KIF1A mediates axonal transport of BACE1 and identification of independently moving cargoes in living SCG neurons

Neurons rely heavily on axonal transport to deliver materials from the sites of synthesis to the axon terminals over distances that can be many centimetres long. KIF1A is the neuron‐specific kinesin with the fastest reported anterograde motor activity. Previous studies have shown that KIF1A transports a subset of synaptic proteins, neurofilaments and dense‐core vesicles. Using two‐colour live imaging, we showed that beta‐secretase 1 (BACE1)‐mCherry moves together with KIF1A‐GFP in both the anterograde and retrograde directions in superior cervical ganglions (SCG) neurons. We confirmed that KIF1A is functionally required for BACE1 transport by using KIF1A siRNA and a KIF1A mutant construct (KIF1A‐T312M) to impair its motor activity. We further identified several cargoes that have little or no co‐migration with KIF1A‐GFP and also move independently from BACE1‐mCherry. Together, these findings support a primary role for KIF1A in the anterograde transport of BACE1 and suggest that axonally transported cargoes are sorted into different classes of carrier vesicles in the cell body and are transported by cargo‐specific motor proteins through the axon.


| INTRODUCTION
The kinesin superfamily proteins (KIFs) are a large gene family of microtubule-dependent motors; over 45 members have been identified in the mammalian genome. 1 A wide array of cellular cargoes are simultaneously transported within an axon by different motor proteins, including transport vesicles, membranous organelle and nonmembranous cargoes such as cytoskeletal filaments, cytosolic protein complexes and messenger RNAs. 2 As motor-cargo interactions are specific, it is difficult to recognise the specific cargoes of a motor protein among the multitude of other cargoes that are present in the sheer volume of axons. 3 As the types of cargo greatly outnumber the motors, motors must recognise and carry more than one type of cargo for efficient transport, but there appears also to be some redundancy in which motor protein carries which cargo. The identification of the cargoes that each motor carries is one of the most poorly developed areas in the field of microtubule-dependent trafficking. 4 Previous studies addressing this question have been largely dependent on co-immunoprecipitation strategies. These studies have led to the identification of important binding partners, 5-8 but with two major caveats. First, as stationary material will also be precipitated, this approach alone does not show that a motor actually moves the associated cargo. Second, this method may fail to pull-down membranous cargoes that do not interact directly with the motor protein, especially if the membranes are disrupted during extraction. Transient interactions with motor protein during transport may also be missed.
Alternative methods to address this question, such as co-migration analysis in live imaging and the effect of motor protein depletion on cargo movement, have been relatively little used.
KIF1A belongs to the Kinesin-3 family. 9 KIF1A is a neuronspecific kinesin and it also has the fastest reported anterograde motor activity. 10 It is suggested to act as a monomer, 11 although Hammond et al reported that KIF1A motors exist in a dimeric state and only dimerized motors display ATP-dependent processive motility. 12 Previous studies have shown that KIF1A transports a subset of synaptic proteins 1,4 and dense-core vesicles. 13 However, the identity of other cargoes carried by KIF1A is still largely unknown. As a good live This research was supported by grants from BBSRC and the Croucher Foundation (Hong Kong). imaging method is already available for KIF1A, 14 we decided to study which cargoes co-migrate with this motor in dissociated superior cervical ganglions (SCG) neurons. SCG neurons have been used extensively to study axonal transport, largely because the polarity of cell processes (axons vs dendrites) can be easily determined in the absence of markers. This, together with their convenient genetic manipulation by microinjection, 15 provides us with a good in vitro model for evaluating transport kinetics, such as the percentage of moving vesicles and the average and maximum velocities in both the anterograde and retrograde directions.
The deposition of amyloid β (Aβ) peptide is the major pathological hallmark of Alzheimer's disease. Beta-site amyloid precursor protein (APP) cleaving enzyme 1 [beta-secretase 1 (BACE1)], a transmembrane aspartyl protease, initiates the processing of the APP during Aβ generation and therefore is the rate-limiting enzyme in the amyloidogenic pathway. 16 As APP and BACE1 are highly expressed in neurons, it is reasonable to assume there should be a mechanism that segregates the transport of APP and BACE1 to prevent overproduction of Aβ. Despite extensive research, the problem of where wild-type APP and BACE1 meet and the location of the β-cleavage site remain controversial. Previous studies have suggested that APP is transported in the same axonal vesicle as BACE1 and γ-secretase, that APP can then be cleaved by them, and consequently Aβ is produced in these vesicles. 17 This view, however, has been challenged by several studies. A ligation experiment on sciatic nerves found that APP is not co-transported with BACE1. 18 This finding was further supported by a video microscopy study showing that APP is transported on a vesicle distinct from BACE1. 19 A recent study using an optical assay, however, found that APP and BACE1 were co-transported in axons and that they interacted during transit. 20 Understanding axonal BACE1 transport is important because abnormal accumulation of BACE1 at the presynaptic terminals has been reported in post-mortem AD brain. 21 This finding suggests that elevation of local BACE1 level could promote the generation of Aβ and this could contribute to the development of AD. 22 However, the molecular players involved in BACE1 transport in axons are still not completely identified.
In this study, we first characterized the axonal transport of a panel of cargo candidates, ranging from signalling molecules to proteins that are involved in neurodegenerative disorders. Two-colour live imaging showed that BACE1-mCherry moves together with KIF1A-GFP in both the anterograde and retrograde directions. We identified several cargoes that show little or no co-migration with KIF1A-GFP and demonstrated that these also move independently from BACE1-mCherry. Our findings support a primary role for KIF1A in the anterograde transport of BACE1 and suggest the existence of independently moving groups of cargoes in neurons. KIF1A is only responsible for the transport of a sub-population of these moving cargoes.

