The present study compared different clinical subtypes associated with PSP tau pathology and evaluated which cytopathologies in which anatomical regions precede others in PSP-RS. We show the following:
Common early vulnerability patterns characterize all PSP clinical subtypes jointly, i.e. affecting mainly the pallido-nigro-luysian axis;
Tau pathology propagates rostrally to neocortical regions and caudally to the cerebellum including the dentate nucleus;
Neuronal tau accumulation is the first step in the early affected pallido–nigro–luysian axis, but in some regions astroglial or oligodendroglial tau pathology precedes neuronal tau pathology;
Significant differences in tau burden, but particularly different tau cytopathologies, distinguish clinical subtypes.
Common early vulnerability but distinct propagation patterns in PSP clinical subtypes
We applied different methods to confirm that the pallido–nigro–luysian axis is the early vulnerable region in PSP subtypes. This has been described in the original description of PSP  and emphasized in the diagnostic criteria  and in a model based on regional tau scores . Importantly, neuronal tau pathology was observed in the locus coeruleus and hippocampus also; however, at least for PSP-RS, we showed that this was independent from the involvement of strategic subcortical and brainstem nuclei, and might represent or overlap with other pathogenic events such as AD or PART or, particularly for the hippocampus and amygdala, with the presence of AGD, which was frequently seen in our cohort (40.2%). On the other hand, observations on distinct subregional distribution of tau pathology in the hippocampus in PSP suggest that there is an AD/PART-independent pathogenic process of the involvement of the hippocampus .
Recently, several neuropathologic studies reported incidental or early stage PSP cases, supporting the importance of the pallido–nigro–luysian axis. Evidente et al.  reported five cases (6.9% of their cohort) with Gallyas-positive pathology fulfilling criteria of PSP but lacking clinical symptoms. They reported that the mean severity scores of Gallyas-positive PSP features were significantly lower in subjects with neuropathological incidental PSP than subjects with clinical PSP, with the subthalamic nucleus and putamen showing the greatest difference . Kovacs et al.  reported five cases (2.1% in their cohort) with PSP pathology in an aging-study (not included in the present study), three of them without clinical symptoms but with tau pathology involving subcortical areas. Dugger et al. (2014) reported four cases (5% of their cohort, three of which were reported by Evidente et al. ). Three of their cases showed neuronal tau in the substantia nigra, two in the subthalamic nucleus and globus pallidus with variable involvement of cortical areas . A single case showed tufted astrocytes in the cortex but no neuronal tau pathology in subcortical areas , which did not fulfil neuropathologic criteria of PSP. Nogami et al.  reported eight cases (2.5% in their consecutive autopsy cohort), which they termed preclinical PSP. All their cases showed neuronal tau pathology in the substantia nigra and either in the globus pallidus (6/8), subthalamic nucleus (4/8), putamen (5/8) or dentate nucleus (7/8) . A single case contained tufted astrocytes in several regions . Yoshida et al.  described 29 PSP cases (2.9%) in a forensic autopsy cohort, 13 of which showed low amount of pathology involving the globus pallidus, subthalamic nucleus, substantia nigra, and pontine nucleus. Many of them had clinical symptoms in spite of not being diagnosed as PSP . Finally, a further case who deceased 2 months after the clinical onset showed neuronal degeneration only in the subthalamic nucleus and substantia nigra with more widespread tau pathology . Importantly, the subthalamic nucleus is predominated by neuronal tau pathology, while other thalamic nuclei are affected by various cytopathologies showing differences between thalamic nuclei .
We theorize that from the common initiating sites in PSP clinical subtypes tau cytopathologies then propagate in different dynamics and patterns. Interestingly, intrinsic connectivity networks, anchored by the dorsal midbrain, whose nodes include the brainstem, basal ganglia, diencephalic, cerebellar and cortical regions have been recently established . These outline a comprehensive architecture of node pairwise connections for this system and show that PSP–related connectivity breakdowns emphasize cortico-subcortical and cortico-brainstem interactions . Several studies addressed the issue of distinct anatomical involvements as a basis for clinical variability, however, focusing on tau burden or neuronal tau pathology [17, 27, 36, 37, 39, 45]. The scoring system developed by Williams et al. considered coiled bodies and thread tau pathology as an important feature . As a novel finding we report here that in addition to differences in overall total tau burden, neuronal, astroglial, and oligodendroglial tau pathologies involve clinical subtypes differently. In particular, neuronal tau differs the least, astroglial tau pathology differs mostly in neocortical areas, while oligodendroglial also in subcortical regions (see Fig. 3). Interestingly, PSP-P, which generally shows a slower disease progression than PSP-RS , differs mostly by the lower degree of glial involvement and particularly of cortical regions. This underpins the importance of glial tau pathology, which might reflect distinct propagation mechanisms of tau or differences in the response to neuronal degeneration . The relevance of astroglial tau pathology has been discussed in distinguishing PSP and pallido–nigro–luysian degeneration . Interestingly, early involvement of astrocytes is a feature of CBD  and has been also reported in brain regions not affected by neuronal tau in Pick’s disease [15, 24]. The discrepancy between neuronal and glial tau pathology in different PSP is intriguing also in the context of divergent patterns of transcriptional associations for neuronal and astroglial tau lesions . Indeed, while neuronal tau pathology positively associated with a brain co-expression network enriched in synaptic and PSP candidate risk genes, astroglial tau pathology positively associated with a microglial gene-enriched immune network . Finally, these observations carry a message for tau neuroimaging since the tracers should be able to detect glial tau also if these distribution patterns are targeted to be recognized. Importantly, PSP-PI showed several differences compared to other subtypes. As reported in other studies , we did not find difference in duration of illness for PSP-PI (except when compared to PSP-P). Early PSP-PI might resemble PSP-RS and accordingly does not show longer disease duration but associates with a significant neuropathological burden.
