD might radically alter the spatial and chemical environment of this putative drug binding site. To test this idea, we measured whether the ATPase activity of PfK5ΔL6-MD was susceptible to inhibition by STLC 42 . Consistent with our structural prediction, while STLC inhibits HsK5 ATPase activity, it does not inhibit PfK5ΔL6-MD (Fig 7F). Thus, despite the conserved aspects of PfK5ΔL6-MD mechanochemistry uncovered by our data, evolutionary divergence between PfK5ΔL6-MD and HsK5 mediates differential inhibition of these kinesin-5 motors.

DISCUSSION
We have determined the biochemical properties and MT-bound cryo-EM structures of a spindle-associated kinesin-5 motor from the malaria parasite. Despite considerable divergence from the human host kinesin-5 sequence, our P. falciparum kinesin-5 PfK5ΔL6-MD construct shares with HsK5 a comparatively slow MT-stimulated ATPase, plus-end directed MT gliding activity and nucleotide-dependent conformational changes that support plus-end directed motility. Significantly, however, our structures revealed a radically different configuration of the well characterised loop5-defined drug binding pocket. Further, we also showed that PfK5ΔL6-MD exhibits no sensitivity to the classical HsK5 loop5 binding drug STLC.
The steady state ATPase activity of PfK5ΔL6-MD is ~340 times slower than H. sapiens kinesin-1 38 , 3-25 times slower than other members of the kinesin-5 family 23,24,39 , and its MT gliding activity is similarly and proportionally slow. Our use of mammalian brain tubulin rather than native P. falciparum tubulin might, in principal, contribute to this -however, αβtubulin is well conserved between S. scrofa and P. falciparum, and the two species have identical residues at the kinesin binding site (S7 Fig), suggesting that tubulin source is unlikely to influence PfK5ΔL6-MD activity. In further support of this, experiments comparing ATPase rates of a yeast kinesin-5 motor domain interacting with mammalian and yeast tubulin showed no difference 39 . A previous study of Plasmodium falciparum and vivax kinesin-5s also observed slow ATPase rates for these motors 21 . Our single molecule MT binding data suggest that PfK5ΔL6-MD ATPase activity is rate limited during the MT-bound portion of its ATPase cycle. These findings are reminiscent of the properties of other kinesin-5s and indeed, this may be critical for their function -substitution of the slow motor activity of vertebrate kinesin-5 with faster kinesin-1 was functionally disruptive in the complex context of the spindle 43 . This suggests that PfK5 -like other kinesin-5s -operates in motor ensembles, where slow-moving teams of PfK5 collaborate to drive MT organisation 13 .
The malaria kinesin-5 protein we studied was engineered to remove a low-complexity region in loop6, a strategy that had previously been adopted both in characterising malaria kinesin-5 21 and other malaria proteins 44 . The insertion point of loop6 lies approximately 40 Å from the NBS and, although we cannot exclude that removal of this region influences PfK5ΔL6-MD's behaviour, our structures clearly demonstrate that the engineered protein adopts a canonical kinesin fold and undergoes a structural response to AMPPNP binding. Loop6 residues are therefore not required for protein folding and fundamental kinesin mechanochemistry. Low-complexity regions like PfK5 loop6 are very common in malaria proteins, and are often found inserted in otherwise well-conserved three-dimensional folds 20 . While the role of such low-complexity regions in immune evasion is logical for extracellular parasite proteins 45 , it remains unclear if and how such regions modulate intracellular protein function.
Improvements of our MiRP image analysis procedures allowed us to efficiently handle the incomplete binding of PfK5ΔL6-MD along the MTs in our cryo-EM data (S2A Fig) and to clearly visualise MT-bound PfK5ΔL6-MD at 5-6 Å resolution. Our structures showed that PfK5ΔL6-MD exhibits an open-to-closed conformational change in the NBS, kinesin motor subdomain rearrangements and neck linker docking on ATP analogue binding, typical of a classical plus-end kinesin 38,46,47 . The lowest resolution region of our PfK5ΔL6-MD is its MTdistal surface, which encompasses the potential drug-binding loop5 region. Because we used GMPCPP-stabilised MTs, we do not think that MT lattice discontinuities -that can occur on paclitaxel-stabilised MTs 32,48 -cause this resolution loss. Rather, loop5 of PfK5ΔL6-MD, which is 21 residues longer than in HsK5 and composed mainly of hydrophilic residues, appears to be intrinsically flexible and thus its conformation is more challenging to capture structurally. Density for loop5 is only well defined in the AMPPNP state and not the no nucleotide state, suggesting it is conformationally sensitive to bound nucleotide, as also observed with HsK5 loop5 49 . Strikingly, density attributable to loop5 impinges on the pocket corresponding to the well-characterised drug binding site in HsK5 50 and provides a possible explanation for the lack of sensitivity of PfK5ΔL6-MD to inhibition by the small molecule STLC. Given the strong sequence conservation in loop5 between different Plasmodium species, this encouraging finding raises the possibility of selective inhibition of parasite motors. Indeed, a small molecule screen identified a compound able to inhibit Plasmodium kinesin-5 ATPase activity, but not that of HsK5 21 .
P. berghei kinesin-5 localises to mitotic and meiotic spindles in blood and mosquito stages of the parasite life cycle 51 , consistent with a conserved role for this motor in the parasite cell division machinery. Although we know very little else about the function of this motor, we infer from our biochemical and structural data that Plasmodium kinesin-5 is likely to play a MT-organising role within parasite spindles. Kinesin-5 is not essential during the blood stages of the Plasmodium life cycle 51,52 . However, knockout of P. berghei kinesin-5 substantially reduces the number of sporozoites in oocysts and mosquito salivary glands. This highlights the operational diversity of replication at different parasite life cycle stages in general, and specifically suggests a key role for kinesin-5 in the multiple rounds of mitosis that occur during sporozoite production in the mosquito host 51 .
There is increasing focus on tackling malaria not only during the symptomatic blood stages of the parasite life cycle but also by perturbing Plasmodium transmission between vector and host to facilitate malaria control at the population level 53 . Intriguingly, despite the reduction of sporozoite numbers in kinesin-5 knockout parasites, the residual sporozoites achieved normal infectivity. Nevertheless, the role of kinesin-5 in this life cycle stage sheds light on parasite transmission vulnerabilities. Particularly given the distinct parasite number threshold that supports onward transmission between vector and host 54 , combinations of perturbations in sporozoite production could enable transmission control. Moreover, the diverse mechanisms by which small molecules can inhibit HsK5 function have demonstrated that some modes of motor inhibition can be more functionally disruptive than preventing MT binding or than removing motor function completely, for example by trapping it in a tightly bound MT state. In fact, tight MT binding is the proposed mechanism for the antifungal small molecules that target C. albicans kinesin-5, despite that motor being nonessential 55 . In the context of these promising findings, our data provide a structural basis for future investigations into parasite-specific kinesin inhibitors.

