Parallels between retinal and brain pathology and response to immunotherapy in old, late‐stage Alzheimer's disease mouse models

Abstract Despite growing evidence for the characteristic signs of Alzheimer's disease (AD) in the neurosensory retina, our understanding of retina–brain relationships, especially at advanced disease stages and in response to therapy, is lacking. In transgenic models of AD (APPSWE/PS1∆E9; ADtg mice), glatiramer acetate (GA) immunomodulation alleviates disease progression in pre‐ and early‐symptomatic disease stages. Here, we explored the link between retinal and cerebral AD‐related biomarkers, including response to GA immunization, in cohorts of old, late‐stage ADtg mice. This aged model is considered more clinically relevant to the age‐dependent disease. Levels of synaptotoxic amyloid β‐protein (Aβ)1–42, angiopathic Aβ1–40, non‐amyloidogenic Aβ1–38, and Aβ42/Aβ40 ratios tightly correlated between paired retinas derived from oculus sinister (OS) and oculus dexter (OD) eyes, and between left and right posterior brain hemispheres. We identified lateralization of Aβ burden, with one‐side dominance within paired retinal and brain tissues. Importantly, OS and OD retinal Aβ levels correlated with their cerebral counterparts, with stronger contralateral correlations and following GA immunization. Moreover, immunomodulation in old ADtg mice brought about reductions in cerebral vascular and parenchymal Aβ deposits, especially of large, dense‐core plaques, and alleviation of microgliosis and astrocytosis. Immunization further enhanced cerebral recruitment of peripheral myeloid cells and synaptic preservation. Mass spectrometry analysis identified new parallels in retino‐cerebral AD‐related pathology and response to GA immunization, including restoration of homeostatic glutamine synthetase expression. Overall, our results illustrate the viability of immunomodulation‐guided CNS repair in old AD model mice, while shedding light onto similar retino‐cerebral responses to intervention, providing incentives to explore retinal AD biomarkers.

AD is a fatal neurodegenerative disorder and the most common form of dementia clinically diagnosed by the progressive loss of memory and cognitive function (Alzheimer's Association, 2018).
Aβ burden, astrocytosis, and microgliosis were diminished, brain milieu shifted from a pro-to anti-inflammatory profile, synapses were rescued, and hippocampal neurogenesis was induced, ultimately preventing cognitive decline (Bakalash et al., 2011;Baruch et al., 2015;Butovsky et al., 2006Butovsky et al., , 2007Frenkel et al., 2005;Koronyo et al., 2012Koronyo et al., , 2015Rentsendorj et al., 2018). The therapeutic mechanisms of GA immunization, mainly explored in these transgenic AD model mice (Butovsky et al., 2006(Butovsky et al., , 2007Koronyo et al., 2012Koronyo et al., , 2015Rentsendorj et al., 2018), were attributed to a shift in microglial phenotype and enhanced recruitment of neuroprotective, peripherally derived monocytes and macrophages directly involved in Aβ clearance, immunoregulation, and neuroregeneration (Koronyo et al., 2015;Rentsendorj et al., 2018). While these promising effects were reported in pre-and early-symptomatic (5-to 12-month-old) ADtg mice, they have never been investigated in old, late-stage mouse models of AD. Given that aging is a fundamental factor in AD development (Sperling et al., 2020), mice allowed to progress into old age may better represent the clinical manifestations of human AD. Such old mice could offer greater insights into the potential translation of effective therapies.