Proposed sequential patterns of tau pathologies in Richardson syndrome
We recognize sequential distribution patterns that consider the accumulation of different cellular tau pathologies. Based on this and on reports on early stage, incidental, or preclinical PSP [7, 8, 30, 50], and as well as on reports on mapping of scores of tau pathologies , we propose a staging schema for PSP-RS. The conditional probability analysis used here focuses on the accumulation (dichotomized as no/mild versus moderate/severe) of any cellular tau pathology. Thus, we cannot exclude that single tau cytopathologies are not seen in a specific lower step of the sequential distribution in anatomical areas where accumulation is provided for a higher stage. Based on these concepts six sequential steps of tau accumulation can be recognized (Fig. 7). This is translated to six stages for practical neuropathological diagnosis:
Step 1 of the sequence is characterized by the appearance of neuronal tau pathology in the globus pallidus, subthalamic nucleus, and substantia nigra. This vulnerability pattern has been already emphasized in the preliminary diagnostic criteria in 1994  and by Williams et al. , who added that sparse tau pathology might be seen in the motor cortex as well. In this stage oligodendroglial coiled bodies can be observed in the globus pallidus and some degree of astroglial tau accumulation in astrocytes in the striatum. The locus coeruleus and hippocampus may also show neuronal tau pathology; however, this is most likely influenced by other pathogenic processes also and reflects concomitant AD or PART. Further studies should specify their exact contribution.
Step 2 is characterized by accumulation of neuronal tau pathology in the midbrain tegmentum, medulla oblongata and pontine base, and astroglial tau pathology in the striatum. A few tau positive neurons may be seen in the striatum but this is less than the astroglial tau pathology in the striatum; thus the latter is a more consistent feature of this step. Oligodendroglial tau pathology further accumulates in the globus pallidus.
Step 3 is characterized by the accumulation of neuronal tau pathology in the striatum, the dentate nucleus, and the amygdala, the latter influenced by concomitant AGD pathology. Neocortical areas (motor and frontal cortices, together representing frontal lobe) and subthalamic nucleus and thalamic nuclei show increased astroglial and oligodendroglial tau pathology.
Step 4 is characterized by increased neuronal tau pathology in the frontal lobe; astroglial tau accumulates in the amygdala, parietal, and temporal lobe. Oligodendroglial tau accumulates in the striatum and cerebellar white matter.
Step 5 is characterized by accumulation of neuronal tau pathology in the parietal and temporal lobes (involvement of this region is influenced also by concomitant AD/PART pathology). Astroglial tau increases in the occipital cortex and midbrain tegmentum, together with the accumulation of oligodendroglial tau pathology in frontal and parietal lobes and the brainstem (pons base, medulla oblongata, midbrain), as well as in the hippocampus.
Step 6 is characterized by the rare situation that neuronal tau increases in the occipital cortex, as well as astroglial tau pathology accumulates, but never reaches severe degree, in the substantia nigra, globus pallidus, locus coeruleus, and medulla oblongata. Oligodendroglial coiled bodies may further accumulate in the occipital and temporal lobe.
For the practicing neuropathologists a simplified approach for the staging is summarized in Fig. 8. Single cellular tau immunoreactivity, defined arbitrarily as one tau immunoreactive cell in 20 high-power fields (× 40 objective) is not enough to define a stage. Although this study cannot exclude that tau pathology begins in the substantia nigra or the locus coeruleus, these regions might show single tau positive neurons in aging or concomitant AD/PART and, therefore, not included in the staging system. Apart from these aspects, a specific stage can be recognized if mild degree of tau pathology is seen.
