Tau Protein’s Placement and Interaction with Memory-Related Factors Depend on Its Type and Region in Human Stem Cell–Derived Neurons

Image analysis was performed using Fiji/ImageJ (V1.53t, NIH). All images were blinded using the Fiji blind analysis tool.

Primary rodent cultures often show poor axonal sorting of exogenous Tau (Iwata et al. 2019; Xia et al. 2016), and human neuron models for axodendritic Tau sorting are scarce. We differentiated human Ngn2‐transgenic WTC11 iPSCs (Miyaoka et al. 2014; Zhang et al. 2013) with and without biallelic MAPT knockout (Buchholz et al. 2025) (Figure S1a) into cortical glutamatergic neurons (Buchholz, Bell‐Simons, Cakmak, et al. 2024; Tracy et al. 2022; Wang et al. 2017) to characterize axonal Tau sorting.

First, we quantified Tau sorting in WT iPSC‐neurons at different stages of differentiation using constitutive lentiviral dTomato expression (Figure 1a) as a cellular volume marker due to its unbiased distribution pattern (Figure S1d,e). The axonal enrichment of endogenous Tau in the axon compared to the soma increases strongly after week 1, stabilizes by week 6, and elevates towards week 10 with high replicate variability, while MAP2 shows no axonal enrichment at any age (Figure 1b). Axonal Tau sorting coincides with the detectable formation of the axon initial segment (AIS), indicated by the accumulation of ANKG and TRIM46 (Figure S2), in line with previous studies (Lindhout et al. 2020). Both Tau and MAP2 showed modest dendritic enrichment, with MAP2 being more enriched at week 3 (Figure S1b). Some dendritic Tau presence is expected, as this has been documented previously, and there is evidence for dendrite‐specific functions of Tau under healthy conditions, such as negative regulation of NMDAR‐induced synaptic activity and modulation of spine turnover (Ittner and Ittner 2018; Ittner et al. 2010; Zempel et al. 2013). Both Tau and MAP2 showed little nuclear localization throughout differentiation (Figure S1c).

Next, we checked the axonal sorting efficiency of HA‐tagged 0N3R in MAPT‐KO iPSC‐neurons using doxycycline‐inducible lentiviral expression of 0N3R^HA‐Tau and P2A peptide‐coupled dTomato (Buchholz, Bell‐Simons, Cakmak, et al. 2024) (Figure 1c). Lentiviral particles were transduced into 6‐weeks‐old MAPT‐KO iPSC‐neurons at low titers, and doxycycline was added for equimolar expression of 0N3R^HA and dTomato (Figure 1c,d). We quantified axonal sorting of 0N3R^HA after 12–13 days (Figure 1d) and compared it to endogenous Tau (Figure 1e). Strikingly, 0N3R^HA sorted with high efficiency, similar to or even more efficient than endogenous Tau, despite an increase in overall abundance due to robust overexpression (Figure S3h). The dendritic and nuclear enrichment of 0N3R^HA was similar to or modestly increased, respectively (Figure S3b,c). Axonal, dendritic, and nuclear MAP2 was not enriched after 0N3R^HA overexpression compared to MAPT‐KO iPSC‐neurons (Figure S3d–f). The correlation analysis between sorting and protein levels revealed a modest positive correlation (Figure S3g).

Taken together, exogenous 0N3R^HA showed endogenous‐like axonal sorting in MAPT‐KO iPSC‐neurons without disturbing MAP2 polarization, and we employed this model to study the sorting behavior of truncated and phosphorylation‐mutant Tau^HA (Figure 1c; Figure 2a; Table 1).

First, we validated protein expression and 2A cleavage of all mutant Tau^HA constructs in MAPT‐KO iPSC‐neurons. Unmutated 0N3R^HA appeared at the expected size of ~55 kDa; a larger band of uncleaved dTomato‐2PA‐0N3R‐Tau^HA was barely visible at ~95 kDa, suggesting highly efficient cleavage of the P2A (Figure 2b; Figure S4a–c; Figure S7f). For most of the truncated Tau^HA constructs (Figure 2a), robust expression of correctly sized products was detected, supported by the immunofluorescent data (Figure S4d). For Cterm^HA, PRR‐3R^HA, only3R^HA, and only4R^HA, very weak or no distinct bands were detectable (Figure S4a–c), in line with decreased protein levels compared to 0N3R‐Tau^HA (Figure S4d), indicating posttranslational degradation. All three constructs were thus excluded from further analysis. We observed three distinct bands for noPRR2‐Tau^HA that might hint towards variations in phosphorylation or other posttranslational modifications (PTM). Cterm^HA, PRR + Cterm^HA and only4R^HA showed slightly increased sorting with higher protein amounts; all other fragments had no detectable correlation (Figure S4e,f).

