To investigate the burden and spatiotemporal distribution of AD-related abnormal tau isoforms (Fig. 1a) in the retina, we prepared retinal cross-sections from the ST and IT regions in a cohort of patients with a premortem diagnosis of AD dementia (n = 34, mean age 83.88 ± 12.67 years, 21 females/13 males) or MCI (due to AD; n = 11, mean age 89.55 ± 6.12 years, 7 females/4 males), with postmortem neuropathological confirmation of AD, along with individuals with NC (n = 30, mean age 81.43 ± 9.82 years, 16 females/14 males). No significant differences were found in age, sex, or post-mortem interval (PMI) among the three diagnostic groups. Demographic information, brain neuropathology, and retinal tauopathy histological quantifications are detailed in Tables 1 and 2. The correlations of various retinal tau isoforms with brain neuropathology and cognitive scores are summarized in Table 3. As an additional control group for AD dementia, we also analyzed several tau isoforms in postmortem retinas of patients with non-AD dementia (n = 4), including DLB, FTLD with ALS, FTLD-tauopathy with Pick’s disease, and FTLD-tauopathy with PSP (see Methods above and Suppl. Table 1).

Fig. 1

MC-1-positive tau tangles in the retina of MCI and AD patients. a Illustration of the various abnormal tau isoforms. b, c Representative images of immunofluorescent staining for MC-1+ tau tangles (red), neuronal marker βIII-tubulin (green), and DAPI (blue) in retinal cross-sections from individuals with normal cognition (NC), as well as patients with MCI and AD dementia. Scale bars 20 µm. Additional inserts of high-magnification images show intra-neuronal MC-1+ tau tangles in the MCI and AD retinas (scale bars 10 µm). d Representative images of cortical (A9) stained for MC1+ tau tangles using immunofluorescence (red) or peroxidase-based DAB (brown) labeling. Scale bars 20 µm. e Gallyas silver staining (Scale bars 20 µm) and/or Bielschowsky silver staining (Scale bars 10 µm) in cortical and retinal sections from AD patients. f Quantitative assessment of brain NFT severity scores in patients with MCI, AD, and NC controls (n = 49 in total). g Quantitative analysis of the percentage of retinal MC-1+ immunoreactive area (n = 18 NC, n = 9 MCI, and n = 21 AD). h Spearman's rank correlation coefficient analysis between retinal MC-1% area and brain atrophy severity score. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, ****P < 0.0001, by one-way ANOVA with Sidak’s post hoc multiple comparison tests. Fold changes are shown in red. F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

Table 2 Retinal tauopathy quantification data on human donors in this studyTable 3 Spearman’s correlation analysis between retinal tauopathy markers, brain neuropathology and cognitive declineIncreased MC-1-positive tau tangles in the MCI and AD retina.

To explore the presence and burden of tau tangles in the AD retina, we utilized anti-MC-1 immunostaining and Bielschowsky silver staining in retinal cross-sections from a subset of donors with MCI (n = 9, mean age 89.7 ± 5.1 years, 5 females/4 males), AD (n = 21, mean age 86.1 ± 8.4 years, 11 females/10 males), and NC controls (n = 18, mean age 84.5 ± 9.3 years, 9 females/9 males), comparing them to brain NFTs (Fig. 1, extended data in Suppl. Fig. 1). Specifically, we conducted IHC staining with the MC-1 monoclonal antibody, a tau tangle conformational- and sequence-specific antibody, which identifies the amino acid (aa) 7‐9 and 312‐322 tau epitopes [40]. MC-1 primarily recognizes pre-tangles and mature tangles [57]. We mostly found pretangles and diffuse-type MC-1+ signal in the human MCI and AD retina. Occasionally, we identified retinal MC-1+ mature tau tangles, indicative of the paperclip folding of tau (Fig. 1b, c), with similar structures to NFTs in the AD brain (Fig. 1d, e). These retinal tau tangles were found in ganglion cells and βIII-tubulin+ cells within the INL of MCI and AD patients. Retinal tangles, as revealed by MC-1 and Bielschowsky silver staining, also appear as paperclip tau formations resembling NFTs (Fig. 1b,and e, right panel), as observed in the AD brain by Gallyas and Bielschowsky silver staining (Fig. 1d, ande, left panel; see additional Bielschowsky silver stain images across retinal layers and in paired brain tissues from other AD patients in Suppl. Fig. 1). Histological analysis of brain NFTs measured by Gallyas Silver and Thioflavin staining shows a significant 3.2-fold higher level in MCI and a 4.1-fold higher level in AD, compared to NC controls (Fig. 1f). Quantification of percent MC-1+ immunoreactive area in the respective retinas indicated significant and modest 2.0-fold and 1.8-fold increases in MCI and AD patients compared to NC controls, respectively (Fig. 1g). There was considerable overlap in the levels of retinal MC-1 percent area in aged individuals with normal cognition and those with MCI due to AD and AD dementia. Spearman’s rank correlation coefficient analysis demonstrated that retinal MC-1+ burden weakly associates with the severity of brain atrophy (Fig. 1h), as well as with NFTs, NTs, and cerebral amyloid angiopathy (CAA, Table 3).