Beyond the brain, a growing number of studies have provided evidence for AD-specific protein aggregation, vascular pathology, and markers of neuroinflammation in the neurosensory retina of various transgenic, induced, and spontaneous animal models of AD (Chang et al., 2020;Chiasseu et al., 2017;Do et al., 2019;Doustar et al., 2017;Grimaldi et al., 2018;Hampel et al., 2018;Hart et al., Koronyo et al., 2017;Koronyo-Hamaoui et al., 2011;La Morgia et al., 2016;Schon et al., 2012;Schultz et al., 2020;Shi et al., 2020;Tsai et al., 2014). Biochemical analyses of Aβ 40 and Aβ 42 peptide levels in retinal and brain tissues from several transgenic murine models and AD patients revealed increases in both peptides in the AD retina as compared to controls, with higher levels in the brain and correlations with brain levels (La Morgia et al., 2016;Schultz et al., 2020;Shi et al., 2020). However, assessment of such relationships between paired retinas derived from oculus sinister (OS) versus oculus dexter (OD) eyes and left versus right cerebral hemispheres for Aβ load was not previously undertaken. This is especially true for response to therapy and in old ADtg mice. A recent study showed the accumulation of Aβ 40 and Aβ 42 peptides in whole ocular tissues of 5× FAD transgenic mice with a reduction in ocular Aβ following neprilysin treatment (Parthasarathy et al., 2015). Although these findings are significant, the inclusion of non-neuronal tissue and the lack of analyzed brain tissues hinder the ability to evaluate connections between the neuro-retina and brain. In this context, we previously demonstrated a comparable Aβ-deposit reduction in retinal and cerebral tissues of 12-month-old ADtg mice subjected F I G U R E 1 Experimental design and intervention timeline assessing cerebral and retinal tissues in old, late-stage ADtg mice. (a) Experimental timeline for mouse Cohort 1: 20-month-old APP SWE /PS1 ΔE9 (ADtg) mice underwent weekly, subcutaneous injections of glatiramer acetate (GA, also known as Copaxone ® ; 100 µg) for a 2-month duration (n = 7 mice). Age-and sex-matched ADtg mice were subcutaneously injected with PBS in the same regimen and naïve non-transgenic (WT) littermates were used as controls (n = 7 mice per group). (b) One week following the last injection, mice were sacrificed, and tissues were collected as described. Paired brains and eyeballs from Cohort 1 were allocated for further analyses: the right (R) cerebral hemisphere was used for immunohistochemistry (IHC), the left (L) posterior brain was used for quantitative biochemical Meso Scale Discovery (MSD) and Mass Spectrometry (MS) assays, and both oculus sinister (OS, left) and oculus dexter (OD, right) eyeballs were collected, the neurosensory retinae isolated, and proteins assessed by MSD and MS analyses. OS and OD retinae were separately analyzed by MSD and pooled together for MS analysis. (c) Cohort 2 was comprised of old ADtg mice (n = 15 mice; average age of 18 months). L and R posterior brains as well as OS and OD eyeballs were collected and analyzed separately for Aβ proteins by MSD. (d) Preparation of mouse neuro-retina and posterior brain Aβ proteins for quantification by MSD. Each tissue was prepared separately for analysis (OS retina, OD retina, and left and right posterior brains). The protocol involves suspension in lysis buffer (LB), homogenization via sonication, concentration with speed vac, and protein denaturation with hexafluoroisopropanol (HFIP) followed by evaporation and resuspension in PBS prior to protein concentration analysis to immunomodulation with MOG-45D (altered myelin-derived peptide) loaded on dendritic cells (Koronyo-Hamaoui et al., 2009, 2011) and subsequently showed the feasibility to noninvasively detect progressive appearance and clearance of individual retinal Aβ plaques following GA immunization (Koronyo et al., 2012).
Collectively, these early studies provide the rationale to explore the relationship between retinal and cerebral pathology, including accumulation of both non-amyloidogenic and disease-associated amyloidogenic Aβ alloforms, in old age and later stages of disease.
There is also a need to study the responses to immunomodulation intervention in more clinically relevant, aged murine models and to quantitatively determine parallels between the brain and retina.
Addressing the above unknowns, this study provides evidence for the predictability of cerebral Aβ 42 and Aβ 40 burden via quantitative measurements of OS and OD retinal counterparts in old ADtg mice. Moreover, in addition to identifying similar responses in the neuro-retina and the brain this study demonstrates the efficacy of GA immunomodulation in restricting vascular pathology and neuroinflammation while improving synaptic density at such late-stage disease.