Protein expression and purification
The PfK5 motor domain (PfK5MD, residues 1-493) was engineered such that 105 amino acids of the asparagine/lysine rich insertion in loop6 (residues 175-269) were removed to facilitate protein expression. The resulting construct, which we refer to as PfK5ΔL6-MD, was cloned in a pET-151D-TOPO vector (Invitrogen) with an N-terminal His6-tag and TEV protease cleavage site, and each preparation was expressed in 12 L BL21 Star DE3 E. coli cells (Invitrogen) grown in LB media. Cells were grown at 37 o C until they reached an optical density of 0.8-1.0, and were then induced with 0.

Protein labelling
Gibson assembly 57 was used to clone in-frame a C-terminal SNAP-tag on PfK5ΔL6-MD (PfK5ΔL6-MD-SNAP) for use in total internal reflection microscopy (TIRFM) experiments. Expression and purification of PfK5ΔL6-MD-SNAP was performed as for PfK5ΔL6-MD, except the anion exchange chromatography step was altered as follows: concentrated and desalted sample was added to a 1 mL HiTrap Q HP column, washed with 10 CV IEX W buffer, and eluted with a 20 CV gradient elution to 500 mM NaCl. Eluted fractions containing PfK5ΔL6-MD-SNAP were pooled. PfK5ΔL6-MD-SNAP was biotinylated or fluorescently labelled by overnight incubation at 4 o C with SNAP-Biotin  or SNAP-Surface  Alex Fluor  647 (New England BioLabs) with at least a 3:1 molar excess of these labels to PfK5ΔL6-MD-SNAP. Free SNAP-ligand was removed by 2 repeats of buffer exchange in T50K20 buffer (0.5 mL Zeba 7K MWCO spin columns).