AT8‐mutant Tau^HA constructs were all expressed robustly with expected sizes (Figure 3a and Figure S5c). Higher molecular bands of uncleaved fusion protein were negligibly faint suggesting almost complete P2A cleavage (Figure 3b; Figures S5a,b and S7f). Correlation analysis showed a positive correlation of protein levels and axonal sorting for S199E^HA, T205E^HA, AT8‐allE^HA, and AT8‐allA^HA (Figure S5d,e).

KXGS‐ and double‐mutant Tau^HA constructs were expressed robustly (Figure S6d) and detected with distinct double bands around 50 kDa with either roughly equal proportions (3xKXGA^HA, 3xKXGE^HA, AT8 + KXGS all E^HA, AT8 + KXGS all A^HA) or a dominant larger band (1xKXGA^HA, 1xKXGE^HA) (Figure 4b and Figure S6a–c). Higher molecular bands of uncleaved fusion proteins were negligibly faint for all constructs (Figure 4b; Figures S6a–c and S7f).

In brief, the investigated Tau^HA constructs were mainly expressed in our MAPT‐KO iPSC‐neurons without signs of protein degradation, except for Cterm^HA, Nterm^HA, and PRR + 3R^HA, only3R^HA, and only4R^HA, and with highly efficient cleavage of 2A‐coupled dTomato reporter proteins.

Next, we quantified the sorting efficiency of all truncated Tau^HA constructs in relation to 0N3R^HA‐Tau in MAPT‐KO iPSC‐neurons. When the N‐ or C‐terminal tails were missing (No‐Ntail^HA, No‐Ctail^HA), we saw efficient axonal sorting while lacking larger parts of the C‐terminus (Nterm^HA) led to failed axonal enrichment (~5%–20% efficiency of 0N3R^HA) (Figure 2f). Addition of the PRR domain (PRR + Cterm^HA) rescued axonal targeting (Figure 2d) but the microtubule‐binding domain alone (PRR + 4R^HA) was in turn much less abundant in the axon. Significantly decreased axonal sorting was observed when Tau was lacking the repeat domains completely (no4R^HA) (Figure 2f). Remarkably, the changes in dendritic enrichment of truncated Tau^HA constructs followed a similar pattern (Figure 2g). Direct correlation of the axonal and dendritic enrichment (axon/dendrite ratio, Figure S7b) revealed that most constructs had no changes in axon to dendrite enrichment, compared to 0N3R^HA. Only for noPRR2^HA and Cterm^HA, which showed decreased protein levels (Figure S4d), the axon/dendrite ratio was significantly decreased, indicating axon‐specific impairments of intracellular sorting for those constructs (Figure S7b). The nuclear levels were not changed for any truncated Tau^HA construct except for Nterm^HA, likely due to the somatic inclusion formation (Figure 2h).

We tested the correlation between sorting efficiency and microtubule affinity, which was determined by Tau‐tubulin binding assays (Gustke et al. 1992; Gustke et al. 1994) (Figure S7e), and found that the axonal enrichment increased with higher microtubule affinity (Figure S7c), in contrast to earlier findings (Iwata et al. 2019). We also observed a positive correlation for the axon/dendrite ratio (Figure S7d). This indicates that microtubule affinity is essential for axonal, but also for dendritic enrichment.

For the AT8‐mutant Tau^HA constructs, axonal sorting was equally efficient for all constructs with pseudophosphorylation or dephosphorylation. Only the phospho‐dead AT8‐allA^HA construct was slightly higher axonally enriched than 0N3R^HA (Figure 3e). The AT8‐allA^HA construct and S199A^HA were also modestly increased in dendrites (Figure 3f). The nuclear enrichment was not changed for any of the AT8‐mutant Tau^HA constructs (Figure 3g).