Identification of retinal tau oligomers with increases in MCI and AD patients linked to brain pathology, Braak staging, and cognitive status

We next explored Oligo-tau forms in the retina of MCI and AD patients. In AD brains, toxic Oligo-tau forms are assembled from small p-tau aggregates after dislodging from microtubules in neurons and are shown to propagate from affected to unaffected brain regions [49, 64, 72]. Previous research demonstrated that extracting Oligo-tau from the AD brain using the T22 Ab and injecting these oligomers into wild-type mouse brains caused neurotoxicity and the propagation of abnormal endogenous murine tau [48]. Here, we performed anti-T22 immunolabeling in retinal and brain sections from a donor cohort comprising MCI (n = 10, mean age 88.7 ± 5.7 years, 6 females/4 males), AD (n = 21, mean age 86.1 ± 8.4 years, 11 females/10 males), and NC controls (n = 19, mean age 84.5 ± 9.1 years, 9 females/10 males). Additionally, we analyzed retinas from four cases of non-AD dementia (D-NAD; Fig. 2, extended data in Suppl. Fig. 2a). Compared with NC control retinas, we identified intense and diffuse-like T22+ Oligo-tau signals in the AD and MCI retinas, when labeled in combination with the pre-synaptic marker, vesicular glutamate transporter 1 (VGLUT1) and DAPI (blue) for nuclei (Fig. 2a, upper panel), or with DAPI (Fig. 2a, lower panel). In the retinas of MCI and AD patients compared to NC controls, we observed abundant cellular and diffused Oligo-tau, especially in the synaptic-rich layers (OPL, IPL), alongside scarce VGLUT1+ signals (Fig. 2a, upper panel). In comparison to retinas from AD dementia patients, retinal T22+ signals appeared fewer in D-NAD patients (Fig. 2a, lower panel, and Suppl. Fig. 2a). In the brain, the differences in T22+ Oligo-tau burden between AD and NC are also evident (Fig. 2b). Quantitative IHC analysis revealed a highly significant 9.2-fold increase in retinal T22+ Oligo-tau in AD patients and a significant 5.2-fold increase in MCI patients compared to NC controls. Retinal T22+ Oligo-tau burden in AD dementia patients was significantly 1.8 times higher compared to MCI patients and 2.9 times higher compared to D-NAD patients (Fig. 2c). There was a trend of elevated retinal Oligo-tau burden in D-NAD patients versus NC controls, which reached statistical significance by two-group analysis Student t-test. Notably, Spearman’s correlation analysis indicated that retinal T22+ Oligo-tau strongly and positively correlates with brain NFTs burden (Fig. 2d, rS = 0.66, P < 0.0001), Braak staging—a parameter of tauopathy spread across brain regions during AD progression (Fig. 2e, rS = 0.71, P < 0.0001), CAA severity (rS = 0.69, P < 0.0001), and the A(amyloid-beta plaque) B(NFT stage) C(Neuritic plaque)—ABC scores (Table 3). Retinal Oligo-tau moderately correlates with brain Aβ burden and NTs. Furthermore, moderate to strong correlations were found with the MMSE and the CDR cognitive scores (Fig. 2f, Table 3; rS = − 0.57, P < 0.001 and rS = 0.63, P < 0.0001, respectively).