| Lateralization of Aβ levels in retinae and brain hemispheres from old AD model mice
To investigate levels of disease-associated amyloidogenic (Aβ 1-42 and Aβ 1-40 ) and non-amyloidogenic (Aβ 1-38 ) alloforms in neurosensory retina and brain tissues, as well as evaluate the relationship between retinal and brain Aβ burden, we analyzed two cohorts of 18-and 22-month-old, late-stage transgenic APP SWE /PS1 ∆E9 (ADtg) mice (Figure 1a-c). Further, to assess retinal and brain Aβ pathology in response to immunotherapy in old mice, a cohort of 20-month-old ADtg mice underwent GA immunization and were compared against age-and sex-matched PBS-control ADtg and naive WT mice (Cohort 1, experimental timeline and GA immunization regimen are shown in Figure 1a; n = 7 mice per group). Cohort 2 composed of old, untreated 18-month-old ADtg mice (n = 15) were allotted to validate concentrations of Aβ alloforms in retinal and brain tissues. Retinae derived from OS and OD eyes, as well as left and right posterior brains, were allocated for histological (immunohistochemistry-IHC) or biochemical (Meso Scale Discovery-MSD, mass spectrometry-MS) analyses as outlined in Figure 1b  ratio similarly showed lateralization, albeit higher in OD versus OS retina (p = 0.0004, paired two-tailed Student's t test; Figure 1c).
Despite this asymmetry, tight correlations in levels of Aβ alloforms and Aβ 42/40 ratios were revealed between paired OS and OD retinae (Pearson's correlation coefficient r = 0.75 and p = 0.0021 for Aβ 42 , r = 0.90 and p < 0.0001 for Aβ 40 , r = 0.94 and p < 0.0001 for Aβ 38 , F I G U R E 2 Retinal and cerebral Aβ alloforms in old ADtg mice and following immunotherapy. (a-c) Analysis of Aβ 1-42 , Aβ 1-40, and Aβ 42/40 ratio levels (n = 7 mice per group) in OS versus OD retinae from Cohort 1 of GA-immunized (blue) and PBS-control (black) old ADtg mice. Data indicate Aβ concentrations for individual mouse in OS versus OD retina analyzed by paired Student's t test. Pearson's r correlations between levels of each Aβ 1-42 , Aβ 1-40 , and Aβ 42/40 ratio in OS and OD retinae are also shown (n = 13-14 mice; right graphs). (d) Schematic display of Aβ 1-42 , Aβ 1-40 , and Aβ 1-38 alloform concentrations (Average; ± SEM in brackets) in each retina from Cohort 1. Data presented in pg Aβ per µg total protein. Lower percentages of amyloid levels in OD versus OS retinae are shown in red, and OS to OD ratios of Aβ concentrations are indicated below. (e-f) Aβ 42/40 ratios as assessed by MSD analysis in paired OS versus OD retinae from ADtg mice (e; Cohorts 1 and 2 without GA group; n = 20 mice) and in paired L versus R brains from Cohort 2 ADtg mice (f; n = 15 mice). Lateralization was determined by paired Student's t test. (g-m) Pearson's r correlations between retinal and cerebral Aβ burden in ADtg mice from Cohort 2. (g-h) Contralateral correlations between OS retina versus R brain and OD retina versus L brain for (g) Aβ 42 and (h) Aβ 40 levels (n = 12-14). (i) Correlation between average retinal and average cerebral Aβ 42/40 ratios. (j-l) Unilateral correlations of OS retina versus L brain (green) or versus R brain (red) for (j) Aβ 42 , (k) Aβ 40 , and (l) Aβ 42/40 ratio (n = 13). (m) Schematic illustration portraying strength of associations between each retina and each posterior brain for Aβ 42 , Aβ 40 , and Aβ 42/40 ratio in old ADtg mice. The analyzed posterior brain includes tissue between −1 and −4 mm bregma. Strong associations in red (r > 0.7), moderate associations in blue (r = 0.5-0.7), and weak or no associations (na) in gray (r = 0.0-0.5). (n-q) Analysis of retinal versus cerebral Aβ levels in Cohort 1 old ADtg mice following GA immunization. (n) Brain and retinal Aβ 42/40 ratios in PBS-control versus GA-immunized ADtg mice (n = 7 mice per group). (o-q) Pearson's r correlations between levels of Aβ 42 in L brain and (o) OD retina, (p) OS retina, and (q) an average of both retinae (n = 14 mice). Strong correlations in Aβ 1-42 burden between brain and retinal tissues are especially apparent following GA immunization (q). Graphs display individual data point for each mouse, with bar graphs also indicating group mean and standard error of mean (SEM) values. Mouse sex is designated as filled circles for males and open circles for females (gender not shown in correlation graphs). *p < 0.05, **p < 0.01, ***p < 0.001 assessed by paired Student's t test for twogroup comparisons, and a two-way ANOVA with Sidak's post-test for group analysis for brain and retinal tissues Aβ 42 and Aβ 40 levels (p = 0.0057 and p = 0.0082, respectively, by twotailed paired Student's t test; n = 21, Figure S1G-H). Retinal Aβ 38 alloforms were below detectable levels for most retinae from this cohort.