For our KXGS‐ and double‐mutant Tau^HA constructs, axonal sorting efficiency gradually declined from 1xKXGA^HA to 3xKXGA^HA and from 1XKXGE^HA to 2xKXGE^HA to 3XKXGE^HA (Figure 4e), but without reaching statistical significance. The axon/dendrite ratio was unchanged for all constructs (Figure S7b) as dendritic enrichment was similar to the axonal pattern (Figure 4f). Axonal targeting of double mutants was similar to KXGS^HA mutants, in line with the missing effect of AT8 single mutants (Figure 3e). The nuclear enrichment of all KXGS‐ and double‐mutant Tau^HA constructs was not different from 0N3R^HA (Figure 4g).

Of note, the intracellular sorting behavior of endogenous MAP2 was not changed upon expression of any truncated or AT8/KXGS‐modified Tau^HA construct (Figure S3g–i; Figures 4f–h and 5g–i) indicating that expression of Tau, even though also markedly present in the dendrites, did not displace MAP2.

Taken together, we found that the absence of the PRR2 domain leads to axon‐specific Tau sorting deficits, and that microtubule affinity was not linked to impaired axonal sorting in both truncated and phospho‐mutant Tau^HA constructs.

Next, we studied Tau interactions in MAPT‐KO iPSC‐neurons by using biotinylation‐based TurboID proximity labeling (Branon et al. 2018; Cho et al. 2020). To unravel isoform‐specific interactors, we coupled the Tau isoforms 0N3R and 0N4R to the biotin ligase BirA. The non‐sorting construct Nterm^HA was coupled to detect sorting‐specific Tau binding partners.

First, we successfully confirmed the proper expression of all BirA‐Tau^HA constructs (Figure S10a) and efficient 2A‐based cleavage from dTomato (Figure S8a,b). The axonal sorting of BirA‐0N3R^HA and BirA‐0N4R^HA was similar to that of uncoupled 0N3R^HA in immunostained neurons, while there was, as expected, no axonal sorting of the control constructs BirA^HA and BirA‐Nterm^HA (Figure 5b–e). For BirA‐Nterm^HA, we observed the puncta‐like distribution typical for Nterm^HA (Figure 5b, see also Figure 2c) (Bell et al. 2021). The chosen labeling time ensured sufficient axonal penetration of biotin (Figure 5c,f) without excessive unspecific labeling (Cho et al. 2020). Expression of BirA‐0N3R^HA without biotin application revealed only basal biotinylation by BirA (Figure 5g), and correlation analysis of sorting efficiency and protein levels was negative for all BirA‐Tau^HA constructs (Figure S8c). Relative protein levels were similar for all BirA‐Tau^HA constructs (Figure S7d).

All in all, we could validate the expression, cleavage, and efficient axonal sorting of our BirA‐Tau^HA constructs, thus being suitable for Tau interaction analysis.

For interactome analysis, biotinylated proteins were enriched with avidin‐coated beads, digested, and analyzed with liquid chromatography–coupled mass spectrometry (LC–MS/MS) (Figure 6a).

We detected 744 proteins in our BirA‐Tau^HA constructs (Table S1a), of which 62 proteins were significantly enriched for at least one construct compared to both control constructs, onlyBirA^HA and only0N3R^HA (Table S1b). We compared the identified interactors of BirA‐0N3R^HA and BirA‐0N4R^HA (33 out of 62 proteins) with data from a study using Tau‐APEX2 in the same parental iPSC line (Tracy et al. 2022), and with data from a human Tau interactome meta‐analysis (Kavanagh et al. 2022). We found an overlap of 23 proteins (69.7%) and 24 proteins (72.7%), respectively (Table 2). A gene ontology (GO) terms analysis of all 33 isoform‐related hits (Tables S7 and S8) revealed a strong association with biological processes such as microtubule function, cytoskeletal protein binding and organization, or nucleotide and ribonucleotide binding (Figure 6b,c), matching with the known major interaction partners of Tau (Guo et al. 2017; Kavanagh et al. 2022). In sum, we confirmed the successful enrichment of Tau‐interacting proteins in our experimental setup.