| Identification of axonally transported cargoes that can be easily tracked in SCG axons using live imaging
The aim of this study was to identify the cargoes that KIF1A carried.
Therefore we first identified cargoes that can be easily tracked in SCG axons using live cell imaging. We visualized the axonal transport of a panel of cargoes chosen from different categories, ranging from signalling molecules to proteins that are involved in neurodegenerative disorders (see Table 1). We first fused each cargo to a fluorescent tag (either mCherry or RFP). We then microinjected a low level of construct encoding fluorescently labelled protein into the nuclei of 7DIV SCG dissociated cultures and visualized the transport of the fluorescently labelled protein in the axons at 37 C 6-8 hours after microinjection (schematic in Figure 1A). Microinjected cells that appeared healthy by phase-contrast were imaged by video microscopy. To determine whether the use of microinjection can physically damage the injected cells and thus affect vesicle motility, we evalu- Of the 14 cargo candidates that we tested 6 showed a punctate structure that moved bidirectionally within the axons so movement along the axon could be represented by kymographs. Interestingly, all six cargoes whose movements can be tracked easily are involved either in neurodegenerative diseases or neurodegeneration. For example, both NMNAT2 (Nicotinamide Nucleotide Adenylyltransferase 2) and SCG10 (superior cervical ganglia-10) are involved in the Wallerian degeneration pathway. NMNAT2, which shares the same NAD-synthesizing enzyme activity with WLD S (slow Wallerian degeneration protein), is now well established as an essential survival factor that prevents spontaneous degeneration of healthy axons. Its constant replenishment from the cell body into axons by anterograde transport is essential for axon survival and its specific depletion is sufficient to induce Wallerian degeneration. 15 Surprisingly, we found that most of the signalling molecules showed a diffuse signal throughout the axons, so their movements could not be easily tracked. These diffuse signals may have occurred because the majority of the fluorescently labelled molecules were unattached and could diffuse freely through the axon. These unattached cargoes lead to high background noise that makes it particularly difficult to track any individual microtubule-dependent moving vesicles. 23

