From the prion-like propagation hypothesis to therapeutic strategies of anti-tau immunotherapy

Alzheimer's disease (AD) is a genuine challenge for the pharmaceutical industry. The current drugs on the market show only low efficacy, and the clinical trials of the last 20 years have all been negative in phase 3. With the hegemony of the amyloid cascade hypothesis and the focus on amyloid precursor protein (APP) and its metabolites (Aβ and amyloid precursor protein intracellular domain: AICD), therapeutic strategies targeting tau protein, a major component of neurofibrillary tangles and neuropil threads, have emerged only in recent years. However, tau aggregates not only in AD, but also in many other highly heterogeneous pathologies called tauopathies. Understanding tau and tauopathies is essential before designing a therapeutic approach. Tau may be an excellent therapeutic target, but should the strategy be similar among tauopathies? The hypothesis of the amyloid cascade has oriented most of the research towards APP and Aβ; in the tau field, the hypothesis of a prion-like propagation has similarly captured the research effort and oriented the new therapeutic approaches such as immunotherapy into new directions.

In this review, we combined the data obtained in humans and those generated from experimental models in the context of the prion-like propagation hypothesis for tauopathies. We evaluated the current knowledge of tau biology in intra- and extracellular compartments (brain and peripheral fluids), described recent breakthroughs, and highlighted some unanswered questions. Finally, with this knowledge, we wondered whether we could predict the mode of action and the target engagement of immunotherapy approaches among tauopathies.

The advance of immunohistochemistry has revealed that approximately 20 neurodegenerative diseases, named tauopathies, are characterized by the accumulation of hyperphosphorylated tau (pTau) [165]. Tau is encoded by a single gene (MAPT) located on chromosome 17. The MAPT gene is mostly expressed in neurons [27], and due to alternative splicing of exons 2, 3 and 10, six main isoforms are found in the adult brain: 2 – 3 − 10− (0N3R), 2 + 3 − 10− (1N3R), 2 + 3 + 10− (2N3R), 2 − 3 − 10 + (0N4R), 2 + 3 − 10 + (1N4R), and 2 + 3 + 10 + (2N4R) [5, 81] (Fig. 1a).

Tau protein has four domains with unique biochemical characteristics and specific functions: (i) an acidic amino-terminal domain, (ii) a proline-rich region followed by (iii) microtubule-binding regions (MTBR) and (iv) a carboxy-terminal tail. The microtubule-binding regions contain three or four repeat domains (depending on the inclusion of exon 10) [38, 82, 90, 114] (Fig. 1b). Several post-translational modifications (PTMs) have been described on tau proteins [87]; phosphorylation dynamically regulates the physiological functions of tau [133, 160]. In tauopathies, tau is excessively and abnormally phosphorylated [10]. Other PTMs (acetylation, glycation, glycosylation, methylation, SUMOylation, truncation, ubiquitinylation, etc.) have also been described; some of them, such as acetylation, glycosylation and truncation, may also be related to the pathology and are considered as therapeutic targets [100].

pTau may accumulate in the cell bodies of neurons without forming fibrillary aggregates—a change called a “pre-tangle”. Moreover, pTau may aggregate in the cell bodies of neurons (neurofibrillary tangles = NFTs) or in the cell processes (neuropil threads = NT), and NT may be axonal (as in the corona of the senile plaque) or dendritic. Under electron microscopy, tau aggregates are principally made of paired helical filaments (PHF) in AD (3R and 4R), which is also the case in ‘primary age related tauopathy’ (PART) in chronic traumatic encephalopathy (CTE) and in some less-common disorders (see Fig. 2). In progressive supranuclear palsy (PSP) and cortico-basal degeneration (CBD), tau aggregates are found both in neurons and glia and are made of 4R straight tau filaments. Other specific neuronal tau inclusions are Pick bodies (seen in Pick disease), in which 3R tau aggregates in the neuronal cell body adopt a spherical shape, and argyrophilic grains made of 4R tau (seen in argyrophilic grain disease (AGD)) are located in presynaptic terminals. The glial inclusions may involve astrocytes: in astrocytic tufts, suggestive of PSP, all processes of the astrocyte are filled with pTau; in astrocytic plaques, seen in CBD, pTau immunoreactivity is found at the end of the astrocytic processes [108, 110]; and they may involve oligodendrocytes where they form coiled bodies (abundant in PSP, CBD and AGD). The ratio between 3 and 4R tau explains why specific sets of migration bands have been recognized by Western blotting (Fig. 2). Recent structural studies by cryo-electron microscopy have confirmed the presence of different tau structures among AD, Pick’s disease and CTE [67, 68, 71].