Next, we determined isoform‐specific interactors out of the 33 proteins (Figure 6d; Figure S11g; Tables S2 and S9). We found nine 0N3R‐specific binding partners, of which three were either directly, LUZP1 and ZO1, or indirectly, SPD2B, linked to the f‐actin cytoskeleton. The 0N3R interactor MAP4 is a microtubule‐binding protein that is structurally related to Tau as both are members of the same MAP family (Nishida et al. 2023; Tokuraku et al. 2010).

Out of the 10 specific interactors of 0N4R, at least 5 proteins are connected to pre‐ or postsynaptic plasticity, vesicle release, and neurite growth, namely the GTPases RALA, RAB11A, RAB11B, the signaling protein MARCKS, and the synaptic serine/threonine kinase GSK‐3β (Arendt et al. 2016; Brudvig et al. 2018; Lalli 2009; Lalli and Hall 2005; Sultana and Novotny 2022; Xu et al. 2014; Zempel and Mandelkow 2014). MARCKS and GSK‐3β are both linked to CDC42 (Brudvig et al. 2018; Zhu et al. 2023), a key regulator of postsynaptic f‐actin plasticity that appears upregulated in AD patients (Zhu et al. 2023). MARCKS also controls vesicle docking in the active zone (Xu et al. 2014). The 0N4R interactors RALA and RAB11A/B affect presynaptic function as they interact with the exocyst complex, a key complex of polarized exocytosis in eukaryotic cells (Lalli 2009; Lalli and Hall 2005; Sultana and Novotny 2022). The remaining 0N4R‐specific interactors were either linked to cytoskeletal functions, such as the microtubule acetyltransferase ATAT1 and the f‐actin‐organizing protein ARP2, or to metabolic function (G6PI).

Next, we compared the isoform interaction partners with those of BirA‐Nterm^HA, which does not show axonal enrichment (Figure 6e,f; Tables S10 and S11). The analysis revealed several specific interactors of BirA‐Nterm^HA and a few proteins specifically binding either Tau isoform (Table S3). In the GO term analysis of the Nterm‐specific proteins (Tables S4–S6) the overwhelming number of terms were linked to the cellular compartments peroxisomal membrane and peroxisomal matrix, and to processes including lipid catabolism, fatty acid oxidation, and related peroxisomal processes (Figure 6g). Two proteins of the PTS1‐associated import complex (PEX5, PEX14) were among the interactors. These findings suggest peroxisomal localization of the puncta‐like inclusions of Nterm^HA (Figures 2c and 5b). The mechanism of peroxisomal Nterm^HA import remains unclear since in silico analysis did not predict PTS1‐ or PTS2‐directed import (Brocard and Hartig 2006; Lametschwandtner et al. 1998; Neuberger et al. 2003a, 2003b; Purdue and Lazarow 2001), (PTS1 predictor: https://mendel.imp.ac.at/pts1/↗, PSORTII: https://psort.hgc.jp/form2.html↗).

Out of the proteins enriched for the Tau isoforms (Table S3), the heat‐shock protein HSP110 is known to interact with Tau and the major axonal Tau phosphatase PP2A (Eroglu et al. 2010; Kavanagh et al. 2022). When we inhibited PP2A with okadaic acid, we observed elevated levels of total Tau and AT8‐phosphorylation in the soma (Figure S8e–g), suggesting decreased axonal Tau targeting upon PP2A inhibition. Taken together, the sorting‐specific Tau interaction partners are involved in axon guidance and modulation of the axonal Tau‐targeting phosphatase PP2A, while the data hint at the import and accumulation of BirA‐Nterm^HA inside the peroxisomal lumen.

In our study, we overcame limitations of previous studies (Janke et al. 1999; Kavanagh et al. 2022; Xia et al. 2016) by using rapidly differentiated human‐derived neurons depleted of endogenous Tau and by achieving efficient sorting of overexpressed Tau upon lentiviral delivery. The randomly distributed dTomato, expressed on the same vector but not fused to the Tau constructs due to a 2A cleavage site between the proteins, was used as a reference marker for cellular sorting efficiency.