| KIF1A has a high percentage of co-migration with BACE1
Next, we wanted to find the cargoes that co-migrate with KIF1A.
Toward this, we microinjected dissociated SCG cultures with low levels of KIF1A-GFP/Cargo candidates-RFP (or mCherry) and simultaneously visualized their transport in axons 6-8 hours after microinjection where the orientation of the axons in relation to the cell body was unambiguous (schematic in Figure 1A). This is the time when these fluorescently labelled proteins are just beginning to be expressed and the fluorescent protein signals are at relatively low levels. This could avoid the problem of artefacts, which are produced by high levels of expression. To demonstrate this pattern, we measured the average fluorescent intensity of neurites at three different time points after microinjection (6, 24 and 48 h) and have presented images captured at these time points in Figure S1B, Supporting Information. We found that there was an increase in fluorescent intensity with time after microinjection. At 6 hours, the average level of fluorescent intensity was relatively low compared to its levels at 24 or 48 hours ( Figure S1C). In addition, to ensure that the signals from the GFP and RFP channels are comparable, we generated line fluorescent intensity profiles along the axon co-expressing NMNAT2-GFP and NMNAT2-mCherry. The profiles generated from the GFP and RFP channels are quantitatively similar in terms of fluorescent peak locations ( Figure S2).
To analyse the extent of co-migration between KIF1A-GFP and cargo candidates, kymographs were generated for each channel (GFP and RFP/mCherry) and for their overlays (Figure 2A). To quantify the extent of co-migration, only particles that moved during at least part of the time-lapse recording were taken into account. The percentage of co-migration in the axon was then analyzed by comparing pairs of kymographs generated for each fluorescently tagged cargo. We found that KIF1A had a different percentage of co-migration with different cargo candidates ( Figure 2B): It had the highest percentage of co-migration with BACE1 but a relatively low percentage of comigration with both NMNAT2 and mitochondria. BDNF, APP and SCG10 all showed intermediate levels of co-migration with KIF1A, with BDNF being most strongly associated of these.

| KIF1A is functionally required for the transport of BACE1-mCherry
The demonstration of a high percentage of co-migration between KIF1A-GFP and BACE1-mCherry raised the question of whether KIF1A is functionally required for transporting BACE1 or whether other motor proteins can substitute when KIF1A activity is removed.
To address this question, we first asked whether the transport of BACE1 vesicles could be disrupted by the expression of a mutant kinesin construct. We expressed a KIF1A-GFP mutant with a point mutation (Thr-312 to Met) that impairs the motor domain activity (KIF1A-T312M-GFP) in dissociated SCG neurons. This point mutation occurs in the α-helix 5 in KIF1A that is close to the centre of the surface for binding microtubules. 24 It impairs the motor activity by interrupting the interactions between the microtubules and kinesin. We observed that most of the KIF1A-T312M particles were localized in A labile axon survival factor whose constant replenishment by anterograde axonal transport is a limiting factor for axon survival We believe that this mutant acts by exerting a dominant negative   to bind to BACE1 ( Figure 3A,B). However, the expression of KIF1A-TM-GFP had no effect on the axonal transport of mitochondria ( Figure 3A,B), which was previously found to be a cargo of KIF1B 25 and KIF5 [26][27][28][29] and have 0% of co-migration with KIF1A ( Figure 2B).
Further, to test whether endogenous BACE1 distribution is affected; we monitored the axonal distribution of BACE1 in neurons expressing KIF1A-T312M using immunohistochemistry. Compared to the control, there was a shift of BACE1 protein from neurites to cell bodies ( Figure 4A), suggesting that the transport of endogenous BACE1 was impaired. However, the expression of mutant KIF1A protein did not appear to affect the localisation of endogenous APP, which had a relatively low percentage of co-migration with KIF1A ( Figure 2B). APP protein was still observed in both the cell body and axonal compartment ( Figure 4B).
Second, we asked whether the depletion of endogenous KIF1A could affect the axonal transport of BACE1 vesicles. We imaged the axonal transport of BACE1-mCherry in axons co-expressing siRNA against KIF1A. We observed a significant reduction in BACE1-mCherry moving particles 48 hours after microinjection compared to cells either without siRNA or with non-targeting (control) siRNA ( Figure 5A,B). To further show the specificity of our siRNA, we also looked at the axonal transport of mitochondria in neurons expressing KIF1A siRNA. We found that the axonal transport of mitochondria is not affected by the non-targeting siRNA (control) or knocking down of KIF1A protein ( Figure 5A,B). These findings suggest that KIF1A activity is functionally required for the transport of BACE1-mCherry.