As it refers to disease-relevant Aβ 42/40 ratios, OD retina dominant lateralization was identified in these old ADtg mice (p = 0.0335, by two- An especially strong correlation was found for levels of Aβ 42/40 ratio (r = 0.98 and p < 0.0001; Figure S1L). Moreover, levels of each amyloidogenic Aβ 42 and Aβ 40 alloform strongly predicted co-accumulation in the same retina or brain location (Pearson's r > 0.90 and p < 0.0001 across all retinal and brain tissues; Figure S2A-F).

| Association between retinal and brain Aβ levels
Next, to determine the feasibility of predicting cerebral Aβ burden via its levels in the retina, brain levels were correlated against contralateral or ipsilateral, each OS and OD, and both retinae in

| Retinal and cerebral Aβ levels strongly correlate in response to intervention
To evaluate effects of GA immunization on levels of retinal and brain Aβ alloforms in old ADtg mice (Cohort 1), we initially compared MSD data between GA-immunized and PBS-control groups (Figures 2n and S1D-F). Both the retinae and the posterior left brain did not show significant reductions in Aβ levels following GA immunization at this advanced, late-stage disease ( Figure S1D-F). However, Aβ 42/40 ratio levels in the OD retina were significantly reduced in GA-immunized as compared to PBS-control mice (p = 0.0246, twoway ANOVA with Sidak's post-test; Figure 2n). Despite marked differences between brain and retina levels for each Aβ alloform

| Immunomodulation ameliorates vascular and parenchymal Aβ deposits in old ADtg mice
To further explore the therapeutic potential of immunomodulation in old, advanced disease stage mice, we performed in-depth histological analyses on brain tissues from Cohort 1. As outlined in Figure 1b, both retinas and left posterior brain tissues were allocated for biochemical MSD and MS analyses, and additionally, the right brain hemispheres were evaluated by IHC analyses.
Histological examination of various AD-relevant biomarkers, including Aβ-plaque burden, both vascular and parenchymal deposits, ionized calcium-binding adaptor molecule 1 (Iba-1), protein tyrosine phosphatase receptor type C (CD45), glial fibrillary acidic protein (GFAP), and glutamine synthetase (GS), was conducted on brain tissues encompassing the following regions as specified in  Figure S4B-C). GA immunization in these old ADtg mice resulted in a consistently smaller average area, lower quantity, and shortened length of the densecore plaques (p = 0.0054 with 28% reduction in size, p = 0.0010 with 47% reduction in count, and p < 0.0029 with 16% reduction in length of DC plaques; Figure 3i-k). These differences did not reach statistical significance for non-DC diffuse plaques ( Figure S4D-F).
Further stratification by plaque size revealed that GA immunization especially modified large plaques (p = 0.0331 with 24% reduction in area, p = 0.0055 with 25% decrease in width, and p = 0.0017 with 28% decrease in length of large plaques; Figure 3l-n; extended data on other plaque types and measurements see Figure S4G-I).