We could confirm that our Tau^HA constructs were expressed correctly after lentiviral transduction. For all constructs, levels of Tau^HA‐dTomato fusion proteins were negligible, indicating successful cleavage of the 2A peptide, as reported previously (Szymczak‐Workman et al. 2012). Only the heavily truncated constructs only3R^HA, only4R^HA and Cterm^HA were not detected in both assays. Since we could exclude problems with virus production or transduction efficiency due to dTomato co‐expression, the lack of detection results either from rapid degradation, e.g., via proteasomal degradation or recognition by another protein quality control pathway reportedly involved in Tau degradation (Amm et al. 2014; Pohl and Dikic 2019; Ruggiano et al. 2014). Notably, the KXGS‐mutant Tau^HA constructs exhibited a slightly smaller but distinct additional protein species that was more abundant for constructs with more mutated residues. Disturbed interaction with KXGS‐targeting enzymes such as MARK or PP2A (Chudobová and Zempel 2023; Gong et al. 2000) could affect the posttranslational state of adjacent residues. The Tau^HA mutants could also become targets of proteolytic cleavage, which is known to produce similarly sized fragments under pathological conditions (Corsetti et al. 2008; Corsetti et al. 2015; Park and Ferreira 2005).

In our sorting analysis, we observed poor axonal sorting of both individual protein halves, suggesting that both N‐terminal and C‐terminal parts of Tau are required for axonal sorting, or the AIS as main axonal traffic checkpoint (Leterrier 2016, 2018; Rasband 2010) prevents undisturbed axonal transit. For the N‐terminal half, we confirmed findings from human SH‐SY5Y‐derived neurons and mouse primary neurons showing somatic puncta‐like inclusions (Bell et al. 2021). Of note, some constructs (e.g., PRR + 3R‐Tau or no4R‐Tau) showed changes in both axonal and dendritic sorting, hinting at confounding processes independent of axon‐specific targeting. The considerable dendritic Tau enrichment could be caused by maturation deficits or by technical difficulties of single‐dendrite measurements in dense neuronal cultures.

Previous studies with GFP‐coupled Tau demonstrated that Tau lacking the N‐terminal tail failed to interact with the axonal annexins A2 and A6 and was abnormally redistributed to the soma (Gauthier‐Kemper et al. 2018). However, we observed efficient axonal sorting in the absence of the N‐terminal tail. As we determined the net axonal enrichment instead of only retrograde protein flux (Gauthier‐Kemper et al. 2018), the results suggest that anterograde transport can compensate for retrograde retention deficits. Data from KXGS‐mutant Tau in primary rodent neurons (Li et al. 2011; Zempel et al. 2017) corroborate our findings.

When Tau gets missorted in early AD stages, the KXGS and AT8 residues appear highly phosphorylated (Braak and del Tredici 2011; Islam et al. 2025; Regalado‐Reyes et al. 2019; Zheng‐Fischhöfer et al. 1998). These motifs are crucial for MT binding, which hints at a potential role of changed MT affinity in the early missorting cascade. A recent study with PRR2‐mutant Tau claimed that tight MT binding affinity of Tau impairs axonal sorting as PRR2 phosphorylation‐dead Tau failed to enrich axonally (Iwata et al. 2019). Our results with highly diffusible phosphorylation‐mimetic AT8‐ and KXGS mutants, and previous studies in rodent neurons seem to support this notion (Iwata et al. 2019; Zempel et al. 2017). However, the phosphorylation‐dead AT8 and KXGS mutants did not show any sorting deficits in our MAPT‐KO iPSC‐derived neurons. This contradicts the relevance of MT affinity in general. The marginal decrease in bound MTs for PRR2 phosphorylation‐dead mutant Tau cast further doubts in this respect (Iwata et al. 2019). And while depletion of the entire PRR2 domain, thus decreasing its MT affinity (Gustke et al. 1994; Iwata et al. 2019), resulted in poor axonal sorting, depletion of the C‐terminal repeat domains led to less MT binding but no changes in axonal Tau sorting in our and previous studies (Iwata et al. 2019).

Taken together, our results suggest that loss of MT binding affinity, due to hyperphosphorylation of KXGS, AT8, or other motifs, does not cause somatodendritic missorting, as it is observed in early stages of AD. In contrast, changes in phosphorylation and MT affinity may be evoked by cellular responses downstream of the initial missorting trigger.