To diagnose stage 1 detection (mild/moderate degree) of neuronal tau pathology in the subthalamic nucleus and neuronal and/or oligodendroglial tau pathology in the globus pallidus and/or astroglial tau pathology in the striatum is required. Stage 2 is characterized by prominent tau pathology in these regions with single cellular tau pathologies in the frontal cortex and/or dentate nucleus/cerebellum. Stages 3 and 4 can be diagnosed if astroglial tau pathology accumulates in the frontal cortex and/or neuronal tau in the dentate nucleus and/or oligodendroglial tau pathology in the cerebellar white matter. Due to interindividual variability (i.e., rostral or caudal predominant progression), the dentate nucleus and cerebellar white matter might be discrepant from the frontal cortex and this can be indicated as rostral (i.e. cortical) or caudal (i.e. dentate/cerebellar) predominant. The difference between stages 3 and 4 is based on the amount of tau pathologies in these regions: either in the frontal cortex or in dentate nucleus/cerebellum or in both, tau pathology has to reach moderate or severe degree to allow recognition of stage 4. Single tau-positive astrocytes may be noticed in the occipital cortex. Stages 5 and 6 can be recognized if astroglial tau pathology accumulates (first mild then to moderate/severe degree) in the occipital cortex. This will less likely be seen in caudal predominant forms but will parallel increased amount of tau pathology in all other regions, however, with interindividual variability (Fig. 5).
The rationale to include occipital lobe in the staging is that when it is involved subcortical areas are so heavily affected that they cannot be evaluated to distinguish further stages. Although FDG-Positron Emission Tomography (PET) studies on PSP do not show significant hypometabolism in the occipital lobe , it must be noted that we also observe only relatively mild tau pathology. Furthermore, in this neuropathology staging system we focus on astroglial tau pathology, which might not associate with significant hypometabolism detectable by FDG-PET. Future studies using tau PET imaging are required to see whether this astroglial tau pathology is detectable in tau-imaging. Indeed, a study showing small amount of microscopically detectable tau pathology did not detect alterations in FDG-PET or tau-imaging in the occipital lobe . Supporting our observations, current tau-imaging studies show that the subcortical areas are the primary affected regions in different clinical subtypes ; however, current tau-imaging may show off target binding and they have not been performed in end-stage PSP cases.
This staging requires five blocks to be stained for phospho-tau: (1) a block containing the globus pallidus and putamen, (2) the subthalamic nucleus, (3) frontal cortex, (4) cerebellum with dentate nucleus, and (5) occipital cortex. This staging system overlaps with the scoring strategy developed by Williams et al. , by emphasizing the central involvement of the pallido–nigro–luysian axis, basal ganglia, and dentate nucleus. However, contrasting that study, which focuses only on the accumulation of oligodendroglial coiled bodies and threads, we included astroglial tau pathology in our staging and define cortical areas also as important regions. Finally, we attempted to stage cases of various PSP clinical subtypes and acknowledged the practicality of this staging (online supplemental file Fig. 9). We noted that occasionally the parietal cortex might show more astroglial tau and can be additionally examined to diagnose stage 3. The overall aim of this staging is to be able to recognize early stages without or with only mild degree of clinical symptoms and these cases then can be evaluated as early or preclinical forms to understand earliest pathogenic events. Distinguishing frontal versus dentate/cerebellum predominant stages acknowledges different dynamics of propagation in various clinical subtypes.
Limitations of the study
First, since we examined cases with obvious clinical symptoms and showing considerable tau pathology, our analysis is not able to predict precisely where exactly neuronal tau pathology begins in the brainstem or subcortical nuclei. Second, this model is based on the accumulation and not the presence of a single tau cytopathology; thus the difference between regions includes no/mild pathology compared to moderate/severe. Accordingly, the exact thresholds might need further validation. Third, the number of cases in less frequent clinical subtypes did not allow the application of a conditional probability matrix approach for each subtype. However, due to the common early vulnerability patterns as seen in the heatmaps, we evaluated the staging described for PSP-RS in other clinical subtypes and found that the cases can confidently be included in these stages. Fourth, we used a semiquantitative approach to evaluate tau pathology, which might not be able to distinguish differences within cases of the same scores. However, since image analysis methods are not yet able to distinguish tau cytopathologies and provide data only for total tau load, we used this strategy to assess cell-specific differences. Finally, tract-specific evaluation of white matter pathology was not specifically addressed since we were interested in distribution patterns in major anatomical regions, which are evaluated in the neuropathological practice, and according to this study, reflects well the progression in all clinical subtypes. However, this aspect can be evaluated in further studies to fine-tune subregional distribution patterns, which might better reflect interindividual variability.