Overall, these results may uncover critically beneficial effects of immunomodulation on limiting pre-existing Aβ deposits at an old age and advanced disease stage.

| Immunomodulation curbs microgliosis and astrogliosis with increased recruitment of peripheral myelomonocytes
Neuroinflammatory responses mediated by innate immune cells surrounding Aβ plaques have been implicated in modulating plaque structure and subsequent toxicity in both animal models of AD and human patients (Rasmussen et al., 2017;Vilella et al., 2018;Wang et al., 2016).
There were substantial increases in cerebral Iba1 + microgliosis, with activated-type cell morphology surrounding plaques, as well as in Iba1 + CD45 hi myeloid cell population in old PBS-injected ADtg mice when compared to matched naïve WT mice (Figure 4a-b; extended representative images in Figure S5-S7). This was evident throughout the analyzed brain regions, including Hipp, Ctx, Ent, and total brain regions (p < 0.0001 for all comparisons, by one-way ANOVA and Sidak's post-test; Figure 4a-b). Importantly, even at old age, our immunomodulation strategy considerably diminished Iba1 + microgliosis in ADtg mice, with an average reduction of 56% in total brain regions (p < 0.0001, by one-way ANOVA and Sidak's post-test; Figure 4a). In contrast, immunization further induced recruitment of peripherally derived Iba1 + CD45 hi myeloid cells into the Hipp, Ctx, and Ent regions, by an average 78% increase in total brain in GAimmunized vs PBS-control ADtg mice (p < 0.001-0.0001; Figure 4b).
Colocalization analysis of Iba1 hi CD45 hi out of total Iba1 + population revealed that the portion of infiltrating innate immune cells was significantly increased from 27% in PBS-control mice to 40% in GAimmunized old ADtg mice (p = 0.0061, unpaired Student's t test; Representative microscopic images in Figure 4g demonstrate the abundance of cortical Iba1 + CD45 hi tagged peripheral immune cells following GA immunomodulation compared to PBS treatment (extended images in Figures S5-S7). Quantification of the ratio between Iba1 + CD45 hi area and 6E10 + Aβ-plaque area revealed a substantial ~3-fold increase in these infiltrating myelomonocytes per plaque site in the GA-immunized versus PBScontrol mouse brains (p < 0.0001, by unpaired Student's t test; GFAP is one of the key biomarkers for detecting reactive, scar-tissue related astrocytes (De Strooper & Karran, 2016;Osborn et al., 2016). Quantitative IHC analysis of GFAP + reactive astrocyte area showed greater cerebral astrogliosis in PBS-control old ADtg mice as compared to the old wild-type mice (p < 0.0001; Figure 4k). GA immunization consistently curtailed astrocytosis by 30 and 35% in various brain regions, including Hipp, Ctx, Ent, and total brain (p < 0.0001, p < 0.0001, p = 0.0361, and p < 0.0001, respectively, by one-way ANOVA and Tukey's post-test correction; Figure 4k). Moreover, levels of cerebral GFAP + astrocytosis directly correlated with 6E10 + -Aβ burden (Pearson's r = 0.84 and p = 0.0006; Figure 4l; extended correlation analyses per brain region in Figure S8C).
Next, we evaluated the astrocyte-specific biomarker glutamine synthetase (GS), an enzyme associated with GFAP reactivity and responsible for synaptic recycling of extracellular glutamate (Rudy et al., 2015;Son et al., 2019;Zou et al., 2011). Peroxidase-based immunostaining for GS in coronal brain sections demonstrated morphological and intensity differences between GA-immunized and PBS-control old ADtg mice ( Figure S8D). The GA-immunized group showed patterns indistinguishable from those of the non-transgenic WT mice. Across the three experimental groups, GFAP + reactive astrocyte area was associated with levels of GS immunoreactive area (Pearson's r = 0.72 and p = 0.0005; Figure   S8E). Quantitative IHC analysis revealed a significant increase in GS + area in PBS-control ADtg mice when compared to naïve WT mice (p = 0.0244), with levels of GS normalized to WT in the Ent of GA-immunized ADtg mice (p < 0.05; Figure S8F). Overall, our data suggest that GA immunization in old AD model mice has multiple beneficial effects toward reducing Aβ plaques and restoring homeostatic microglia and astrocytic milieu, along with recruitment of Iba1 + CD45 hi myeloid cells.