Fig. 2

Identification of oligomeric tau in the retina of MCI, AD, and non-AD dementia patients. a Representative images of immunofluorescent staining for T22+ oligomeric tau (Oligo-tau, red), vesicular glutamate transporter 1 (VGLUT1, green), and nuclei (DAPI, blue) in retinal cross-sections from NC, MCI, and AD; representative images from patients with non-AD dementia (D-NAD), with frontotemporal lobar dementia (FTLD) with either progressive supranuclear palsy (PSP) or Pick’s disease, are also shown. Dashed lines demarcate the area of the quantitative IHC analysis, between the inner limiting membrane (ILM) and the outer limiting membrane (OLM). b Representative images of T22+ Oligo-tau immunofluorescence (red) and DAPI (blue) in cortical (A9) sections from NC and AD patients. Scale bars 20 µm. c Quantitative IHC analysis of the percent retinal T22+ Oligo-tau immunoreactive area (n = 19 NC, n = 10 MCI, n = 21 AD, and n = 4 D-NAD patients). Spearman's rank correlation coefficient analyses of retinal T22+ Oligo-tau against d brain NFTs severity score, e Braak stages, and f CDR scores. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, ***P < 0.001, ****P < 0.0001, by one-way ANOVA and Tukey’s post-hoc multiple comparison test, or unpaired 2-tailed Student’s t test (in parenthesis). Fold changes are shown in red. D-NAD, non-AD dementia; F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

GeoMx profiling of total tau and p-tau isoforms in the retina and brain of MCI and AD patients.

We employed the high-throughput NanoString GeoMx® digital spatial profiling (DSP) technique (Fig. 3a) to determine quantities of total tau protein and various p-tau forms in retinal cross-sections and corresponding brain cortical sections prepared from a donor cohort comprising MCI (retina: n = 6, mean age 88.5 ± 5.0 years, 3 females/3 males; brain: n = 4, mean age 87.5 ± 5.8 years, 3 females/1 male), AD (retina: n = 9, mean age 85.1 ± 7.8 years, 5 females/4 males; brain: n = 4, mean age 86.75 ± 4.3 years, 3 females/1 male), and NC controls (retina: n = 9, mean age 89.3 ± 9.4 years, 6 females/3 males; brain: n = 5, mean age 90.4 ± 7.3 years, 3 females/2 males). The list of individual patients is detailed in Supplementary Table 1.

Fig. 3

Tau isoforms quantified by GeoMx® digital spatial profiling in retinas and brains from MCI and AD patients. a Graphical illustration of NanoString GeoMx® digital spatial profiling (DSP) analyses for tau protein and tau isoforms in retinal and respective brain samples. b Quantitative analysis of retinal total tau and p-tau at phosphorylation sites of S199, S214, S396, S404, and T231 detected by GeoMx® in retinas from AD (n = 9) and MCI (n = 6) patients, and NC controls (n = 9), and paired-brain tissues (frontal cortex region A9; n = 13 in total). Spearman’s rank correlation coefficient analyses are shown between retinal pT231-tau and the severity of brain c Aβ plaques, d CAA, e ABC, f NFTs, and g NTs scores. Individual data points (circles) and median, lower and upper quartile are shown in violin plots; n, sample size. *P < 0.05, **P < 0.01, by one-way ANOVA and Sidak’s post-hoc multiple comparison test for group analyses. Two group comparisons by unpaired 2-tailed Student t test are indicated in parenthesis. Fold changes are shown in red. Illustrations created with Biorender.com

The GeoMx tau module included the analysis of total tau and p-tau isoforms at sites of serine 199 (S199), serine 214 (S214), serine 396 (S396), serine 404 (S404), and threonine 231 (T231) (Fig. 3b). In the AD retinas and corresponding brains, there were trends of higher total tau levels compared to NC controls, which reached statistical significance for the AD brain by Student t test. While no difference was detected in total tau levels in the MCI brains versus NC controls, the total tau levels in the MCI retinas were 1.8-fold higher, reaching statistical significance by Student t test (Fig. 3b; top left). Significant increases in brain pS214- (21.5-fold), pS396- (29.2-fold), and pS404- (twofold) tau forms were found in AD patients compared to NC controls, while brain p-tau forms at sites S199 and T231 showed a non-significant trend of increases in the AD patients. For the MCI brain, only pS214- and pS396-tau forms showed a non-significant trend of increases compared to NC controls, and the other forms showed no differences.