MT Preparation
Purified and lyophilised unlabelled, X-rhodamine labelled, or biotin labelled Sus scrofa brain tubulin -except X-rhodamine which was from Bos taurus -(catalogue numbers T240C, TL620M T333P, respectively, all >99% pure, Cytoskeleton Inc.) was reconstituted to 10 mg/mL in BRB80 (80 mM PIPES pH 6.8, 2 mM MgCl2 1 mM EGTA pH 6.8), centrifuged at 611453 g for 10 mins at 4 o C, and the supernatant snap frozen in liquid nitrogen and stored at -80 o C. Double cycle GMPCPP polymerisation was performed as follows: reconstituted tubulin was supplemented with 1 mM GMPCPP (Jena Biosciences) and incubated for 5 mins on ice. Tubulin was then polymerised at 4-5 mg/mL for 20 mins at 37 o C. MTs were then pelleted at 611453 g for 10 mins at 23 o C, washed twice, then resuspended, both with BRB80. MTs were then depolymerised on ice for 15 mins, and a second round of polymerisation performed as above. MTs for the ATPase assay were pelleted on a 50 % v/v sucrose/BRB80 cushion after the second polymerisation, to aid separation of MTs from unpolymerised tubulin.
For labelled MTs, X-rhodamine and unlabelled tubulin were mixed in a 1:9 ratio, or Xrhodamine, biotinylated, and unlabelled tubulin were mixed in a 1:1:8 ratio and polymerised at approximately 4 mg/mL total tubulin with 1 mM GTP at 37 o C for 20 mins. 40 M Paclitaxel (Merck) dissolved in DMSO was then added, followed by a further incubation at 37 o C for 15 mins, and then incubation at room temperature for at least one day before use.
To prepare polarity marked MTs 58 , long, dimly labelled MTs (1:9 ratio of X-rhodamine to unlabelled tubulin) were polymerised at 2 mg/mL total tubulin for 2 hours with 1 mM GMPCPP. To prepare NEM-tubulin, by which minus-end MT growth is blocked, 8 mg/mL unlabelled tubulin was incubated with 1 mM N-ethyl maleimide (Sigma) on ice for 10 mins, then with 100 mM ß-mercaptoethanol (Sigma) for 10 mins. To polymerise the bright plus end MT cap, NEM-tubulin was mixed 1:1 with X-rhodamine tubulin and incubated at 37 o C for 15 mins. Finally, long, dim MTs were pelleted (15 mins, 17,000 g, room temp.), the pellet resuspended with bright MT caps, incubated 37 o C for 15 mins, then 40 M paclitaxel added.

Steady-state ATPase activity
An NADH-coupled ATPase assay 59  MT gliding and single molecule TIRFM assays TIRFM assays with fluorescently labelled protein were performed in a flow-chamber, created by adhering biotin-PEG coverslips (MicroSurfaces Inc.) to glass slides with doublesided tape. To prepare flow-chambers for the MT gliding assay, the following treatments were made. (1) 0.75 % Pluronic acid + 5 mg/mL casein was added for 5 min, and the chamber was washed with T50K20 buffer + 20 M paclitaxel (T50T). (2) 0.5 mg/mL neutravidin was added for 2 mins, then the chamber was washed twice with T50T + 1 mg/mL casein (T50TC). Imaging was manually or automatically performed on a G2 Polara (FEI/ThermoFisher) operating at 300 kV using SerialEM 61 . Images were collected on a K2 summit detector in counting mode, with a GIF Quantum LS Imaging Filter (Gatan). The sample was exposed with 58 e -/ Å 2 for 18 sec and 60 frames collected with a pixel size at the sample of 1.39 Å.
All image processing steps were performed with RELION v3.0 and 3.1 27,62 except where otherwise noted. Beam induced motion in micrographs was corrected using RELION's implementation of MotionCor2, and the CTF was determined for each micrograph using Gctf 63 . The start/end coordinates of each MT were manually assigned, and MT particles with a box size of 432 pixels were extracted every 82 Å and normalised. Alignment and asymmetric reconstruction of 14 protofilament (PF) MTs (which are the dominant MT type in GMPCPP preparations 64 ) was performed using MiRP 28 , as follows, with new MiRP updates noted. RELION parameters used for MiRP PF sorting, initial seam alignment and seam checking steps are listed in supp. table 1. Briefly, PF number assignment for each MT was performed with supervised 3D classification. As part of the MiRP update undertaken during this work, after 3D classification, PF number class assignments for each MT were smoothened by calculating the mode of class assignment over a 7 particle window. Where changes in class assignment occurred within a single MT -due to for example, changes in PF number or major defects within a single MT -MT regions were subsequently treated as distinct MTs. This improved the homogeneity of each MT, increasing confidence in protofilament number assignment and seam location determination. Initial seam alignment was then performed with several iterations of 3D alignment. This was followed, as previously, by Rot angle and X/Y coordinate fitting -however, a local search step was added to improve Rot angle and X/Y shift assignment. Seam checking via supervised 3D classification was then performed, and MTs with less than 50 % confidence in seam class assignment were removed. C1 reconstructions were obtained with a 3D auto-refine run (using the parameters for X/Y refine in supp. table 1, with a solvent mask obtained from a 3D reconstruction of seam checking results), followed by per-particle CTF refinement, Bayesian polishing, and beam-tilt estimation, then a second 3D auto-refine with these new corrections. MiRP was also updated to improve useability, by creating three programs for each MiRP step that can be operated from the RELION v3.1 GUI (https://github.com/moores-lab/MiRPv2).
Symmetrised maps were obtained by first performing 2D classification without alignment (200 classes, T = 8), and selecting well-aligned classes with many particles and an estimated resolution better than 6 Å. A 3D auto-refine run was then performed where 14-fold local symmetry was applied, as previously described 65 . To address PfK5ΔL6-MD heterogeneity, 3D classification was performed at the level of a PfK5ΔL6-MD:αβ-tubulin asymmetric unit. For this, symmetry expansion was applied to all particles, then 3D classification (3 classes, T = 256) without alignment and a mask around one PfK5ΔL6-MD site opposite the seam was applied. This resulted in one class with clear PfK5ΔL6-MD decoration, which was selected and subjected to a 3D auto-refine procedure. The MT-bound nucleotide-free and AMPPNPbound PfK5ΔL6-MD reconstructions are deposited with the Electron Microscopy Data Bank, deposition number 12257 and 12258 respectively.