Our findings give, together with former studies, a clear picture: Highly diffusible Tau can be sorted efficiently (KXGE mutants, 4R‐lacking construct) (Iwata et al. 2019; Zempel et al. 2017) or inefficiently (PRR2‐lacking, our study, (Iwata et al. 2019)), and also MT‐affine Tau can be sorted efficiently (KXGA mutants, AT8 Ser‐ > Glu mutants, our study) or inefficiently (PRR2 Ser‐ > Glu construct) (Iwata et al. 2019). Thus, we postulate that both the C‐terminal repeat domains and the MT affinity of Tau play no key role in the axonal Tau sorting process.

In contrast, certain residues or motifs of the PRR2 domain appear critical for the axonal sorting of Tau as underlined by the Tau sorting failure of PRR2 dephosphorylation‐mimetic Tau mutants (Iwata et al. 2019). The unaffected sorting of all AT8 mutants makes this motif unlikely to be critical. The PRR was shown to contribute to the MT‐binding geometry of Tau by binding the outer tubulin surface (Amos 2004; Kar et al. 2003). These results question the idea that hyperphosphorylation of the AT8 and KXGS motifs, hallmarks of pathological Tau missorting (Zempel and Mandelkow 2014), is causative for the abnormal cellular localization. Our results rather hint at hyperphosphorylation of AT8 and KXGS residues as a consequence of somatodendritic missorting.

The human Tau isoforms differ in their intracellular localization (Bachmann, Bell, et al. 2021; Zempel et al. 2017), and studies in rodent neurons hint at isoform‐specific interactors and functions (Liu et al. 2016). Comparable data are lacking for human Tau, although there is evidence for differential roles of Tau isoforms in health and disease conditions (Buchholz et al. 2025; Buchholz and Zempel 2024).

We aimed to identify specific interactors of the 0N3R‐Tau and 0N4R‐Tau, two isoforms with similar cellular distribution (Bachmann, Bell, et al. 2021; Zempel et al. 2017), in MAPT‐KO iPSC‐derived neurons using TurboID proximity labelling (Branon et al. 2018; Cho et al. 2020). BirA coupling did not impair the axonal sorting efficiency, and the interactome uncoupled Tau and APEX2‐bound Tau was shown to be consistent (Kavanagh et al. 2022; Tracy et al. 2022), indicating the suitability of biotin‐based Tau interactomics. In our study, the protein number was notably smaller than reported for Tau‐APEX2 constructs (Tracy et al. 2022), probably due to our exclusion of background hits found in control constructs. Still, our GO term analysis and the strong hit consistency with previous studies (Kavanagh et al. 2022; Tracy et al. 2022) demonstrated the validity of our data sets (see Table 2 for comparison of our data to previous studies).

We found evidence for differential isoform effects on MTs, namely 0N3R‐Tau interacted with MAP4, which controls MT branching in neurons (Nishida et al. 2023; Tokuraku et al. 2010) while 0N4R‐Tau was associated with the MT acetyltransferase ATAT1 (Iuzzolino et al. 2024; Li and Yang 2015). This suggests that 0N4R‐Tau promotes MT stability while 0N3R‐Tau rather enhances plasticity, in line with the previous claim of labile plasticity being the key benefit of Tau‐MT binding (Qiang et al. 2018). However, overexpression of individual isoforms in SH‐SY5Y cells did not result in changed MT stability or growth (Bachmann, Bell, et al. 2021), but evidence from more relevant human neuronal cell models is missing.

The C‐terminal repeat domains mediate f‐actin interactions of Tau (Correas et al. 1990; Elie et al. 2015). Accordingly, we saw different f‐actin‐related proteins for 0N3R‐Tau (LUZP1, ZO1) and 0N4R‐Tau (ARP2, MARCKS). The 0N4R‐specific signaling protein MARCKS regulates, besides other cellular functions like cell motility, adhesion, or phagocytosis (Huber et al. 2022; Rodriguez Pena et al. 2013), dendrite growth and spine formation via activation of the CDC42 pathway (Brudvig et al. 2018). Pathological CDC42 overactivation, as observed in AD patients, can induce f‐actin breakdown, synapse defects, and Tau hyperphosphorylation via recruitment of the Tau kinase GSK‐3β (Ying et al. 2022; Zhu et al. 2023). Strikingly, we also identified GSK‐3β, a crucial downstream effector of CDC42 (Zhu et al. 2023), as a 0N4R‐specific interactor, although it targets the PRR2 and C‐terminal tail domains of Tau (Zempel and Mandelkow 2014) but not its repeat domains. These findings suggest that 0N4R‐Tau acts on both upstream regulation of CDC42‐mediated postsynaptic function and downstream effects of CDC42‐induced AD‐related spine dysfunction.