Genetic frontotemporal lobar degeneration (FTLD-MAPT formerly called FTDP-17, [74]) is of particular interest. It is caused by autosomal dominant MAPT mutations mainly located in the sequence regulating exon 10 alternative splicing or encoding microtubule-binding regions. Most coding region mutations often act on both nucleation and fibrillogenesis [12, 36]. They are widely used for studying the pathological mechanisms of tauopathies, especially in animal models [61].

Altogether, tauopathies are the consequence of several molecular dysfunctions likely due to PTM and protein-folding deregulations. These deregulations are associated with the development of conformational changes, oligomerization and finally, intracytoplasmic aggregation of tau. All these processes have a significant impact on cell physiology and particularly on tau functions [164] (Fig. 3). Neuropathology, however, shows the large variety of cerebral areas and cell populations (glial or neuronal) affected; it also indicates that the type of tau aggregation is diverse, as shown by the multiplicity of tau inclusions and their molecular composition.

Tauopathies and other neurodegenerative disorders are reminiscent of prion diseases. The latter are indeed related to a pathological change (misfolding) in the molecular conformation of the normally present prion protein and to the capacity of misfolded prions to induce misfolding by contact with a normally folded prion. Interestingly, different strains of prion exist with dissimilar fibre conformations that are transmissible [175]. There are at least two steps in the development of prion diseases: (1) misfolding induction and (2) transport of the misfolded prion that permits contact with still normal prions. Step 1 is called seeding, and step 2 is propagation. Propagation can take place by diffusion in the extracellular space or along the axonal paths.

Seeding and propagation have been reported not only with prions, but also with Aβ [102] and alpha-synuclein [97]. The term “prion-like” has been attributed to this molecular behaviour when applied to proteins other than prions; such proteins have been called “propagons” and may induce protein misfolding in test-tubes, tissues, organisms or between subjects (infectious). Although there is no clear evidence of such interindividual transmission for tau, in several tauopathies, tau pathology progresses by well-defined “stages”, which may support prion-like propagation. Here, the term “prion-like propagation hypothesis” includes the two steps: seeding and propagation.

Data from humans allowed the development of experimental models to assess whether the ‘prion-like’ propagation hypothesis is implicated in tauopathies and especially whether extracellular tau is relevant to this hypothesis. These models were thus designed to investigate the prion paradigm in regard to human neuropathology. For instance, with the upregulation of Rab7A in AD patients [78‐80], tau secretion seems to be regulated by this small GTPase involved in the trafficking of endosomes, autophagosomes, and lysosomes [149]. Similarly, tau secretion is correlated with Golgi dynamics [128], consistent with its fragmentation noted in AD [167]. This hypothesis involves a multi-step mechanism (tau secretion, uptake and subsequent seeding processes) that has been widely explored.

Historically, tau therapeutic interventions were designed to target intraneuronal mechanisms such as modulating PTMs, breaking tau aggregates or decreasing tau concentrations [100]. Many of them have already failed and a global tau silencing may have side-effects due to multiple tau functions [100, 164]. More recently, tau immunotherapy showed promising preclinical results. While different mechanisms, such as microglial activation and the generation of different anti-tau antibodies, are involved in vaccination, we focus here on mechanisms involved in passive immunotherapy that have been widely explored in different experimental models [2, 7, 17, 33‐35, 47, 50, 52‐54, 155, 176, 178, 179, 194]. Early studies have shown that tau vaccination and tau immunotherapy could reduce intraneuronal tau pathology. Most of them decreased the amount of insoluble tau materials [7, 176, 194]. Cognitive decline was also reduced in some [176, 194], including those using MC1, an antibody recognizing a conformational epitope involving the amino terminus of tau [101] and HJ8.5, an antibody recognizing the amino acids 25–30. Both have been further developed for human immunotherapy and are now in clinical trials (LY3303560, ABBV 8E12-Table 2). Peripheral administration of this MC1 antibody in murine models of tauopathy has not only reduced tau pathology quantified by biochemical and immunohistochemical analyses but also delayed the onset of decreased motor impairment and weight loss [35]. Such experiments were also performed with a phospho-dependent antibody (anti-pS422, also further developed for human immunotherapy, RG7345-Table 2) showing similar effects with the involvement of the endosome-lysosome pathway [47]. The finding of tau secretion and its compatibility with the prion-like propagation hypothesis have modified the strategy of immune therapy: research was then oriented towards antibodies that did not penetrate the cells and principally acted in the extracellular space.

The Fc part of immunoglobulins G (IgGs) binds to specific Fc gamma receptors (FcγR), which are variously expressed in neurons, microglia or astrocytes. Different classes of FcγR have been described with high affinity (binding to monomeric IgG complexes) or low affinity (binding only to multimeric IgG complexes) to different Fc. Microglial cells highly express all classes of FcγR [4], whereas only FcγRI has been reported in astrocytes. In neurons, the expression of FcγR is still debated [136], but neurons [69] and microglia [4] seem to upregulate FcγR in response to extracellular IgG with a functional Fc domain.