In the retina, significant increases in retinal S214 (2.3-fold), S396 (2.6-fold), S404 (2.4-fold), and T231 (1.8-fold) p-tau forms were detected in MCI patients compared to NC controls. Retinal pT231-tau was significantly elevated (1.6-fold) in AD patients compared to NC controls (Fig. 3b). Interestingly, the levels of retinal pS396- and pS404-tau were significantly higher in MCI versus AD patients. Furthermore, our quantitative GeoMx analysis in this cohort indicated that retinal pT231-tau significantly and weakly correlated with brain Aβ plaques and moderately correlated with CAA, ABC, brain NFTs, and brain NTs severity scores (Fig. 3c–g).

Histological evaluation of total tau and p-tau isoforms in the retina of MCI and AD patients

The detection of significant changes in retinal and brain tau isoforms in GeoMx DSP analysis prompted an additional histological examination of total tau and other retinal p-tau forms at epitopes T202/S214 (AT100+), S202/T205 (AT8+), and S396 (pS396+), in larger cohorts (Figs. 4 and 5; extended data in Suppl. 2–4). IHC analysis of total tau using both HT7 and 43D antibodies revealed a considerable retinal tau signal, mostly restricted to the OPL in NC subjects, and across all retinal layers in MCI and AD patients (Fig. 4a, upper and middle panels). Analysis of retinal AT100+ p-tau forms was performed in a subset cohort consisting of AD (n = 6, mean age 80.67 ± 14.73 years, all females), MCI (n = 4, mean age 92.75 ± 4.99 years, all females), and NC controls (n = 9, mean age 81.22 ± 12.20 years, 8 females/1 male). Our histological analysis showed retinal AT100-positive immunoreactivity in the innermost retinal layers, including nerve fiber layer (NFL) and GCL (Fig. 4b), and a non-significant trend of increases in MCI and AD patients compared to NC controls (Fig. 4c). Retinal AT100+ p-tau forms were strongly associated with the NFTs burden (Fig. 4d, Table 3).

Fig. 4

Total tau, AT100-positive, and AT8-positive p-tau isoforms detected by histological examination of retinas from MCI and AD patients. a Representative images of immunofluorescent staining for total tau (HT7+ and 43D+) and AT8+ p-tau (S202/T205) in retinal cross-sections from human donors with NC, MCI, and AD. Scale bars 20 µm. Additional inserts of high-magnification images show pronounced intra-neuronal 43D+ tau and AT8+ p-tau staining in the AD retina. b Representative images of peroxidase-based DAB staining of retinal AT100+ p-tau (T202/S214) in an AD patient; a high magnification image for the ganglion cell layer (GCL). c Quantitative analysis of percent retinal AT100+ p-tau immunoreactive area (n = 9 NC, n = 4 MCI, and n = 6 AD). d Spearman's rank correlation coefficient analyses between retinal AT100+ p-tau and brain NFTs severity scores. e Representative images of peroxidase-based DAB staining of AT8+ p-tau (S202/T205) in retinal cross-sections from AD patients. f Quantitative analysis of retinal AT8+ p-tau immunoreactive area (n = 18 NC, n = 8 MCI, n = 21 AD, and n = 4 D-NAD). g Mapping of AT8+ p-tau in central (Cen), Mid-, and Far-peripheral retinal subregions in the same cohort. h Spearman's rank correlation coefficient analysis between retinal AT8+ p-tau and the cognitive status as assessed by CDR scores. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, **P < 0.01, by one-way ANOVA and Tukey’s or Sidak’s multiple comparison tests, or unpaired 2-tailed Student t test for two group analysis (in parenthesis). Fold changes are shown in red. F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