| BACE1, NMNAT2 and mitochondria are transported in distinct carriers
We previously reported that NMNAT2 and mitochondria move independently of one another. 30 Interestingly, our data showed that both NMNAT2 and mitochondria also have a very low percentage of comigration with KIF1A-GFP ( Figure 2B). This prompted us to ask whether BACE1 is transported by a different carrier from those carrying either NMNAT2 or mitochondria as this may suggest the existence of several independently moving groups of cargoes. We examined the percentage co-migration of BACE1-mCherry with NMNAT2-GFP and also the percentage co-migration of BACE1-mCherry with Mitochondria-RFP using live imaging of each pair of cargoes in the same axon.
In SCGs neurons co-injected with BACE1-mCherry and NMNAT2-GFP ( Figure 6A), the localisation of mobile BACE1 and NMNAT2 was mostly non-overlapping. Consecutive images from a 2.5-second time-lapse series are shown in Figure 6B with yellow arrowheads in the GFP channel ( Figure 6B) marking an anterogradely moving NMNAT2-GFP vesicle not containing BACE1-mCherry. Grey arrowheads in the RFP channel ( Figure 6B) mark an anterogradely moving BACE1-mCherry vesicle not containing NMNAT2-GFP. Thus, these particular vesicles moving BACE1 and NMNAT2 are transported by distinct carriers and the quantification ( Figure 7B) shows this is typical of the large majority of such vesicles. Similar results were found when BACE1-GFP and mito-RFP were co-injected ( Figure 7A). Yellow arrowheads in the GFP channel ( Figure 7A) mark an anterogradely moving BACE1-GFP vesicle not containing RFP and grey arrowheads in the RFP channel ( Figure 7A) mark an anterogradely moving mitochondria not containing GFP. Thus, BACE1 and mitochondria are also moving independently from one another.
We also compared the transport kinetics of mobile BACE1-mCherry/NMNAT2-mCherry/Mitochondria-RFP vesicles using a plugin developed by our group in the software IMAGEJ. 31 Table 2). Taken together, these findings suggest that BACE1, NMNAT2 and mitochondria are transported in distinct carriers and that KIF1A is a selective kinesin responsible for moving one group of independently moving cargoes in axons.

| BACE1, NMNAT2 and Mitochondria are markers for non-overlapping components of axonal transport
The live imaging technique was introduced in the late 1980s to overcome the limitations of using whole-nerve radiolabelling in This method achieves a much higher temporal and spatial resolution than the traditional radiolabelling approach. Most importantly, individually moving cargoes can also be tracked. 32 In this study, we showed that KIF1A is the primary anterograde motor protein mediating the axonal transport of BACE1. To the best of our knowledge, this is the first study to identify the kinesin responsible for moving BACE1 using live cell imaging, a result further of most axonal transport cargoes, as their movement may be influenced as much by local energy demands as by transport capacity. Moreover, only a minority of mitochondria move in axons. To address this limitation, we generated a new transgenic mouse to study the axonal transport of NMNAT2, which is an essential survival factor for the maintenance of healthy axons. The transport of NMNAT2 can be assayed in the central nervous system explants and in peripheral nerves. 35,36 Our data showed that BACE1 also moves independently of both NMNAT2 and Mitochondria. Hence, BACE1 appears to be a useful marker for studying a third, non-overlapping component of axonal transport. A useful future research direction would be to cross the transgenic mice expressing fluorescently tagged BACE1 37 with both MitoS and NMNAT2 transgenic models to simultaneously report on multiple axonal transport motors. We could then follow the axonal transport of three independently moving cargoes using time-lapse imaging. In this way, we would be able to obtain a more comprehensive view of how axonal transport is affected in aging 36 or in models of neurodegenerative diseases. 38

| Retrograde transport of KIF1A and BACE1
In this study, we show that KIF1A has a high percentage of comigration with BACE1 in both the anterograde and retrograde directions. As all of the axonal microtubules are orientated uniformly, with the plus ends (polymerising ends) facing the axon terminals, the observation in this and other studies 13,14 that the plus-end driven kinesin KIF1A is capable of moving retrogradely is rather surprising. It has been proposed that cargoes move bidirectionally due to a tug-ofwar with oppositely directed motor proteins (ie, dynein complex) attached to the same cargo. 39,40 However, the results that both knocking down KIF1A and expressing the T312M mutant (probably a dominant negative) diminished BACE1 motility in both the anterograde and retrograde directions challenges this model. It may be that without anterograde transport there is less BACE1 available for retrograde transport but this observation also fits with the previously suggested "co-dependence" model 39 and with previous studies in which inhibition of Drosophila Unc-104 (KIF1A) disrupts axonal transport of dense core vesicles in both directions. 41 Instead of competing mechanically with dynein motors, we suggest that KIF1A itself could act as a cargo for the dynein complex. Indeed, a previous study found that KIF1A is a component of the cytoplasmic dynein complex. 42