Sequence analysis, comparative modelling and flexible fitting
To obtain a kinesin-5 family sequence alignment, the motor domains all kinesin-5 family members in the Swiss-Prot database were aligned with MAFFT 66 . A hidden Markov model of this alignment was created, and queried against the UniProt Reference Proteomes database using HMMER 67 . Sequences obtained were then compared to a kinesin profile from the Pfam database 68 , and those with less than 400 identities were removed. Finally, the sequences were aligned with MAFFT, using the L-INS-i method (S12 Data). Secondary structure prediction was performed using Quick2D 69 , using various prediction algorithms [70][71][72][73][74] . A residue was assigned helical or beta-sheet identity if 3 or more prediction algorithms agreed.  78 . During this procedure, the nucleotide binding site (AMPPNP, Mg 2+ , switch-I/II loops, and the P-loop) was defined as a rigid body, while loop regions were treated as "all atoms". Models of the N-terminus, loops 2, 8, 9, 10, 11, and 12, and the neck linker were predicted using Rosetta, firstly using a coarse method (500 models using cyclic coordinate descent with fragment insertion in the centroid modelling step 79 ), then the model with highest cross-correlation was selected for a second prediction (500 models using kinematic closure with a fit to density term in the centroid modelling step 80 ). A local all-atom fit to density step was then performed using the Rosetta Relax procedure including a fit to density term 81 . Finally, the interface between PfK5ΔL6-MD and αβ-tubulin was refined with protein-protein docking restrained by cryo-EM density in HADDOCK 82 as described previously 31 , using the PDB ID 3JAT 83 as αβ-tubulin atomic model. SMOC scores were calculated using the TEMPy software package 84,85 . The molecular models of MT-bound nucleotide-free and AMPPNP-bound PfK5ΔL6-MD are deposited with the Worldwide Protein Data Bank, deposition number 7NB8 and 7NBA respectively.

Visualisation and analysis
Plotting was performed with GraphPad Prism 8. Cryo-EM density and model analysis was done in Chimera 86 and ChimeraX 87 . Protein sequence analysis was done in Jalview 88 . Protein interface areas were calculated with PDBe PISA v1.52 89 .  84,85 for the no nucleotide and AMPPNP state models, that indicate the fit of the model to cryo-EM maps. The SMOC score for the homology model is also show, to demonstrate how the flexible fitting process has improved the models fit to density.

SUPPORTING
S4 Figure. PfK5ΔL6-MD AMPPNP state nucleotide density. To illustrate cryo-EM density corresponding to AMPPNP, synthetic density corresponding to the protein components of the PfK5ΔL6-MD model -i.e. without AMPPNP -was calculated at 6 Å and was subtracted from PfK5ΔL6-MD AMPPNP state cryo-EM reconstruction. The resulting difference density corresponds to the bound nucleotide (modelled as AMPPNP given the sample preparation conditions), αβ-tubulin, and loop5, which was not included in the model.   Table. Restraints used in PfK5ΔL6-MD homology model generation S10 Table. Comparative rotation of PfK5ΔL6-MD helices between the no nucleotide and AMPPNP states. S11 Table. Residue-residue contacts between the PfK5ΔL6-MD no nucleotide and AMPPNP states, and αβ-tubulin. Residues within contact distance were detected in Chimera.