Fig. 5

Retinal pS396-tau in MCI and AD patients. a–c Representative images of immunofluorescent and peroxidase-based staining of pS396-tau in retinal cross-sections from AD and MCI patients as compared to NC controls. Scale bars 20 µm. Higher magnification images show intracellular pS396-tau in the INL and ONL (b red arrowheads and c white arrowheads). d Quantitative analysis of the percent pS396-tau immunoreactive area in the retina (n = 15 NC, n = 9 MCI, n = 25 AD, and n = 4 D-NAD). e Percent pS396-tau area separated to central (Cen), Mid-, and Far-peripheral retinal subregions in the same cohort. Spearman's rank correlation coefficient analyses are shown between retinal pS396-tau and the severity of brain f Aβ plaques, g NFTs, h NTs, and i ABC scores. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, **P < 0.01, ***P < 0.001, by one-way ANOVA and Tukey’s post-hoc multiple comparison test, or unpaired 2-tailed Student t test for two group analysis (in parenthesis). Fold changes are shown in red. F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

We next examined retinal AT8+ p-tau signals (Fig. 4a, lower panel and Fig. 4e) in a subset of donors with MCI (n = 8, mean age 89.75 ± 5.50 years, 4 females/4 males), AD (n = 21, mean age 82.81 ± 13.40 years, 10 females/11 males), and NC controls (n = 18, mean age 81.50 ± 8.96 years, 9 females/9 males). AT8+ pS202/T205-tau isoforms were frequently detected in the OPL, and to a lesser extent, in the IPL (Fig. 4a, e); a staining pattern that is comparable with previous reports in the AD retina [34, 87]. Quantitative IHC analysis revealed a 3.5-fold trend of increases in MCI and a 2.9-fold in AD, compared to NC controls (Fig. 4f), reaching significance by Student’s t test. Retinal AT8+ p-tau burden in D-NAD patients was at similar levels as those observed in MCI and AD patients and had a trend of a 2.7-fold increase compared to NC controls (Fig. 4f; representative images in Suppl. Fig. 2b). Examination of AT8+ signals in the central, mid-, and far-peripheral retina indicated significant increases in MCI compared to NC in all three retinal subregions (Fig. 4g). Retinal AT8+ p-tau forms significantly and weakly associated with brain NFTs, ABC, and CAA severity scores, while showing a moderate correlation with the CDR cognitive scores (Fig. 4h, Table 3).

Analysis of retinal pS396+ tau was performed in a cohort consisting of donors with MCI (n = 9, mean age 89.67 ± 5.15 years, 5 females/4 males), AD (n = 25, mean age 86.80 ± 8.25 years, 16 females/9 males), and NC controls (n = 15, mean age 84.33 ± 8.96 years, 6 females/9 males). Using immunofluorescence and peroxidase-based immunostaining, we found substantial pS396-tau depositions across all retinal layers in MCI and AD patients compared to NC controls (Fig. 5a–c; arrowheads indicate intraneuronal p-tau structures). In agreement with the quantitative GeoMx DSP findings (Fig. 3b), stereological analysis of pS396-tau revealed a 2.2-fold increase in MCI compared to NC retinas, reaching significance by Student t test. The AD retinas exhibited a significant 2.6-fold higher pS396-tau burden compared to NC controls (Fig. 5d). Retinas from D-NAD patients have on average, similar levels of pS396-tau burden as those observed in MCI and AD patients (Fig. 5d; representative images in Suppl. Fig. 3a), with a trend of 2.2-fold increase compared to NC retinas, reaching significance by Student t test. Examination of retinal pS396-tau isoforms per retinal subregion indicates that the far-peripheral retina exhibits an earlier and more significant increase of these p-tau isoforms, providing clearer separation between the diagnostic groups (Fig. 5e; representative images per retinal central, mid- and far-peripheral subregions in Suppl. Fig. 4). Notably, strong and highly significant Spearman’s correlations were identified between retinal pS396-tau burden and brain Aβ plaques, and moreover, NFTs and NTs severity scores (Fig. 5f–h, P < 0.0001; rS = 0.61, rS = 0.72, and rS = 0.74, respectively). Moderate-to-strong correlations were also found between retinal pS396-tau and ABC scores (Fig. 5i) and Braak staging, with a weak correlation to the CDR cognitive scores (Table 3).