0N4R‐specific interactors also act at the presynapse: MARCKS controls synaptic fusion of RAB10‐positive vesicles (Xu et al. 2014). The Ras‐related GTPase RALA controls vesicle trafficking by interacting with the exocyst (Lalli 2009; Lalli and Hall 2005), a key complex for polarized exocytosis in eukaryotic cells (Hsu et al. 1998). In neurons, exocyst‐dependent exocytosis regulates neurite outgrowth, synaptic receptor transport, and neuronal polarity (Lalli 2009; Lalli and Hall 2005; Mehta et al. 2005; Sans et al. 2003; Vega and Hsu 2001). Two more 0N4R‐interactors, RAB11A and RAB11B, interact with multiple exocyst proteins and mediate vesicle tethering, as shown in yeast (Das and Guo 2011; Sultana and Novotny 2022). Of note, RAB11 has implications in AD pathogenesis as it controls the endosomal recycling of beta‐secretase (Li et al. 2012), and reduced RAB11 levels promote BACE‐mediated APP cleavage and accumulation of pathogenic amyloid‐β (Sultana and Novotny 2022; Udayar et al. 2013).

While it is known that Tau binds to presynaptic proteins and contributes to synapse dysfunction in AD (McInnes et al. 2018; Tracy et al. 2022; Zhou et al. 2017), we provide the first evidence that presynaptic interactions are isoform‐specific and that 0N4R‐Tau specifically binds signaling proteins modulating the exocyst‐dependent exocytosis and RAB11 protein function.

The identification of sorting‐specific Tau interactors is crucial to understand the Tau sorting process and to tackle pathological Tau missorting, an early pathological hallmark (Arendt et al. 2016; Guo et al. 2017; Tracy and Gan 2018).

When we compared the interactome of Tau isoforms (0N3R, 0N4R) and non‐sorting Nterm‐Tau, we found numerous Nterm‐specific binding partners associated with peroxisomes, the major organelle for metabolizing fatty acids and other lipids (Eckert and Erdmann 2003; Wanders and Waterham 2006). Interactions between Tau and peroxisomes were not described for both healthy and disease conditions (Kavanagh et al. 2022). When Tau undergoes proteolytic cleavage in disease, fragments similar to our constructs are generated (Guo et al. 2017), but they show mitochondrial (Corsetti et al. 2008; Corsetti et al. 2015) or untargeted localization (Ferreira and Bigio 2011; Park and Ferreira 2005). We found interactions with components of the major peroxisomal import complex (Platta and Erdmann 2007), but the mode of import remains unclear since all required recognition motifs are missing (Brocard and Hartig 2006; Lametschwandtner et al. 1998; Neuberger et al. 2003a, 2003b) and the construct was classified as 'on‐targeting' by two in silico prediction tools.

Out of the proteins that were specifically enriched for the Tau isoforms, the heat shock protein HSP110 is a well‐described interactor of axonal Tau (Kavanagh et al. 2022) and the major Tau phosphatase PPA2 (Arendt et al. 2016; Eroglu et al. 2010). Axonal PP2A exhibits manifold functions, as it is involved in cytoskeletal organization (Bhattacharjee et al. 2022) and promotes axon elongation, guidance, and branching (Lu et al. 2013; Ogura et al. 2010; Zhu et al. 2010). Downregulation of PP2A is associated with AD pathology, as it leads to Tau hyperphosphorylation, enhanced Aβ levels, increased neurodegeneration, and impaired spatial memory in rodents (Martin et al. 2013; Wang et al. 2015). Axonal Tau shows enhanced retrograde diffusion after PP2A loss‐induced hyperphosphorylation (Li et al. 2011). Our findings of somatic Tau accumulation upon treatment with okadaic acid corroborate the role of PP2A in axonal Tau retention.

Interestingly, depletion of HSP110 causes hyperphosphorylation of axonal Tau (Eroglu et al. 2010). Thus, HSP110 acts as a PP2A activator and could be critical for efficient axonal retention of Tau by promoting its MT binding affinity. Whether the binding of axonal Tau to HSP110 influences PP2A regulation, remains elusive.