The different isotypes of human IgG (IgG1, IgG2, IgG3 and IgG4) have, on the other hand, different affinities for the receptors. For example, human IgG1 has a higher affinity for activating FcγRs present on microglia compared to human IgG4 and may induce a more pro-inflammatory response [26]. IgG4 may thus theoretically bind to its antigen, lowering its toxicity, while at the same time, limiting its cell internalization through Fc receptors and the potential inflammation, it may induce [111]. Most tau antibodies currently being tested in clinical trials use this isotype to target extracellular tau (Table 2). Nevertheless, most of the preliminary studies are performed in mice, and it is often difficult to translate these data to humans. In mice, four classes are also found, but they do not correspond to human IgG. In fact, murine IgG1 is closer to human IgG4 than murine IgG2. Thus, when murine antibodies and their effector functions are tested in animal models, it is always difficult to draw conclusions. For instance, Funk and collaborators showed that the murine version of ABBV 8E12 (HJ8.5 IgG2a) drives uptake of tau species into BV2 microglial cells [76]. However, single-chain HJ8.5 antibodies without Fc fragment (scFvs) significantly reduced levels of hyperphosphorylated, aggregated tau in brain tissue of tau transgenic mice [98]. Such data suggest that the Fc fragment may not be required for the efficacy of this family of anti-tau antibodies in tau immunotherapy. The use of different approaches and different models may thus sometimes lead to different conclusions on the impact of Fc fragment and receptor.

The most widely accepted scenario is that therapeutic antibodies bind extracellular pathological tau species in the ISF responsible for the spread of pathology but are not necessarily required to directly bind intraneuronal tau. The antibodies would block the seeding activity of extracellular tau, most likely by blocking the initial uptake of these seeds and slowing down the spread of pathology [134]. Nevertheless, the future of such anti-tau antibody-tau complexes and their clearance are still puzzling, and many hypotheses have been proposed, as summarized in Fig. 6.

Therapeutic antibodies currently in clinical development have shown their capability to target extracellular tau (by modulating cell-to-cell tau transfer) and/or to reduce tau seeding in experimental models. The murine version of BIIB076 has been shown to block neuronal uptake of human tau species in microfluidic chambers [134]. BIIB076 was described to be more efficient in blocking tau uptake than an antibody directed against the C-terminus of tau protein [134]. The murine version of UCB0107 (binding a region before the first MTBR) displayed superior efficacy in an in vitro model of seeding and aggregation induced by human AD tau than antibodies targeting N-terminal, pS202/pT205, or pS422 conformational epitopes [49]. It also effectively prevented the induction of tau pathology in the brain of transgenic mice injected with human AD brain extracts, in contrast to an amino-terminal tau antibody recognizing the same epitope as BIIB092 [2]. Other antibodies have also been tested by Vandermeeren and collaborators from Janssen [179]. Immunodepletion assays strongly suggest that difference exist between epitopes. In fact, antibodies that recognized the mild-region of tau are most effective while N-terminal antibodies could not fully block seeding in vitro (FRET assays) as well as in vivo (AD brain material enriched for PHFs injected into the hippocampus of P301L transgenic tau mouse to spur fibrillization immunized by various antibodies). Together, these results suggest that the murine version of JNJ-63733657 which is targeting the pS217, a central part of tau may also act in this way [107, 150]. Extracellular tau may also lead to neuronal hyperactivity and facilitate Aß peptide secretion; in this paradigm, it was reported that the murine version of BIIB092, IPN002, is protective by blocking extracellular tau [25].

Finally, all these antibodies have shown some effects in mouse models, although few reports are available in humans. The next challenge is to validate the target engagement and efficacy of drug candidates.

Some studies on transgenic models suggest that plasma tau is a potential biomarker for therapeutic monitoring. Indeed, the active immunization of tau transgenic mice with peptides containing a pathological phosphorylated epitope (pS422) reduced pathological tau species in the brain and delayed cognitive deficits but was importantly associated with a significant increase in plasma tau concentrations [176]. Similarly, it was reported at the 2017 AD/PD conference that injection of RO7105705 in tau transgenic mice (13 weeks treatment/3–30 mg/kg) increased plasma tau [8]. This concept of tau/Ab complex stabilization in the plasma has also been validated by another study demonstrating that peripheral administration of the anti-tau antibody HJ8.5 (ABBV 8E12) increased plasma tau not only in transgenic mice, but also in PSP patients [195]. Such stabilization in plasma may be explained by high plasma concentrations of therapeutic antibodies. For instance, at the 2017 AD/PD and AAIC conferences, it was shown that BIIB076 concentrations were 1000-fold higher in plasma than in CSF when a single injection was administered in blood in cynomolgus monkeys [84]. Tau analysis is also performed in CSF in clinical trials and may help to show target engagement. For instance, the BIIB092 antibody in a phase 1b clinical trial in PSP is well tolerated, and its complex with tau is found in CSF [18]. In fact, most of the antibodies described in Table 2 are in clinical trials, and their outcome should be published in the months to come. Nevertheless, we know that the ABBV 8E12 antibody has already been discontinued in a phase 2 clinical trial in PSP. However, it is still ongoing in AD clinical trials. BIIB092, RO7105705 and ABBV 8E12 target the amino terminal domain of tau proteins, whereas UCB0107 recognizes an epitope in the mid-region before the repeats. Only JNJ-63733657 [107] and LY3303560B [35] target so-called pathological intracellular epitopes. Knowing the molecular diversity of tau species among tauopathies, do we have the right tool for the right disease? In fact, the main question is to define which species have seeding- and propagative-competent properties.