| Redundancy among kinesins
As a relatively small number of kinesins have to move a much larger number of transport cargoes, it is clear that each kinesin has multiple roles. Indeed studies in knockout animals have previously implicated KIF1A in the movement of synaptic vesicle precursors. 44 However, it remains unclear how much redundancy exists between different kinesins in the function of moving an individual transport cargo. Mutation of a single kinesin in humans or mice is in many cases sufficient to cause axonal growth or degeneration phenotypes, indicating that not all roles of the respective kinesin can be compensated for by others. [45][46][47] This includes a possible role for KIF5A in the axonal transport of mitochondria, one of the cargoes we find   shows minimal co-migration with KIF1A. 48 However, there are also indications of some functional redundancy in the movement for example of neurofilaments 49 and late endosomes. 50 Thus, it appears likely that the degree of functional redundancy differs on a cargoby-cargo basis.

| Role of KIF1A in separating APP and BACE1 trafficking
Despite intensive investigation, the location where wild-type APP and BACE1 meet and the site of the β-cleavage are still not completely understood. In our study in SCG neurons, we found that KIF1A has a high percentage of co-migration with BACE1 (about 70% in the anterograde direction and about 80% in the retrograde direction) but a relatively lower percentage of co-migration with APP (about 20% in both the anterograde and retrograde direction).
Thus, in this neuron subtype at least, overexpressed BACE1 and APP are transported primarily by different kinesins. Moreover, we found that most moving APP and BACE1 vesicles are nonoverlapping ( Figure S4). We used dual-colour, live-cell imaging of primary SCG neurons to visualise the co-migration of APP with BACE1. The co-migration of APP-GFP and BACE1-mCherry was then quantified using kymographs; only moving particles were counted. We found that the majority (>50%) of the APP particles were moving separately from the BACE1 particles in the anterograde direction. However, it should be noted that the APP construct used in this study was tagged at the carboxy-terminal; therefore, this method does not distinguish between the transport of fulllength APP and fragments generated by cleavage that contain the carboxy-terminal region (CTFs). Similarly, the effect of the overexpression of BACE1 on APP transport is not properly understood.
Thus, the question of where APP and BACE1 meet is difficult to resolve using live imaging alone.
Our results do raise the interesting prospect that changes in KIF1A could influence the axonal transport of BACE1, and through this influence the intracellular site and rate of processing of APP.
Interestingly, an increase in BACE1 was paradoxically found to lower the production of Aβ, 51 leading to the suggestion that shifting processing to the cell body, as for example by reducing KIF1A activity (Fig 4A), could be beneficial. In contrast, lowering expression of KLC leads to more Aβ generation, potentially by increasing the dwell time of APP processing products in the axon. 52

| Dissociated SCG cultures
Dissected SCG ganglia were incubated in 0.025% trypsin (Sigma) in PBS Medium was changed the next day and every 2-3 days after that.

| Plasmids construct and siRNA reagents
Expression vectors encoding EGFP-tagged SCG10 and BDNF were gen-  Membranes were then scanned and quantified using the Odyssey imaging system (LI-COR Biosciences, Lincoln, North Carolina).

| Microinjection of DNA constructs
Microinjection mixes of plasmid DNA were prepared in 0.5× PBS(−) and passed through a Spin-X filter (Costar, Glasgow, UK). Eppendorf Femtotips were loaded with the microinjection mix and injection was performed using an Eppendorf 5171 transjector and 5246 micromanipulator system on a Zeiss Axiovert 200 microscope. All injections were carried out directly into the nuclei of dissociated SCG neurons.
For live cell imaging, a maximum total DNA concentration of 0.05 μg/μL in the injection mix was used. A total of 75 cells were injected per dish and imaging was performed 6-8 hours after microinjection. As on only the injected cells express fluorescent protein, no other marker of injection is required. Time-lapse images were acquired using an Andor wide-field/TIRF imaging system comprising Nikon Ti-E microscope, Nikon 100×1.49
Cultures were maintained at 37 C in an OKO labs environment chamber. Time-lapse images of proximal axon regions were captured for 24 seconds with a 250 ms time-lapse interval using iQ3 live cell imaging software (Andor). A total of 10-12 movies from different neurons were captured from each culture dish. The software IMAGEJ was used for generation of kymograph and analysis of particle velocities.

| Data analysis
Statistical analyses and graph fitting were performed using GRAPHPAD PRISM 6.0 (GraphPad Software Inc., La Jolla, CA 92037 USA)