Retinal PHF-tau increases in AD dementia patients and correlates with brain tauopathy

Upon detecting increased levels of p-tau, Oligo-tau, and tau tangle forms in the retina of MCI and AD patients, we further examined the pre-NFT forms—the PHF-tau aggregates. Previously, brain PHF-tau in AD has been associated with chronic neuroinflammation, including activated microgliosis [53, 85]. Additionally, tau-laden neurons were susceptible to excessive microglial synaptic pruning [85]. Retinal IHC analysis, using the PHF-1 antibody recognizing pS396- and pS404-tau in paired helical filaments, was performed on a subset of patient donors with MCI (n = 5, mean age 89.8 ± 5.8 years, 3 females/2 males), AD (n = 10, mean age 88.1 ± 7.4 years, 5 females/5 males), and NC controls (n = 9, mean age 82.2 ± 7.9 years, 3 females/6 males) (Fig. 6). This analysis showed marked PHF-tau deposition across retinal layers in AD patients, particularly localized in VGLUT1+ synaptic-rich OPL and IPL, alongside INL and GCL, and IBA1+ microgliosis-laden regions (Fig. 6a, b). Quantitative IHC analysis revealed a highly significant 2.3-fold increase in retinal PHF-tau in AD, but not in MCI, compared to NC controls (Fig. 6c). Notably, retinal PHF-tau burden in AD dementia patients was significantly 2.4 times elevated compared to MCI patients and 3.6 times higher compared to D-NAD patients (Fig. 6c; representative images for donors with D-NAD in Suppl. Fig. 3b), with no overlap between the AD and D-NAD groups. Examination of retinal PHF-tau distribution per retinal subregion indicates that the mid- and far-peripheral retina show more significant increases of PHF-tau forms in AD patients versus MCI and NC controls (Fig. 6d). Retinal PHF-tau deposition strongly associated with brain NTs burden and ABC scores (Fig. 6e, f; rS = 0.71, P = 0.0011 and rS = 0.69, P = 0.0014, respectively), and moderately with Braak staging, CAA, brain atrophy, and NFTs severity scores (Table 3).

Fig. 6

Paired-helical filaments of tau in retinas of MCI and AD patients. a, b Representative images of immunofluorescent stainings for PHF-1+ paired-helical filaments of tau (PHF-tau, red), vesicular glutamate transporter 1 (VGLUT1) or IBA1 (green), and DAPI (blue) in retinal cross-sections from patients with MCI and AD, and NC controls. Scale bars 20 µm. c Quantitative analysis of retinal PHF-tau immunoreactive area (n = 9 NC, n = 5 MCI, n = 10 AD, and n = 4 D-NAD). d Mapping of PHF-tau area in central (Cen), Mid-, and Far-peripheral retinal subregions in the same cohort. Spearman's rank correlation coefficient analyses are shown between retinal PHF-tau and the severity of brain e NTs and f ABC scores. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, ***P < 0.001, ****P < 0.0001, by one-way ANOVA and Tukey’s or Sidak’s multiple comparison post-tests for group analyses. Fold changes are shown in red. F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

Identification of citrullinated-tau forms in the MCI and AD retinas and association to cognition

Citrullination is a post-translational modification in which an arginine amino acid is converted to a citrulline amino acid (Fig. 7a). This process is catalyzed by peptidyl arginine deiminase (PAD) enzymes, which play a significant role in several chronic diseases [18]. Neurons expressing PAD4 were found to accumulate citrullinated proteins in AD cortices and hippocampi [2]. A recent study identified altered PAD4 expression and Cit-tau accumulation in the brains of AD patients [52]. Aberrant tau deposition activated PAD4 in neurons, leading to the citrullination of tau at multiple arginine residues. Here, we identified prominent neuronal PAD4 expression along with marked depositions of retinal and cortical citrullinated arginine (R)-209 tau (CitR209-tau) and AT8+p-tau (S202/T205) in MCI and AD patients compared to NC controls (Fig. 7b–d). In both retinal and paired brain tissues, we found increased co-localized signals of AT8+ p-tau and CitR209-tau in MCI and AD patients, with clear intra-neuronal CitR209-tau signals in the INL (Fig. 7e; higher magnification of MCI retina). We conducted quantitative analysis of retinal CitR209-tau in a cohort of donors with MCI (n = 8, mean age 89.13 ± 5.22 years, 4 females/4 males), AD (n = 21, mean age 82.81 ± 13.40 years, 10 females/11 males), and NC controls (n = 18, mean age 80.72 ± 7.89 years, 9 females/9 males). Stereological quantification of retinal CitR209-tau indicates a substantial 3.5-fold and 4.1-fold increases in MCI and AD patients, respectively, compared to NC controls (Fig. 7f). Analysis of CitR209-tau burden per retinal subregion indicated that CitR209-tau appears earlier and more pronouncedly in the central retina (Fig. 7g). Moreover, Pearson’s correlation analysis indicated a strong positive association between the two post-translational modifications of tau (CitR209-tau and AT8+p-tau) in the retina (Fig. 7h; rP = 0.74, P < 0.0001). Spearman’s rank correlation analyses demonstrated weak-to-moderate associations between retinal CitR209-tau and CDR or MMSE cognitive scores (Fig. 7i, j), with no associations to the severity of brain pathology or disease staging (Table 3).