Since it is now acknowledged that extracellular tau (eTau) is a therapeutic target, the next generation of antibodies has to take into account the heterogeneity of tau species. This heterogeneity exists at the level of protein composition (tau sequence and PTMs), compartment (free or vesicular) and seeding competency. In fact, high heterogeneity of aggregated tau species is encountered in the human brain among tauopathies, and some of them may be released in ISF/CSF. As revealed by 2D electrophoresis studies, the tau proteome in CSF is at least as complex as that in the AD autopsy brain [88]. The pattern is composed of tau proteolytic fragments [16, 162, 199]. Barthelemy and collaborators [13, 14] identified peptides in the CSF of AD where they are abundant but also in the CSF of PSP, Lewy body dementia (LBD) and control groups where tau concentrations are very low. Among the identified peptides, those with the central part of tau (amino acids 126–234) were the most abundant ones, 4R-specific MTBR peptides were below the lower limit of detection, and the 2 N peptides were poorly detected, which might suggest that the 1N3R isoform is the most abundant isoform in CSF. When comparing pathologies, the central peptides (amino acids 126–234) were more abundant in the AD group than in the control, PSP and LBD groups. Nonetheless, no difference was observed between the PSP and LBD groups. The same observation was made by Cicognola and collaborators who described an important cleavage site at amino acid 224, especially in CSF from AD in comparison to MCI-AD, PSP and CBS [43].

The characterization of the species and nature of human tau species with seeding properties is a crucial, unanswered question. Are these seeds hypo- or hyperphosphorylated, truncated, or fibrillar? Recent studies looking precisely at seeding-competent tau from human brains with different tauopathies confirmed this molecular diversity in in vitro and in vivo models [44, 104, 105, 132, 154]. As described earlier, tauopathies are heterogeneous and there are some differences among tauopathies but we have to be aware that it is also true within a tauopathy. For instance, very little is known about the case-by-case molecular heterogeneity among a specific tauopathy such as AD and how such heterogeneity would affect the underlying pathophysiological mechanisms such as tau propagation and seeding. In this regard, Brad Hyman group described huge differences among seed-competency of human AD brain inoculates [62]. It does not mean that seeding models are bad, it only means that tau seeds in brain inoculates may be different among samples. It is not surprizing since the clinical course, the disease duration, the postmortem delay and maybe the storage conditions are different. It is important to be aware of this when seeding models are used to test and validate immunological tools. This last section highlights heterogeneity among tauopathies and strongly suggests that each disease may have its propagons. Thus, an anti-tau immunotherapy strategy common to all tauopathies is therefore unlikely.

If we want to summarize our knowledge, we can say that among tauopathies, the work of neuropathologists has made it possible to define specific neurodegenerative pathways suggesting the existence of vulnerable neuronal subpopulations for each disorder. These hierarchical pathways of tau pathology led to the hypothesis of prion-like propagation, which is based on human neuropathology and numerous experimental models and has led to the identification of a new player in tauopathies: eTau. This hypothesis has led to a reorientation of anti-tau therapeutic approaches, particularly immunotherapy. The main remaining questions are around molecular species responsible for propagation or seeding and their targeting by our therapeutic strategies. Do we have the right immunological tools?

However, we should not forget that the prion-like tau propagation hypothesis is still an assumption that is not unanimous. There are first very heterogeneous tau aggregates at the molecular and cellular level among tauopathies. The diversity of neuronal populations and their connectivity are probably an explanation for differences among tauopathies. In any case, as we have seen, the arguments that support the prion-like propagation hypothesis in humans are particularly well documented in AD: the steadiness of Braak stages in AD patients and the high and constant concentration of eTau during AD pathology compared to other tauopathies. Thus, based on our current knowledge, AD is the best target for anti-eTau immunotherapy. With a remaining question, do we have the right approach since the prion-like propagation is still a hypothesis, like the amyloid cascade…