Fig. 7

Identification of citrullinated-tau isoforms in the MCI and AD retinas. a Graphical illustration of post-translation modification of tau citrullination, catalyzed by protein arginine deiminase (PAD) enzymes. b Representative images of immunofluorescent staining for PAD4 enzyme (green), citrullinated tau at arginine R209 site (CitR209-tau, red), and nuclei (DAPI, blue) in retinal cross-sections from MCI and AD patients as compared with NC controls. Scale bars 20 µm. c-e Representative images of immunofluorescent staining for AT8+p-tau (green), CitR209-tau (red), and DAPI (blue) in c brain cortex (A9) and d, e retinal cross-sections from patients with AD, MCI, and NC controls. e High-magnification image shows intra-cellular CitR209-tau labeling in retinal INL of MCI patient. Scale bars 20 µm. f Quantitative analysis of retinal CitR209-tau immunoreactive area (n = 18 NC, n = 8 MCI, and n = 21 AD). g Mapping of CitR209-tau area in central (Cen), Mid-, and Far-peripheral retinal subregions in the same cohort. h Pearson’s correlation coefficient (r) between retinal CitR209-tau and retinal AT8+ p-tau. Spearman's rank correlation coefficient analyses are shown between retinal CitR209-tau and i CDR and j MMSE cognitive scores. k Heatmap of Pearson’s correlation coefficient (r) analyses between retinal Aβ forms [12F4+-Aβ42, intraneuronal scFvA13+-Aβ oligomers (AβOi), arterial (A) 11A50-B10+-Aβ40, and venular (V) 11A50-B10+-Aβ40] and retinal tau isoforms [AT8+, AT100+, and pS396-p-tau, CitR209-tau, Oligo-tau, PHF-tau, and MC-1 tau tangles]. The strength (darker color) and direction (positive-red, negative-blue) of the correlations are shown. Pearson’s rP values are indicated in larger fonts and below are the number (n) of pair-wise correlations. Individual data points (circles) and median, lower and upper quartile are shown in violin plots. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, by one-way ANOVA and Tukey’s or Sidak’s multiple comparison post-tests for group analyses. Fold changes are shown in red. F, female; M, male; age (in years); n, sample size. Illustrations created with Biorender.com

We then investigated the relationships between the various retinal tau isoforms and retinal Aβ alloforms; the latter recently investigated by our group [44, 73]. To achieve this, we performed pair-wise Pearson’s correlation analyses between retinal tau isoforms (AT8+, AT100+, and pS396-p-tau, CitR209-tau, Oligo-tau, PHF-tau, and MC-1 tau tangles) and retinal Aβ species (12F4+-Aβ42, intraneuronal scFvA13+-Aβ oligomers (AβOi), arterial (A) 11A50-B10+-Aβ40, and venular (V) 11A50-B10+-Aβ40) (Fig. 7k). A very strong and significant correlation was found between retinal Oligo-tau and retinal Aβ42 forms (rP = 0.86, P < 0.0001) and retinal arterial Aβ40 burden (rP = 0.76, P < 0.0001). Retinal CitR209-tau strongly correlated with retinal Aβ42 burden (rP = 0.64, P < 0.001). Moderate linear associations were found between Oligo-tau and retinal AβOi as well as between retinal AT8+-p-tau and retinal Aβ42 and arterial Aβ40 alloforms (Fig. 7k).