| Synaptic preservation by GA immunization in old ADtg mice
The critical role of astrocytes in synaptic homeostasis, notably GS activity in regulating extracellular glutamate, prompted us to evaluate synaptic integrity and its relationship to astrocytic GS expression in these old ADtg mice. Histological examination of synaptic density was assessed for the post-synaptic density protein 95 (PSD95) biomarker in three AD-relevant brain regions, including hippocampal subregions, as outlined in Figure 5a. A significant increase in PSD95 + area, reflecting improved synaptic density, was seen in the Ent region following GA immunization in ADtg mice, reaching equivalent levels to those measured in old WT littermates (p = 0.0411 and 98% increase; Figure 5b; extended analysis of various anatomical hippocampal layers is shown in Figure S9A). Post-synaptic area was inversely correlated with astrocytic GS expression per cell, which was independent of treatment or genotype group (r = −0.57 and p = 0.0127; Figure 5c). F I G U R E 3 Decreased cerebral Aβ -plaque pathology and vascular amyloidosis in old ADtg mice following GA immunotherapy. (a) Cerebral map indicating specific brain regions analyzed by IHC; regions include the cingulate cortex/retrosplenial area (Ctx), hippocampus (Hipp), and entorhinal cortex/piriform area (Ent). (b-c) Representative coronal sections of a cortical region stained for astrocytes (GFAP, green), Aβ plaques (6E10, red), and cell nuclei (DAPI; blue) in (b) PBS-control and (c) GA-treated ADtg mice. (d) Quantitative IHC analysis of 6E10 + Aβ-plaque area in total brain, Hipp, Ctx, and Ent of all experimental groups (n = 6 mice per group). (e) Analysis of cerebral amyloid angiopathy (CAA) scores in the Ent of GA-immunized versus untreated (PBS) old ADtg mice (n = 7 mice per group). (f) Representative images illustrating the scoring method used to assess vascular 6E10 + Aβ deposits [termed as cerebral amyloid angiopathy (CAA) scores], with scale ranges from 0 to 4, with higher scores for greater vascular Aβ pathology. (g) Representative images and measurements of perimeter, length (largest diameter), and width (smallest diameter) acquired per Aβ plaque (top); Dense-core and non-dense-core/diffuse plaque subtypes are demonstrated (bottom). (h) Microscopic images showing classification of plaque by size, as defined by area, length and width. Accordingly, plaques were categorized into four subgroups: x-small, small, medium, and large. (i-k) Quantitative analysis of dense-core plaque phenotype within the Ent of GA-immunized versus PBS-control ADtg mice, including (i) total count, (j) average area, and (k) length. (l-n) Quantitative analysis of large Aβ-plaque count, as determined by (l) area, (m) width, and (n) length, within the Ent of GA-immunized versus PBS-control ADtg mice (n = 6-7 mice per group). Bar graphs indicate mean, standard error of mean (SEM), and individual data points. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 assessed by unpaired Student's t test for two-group comparisons, and a one-way ANOVA with Tukey's post-test for three or more groups F I G U R E 4 Reduced cerebral microgliosis and astrogliosis along with peripheral monocyte recruitment following immunomodulation. (a) Quantitative IHC analysis of Iba1 + microgliosis within predetermined brain regions: Hipp, Ctx, and Ent, as well as their average (Brain) in GA-immunized versus PBS-control ADtg mice, and in naïve WT mouse littermates. (b) Quantitative IHC analysis of Iba1 + CD45 hi infiltrating peripheral immune cells in brain regions and their average (Brain) (n = 6 mice per group). (c) Analysis of percent area of Iba1 hi / CD45 hi peripheral monocytes population of Iba1 + myeloid cell area within PBS-control and GA-immunized groups (n = 6 mice per group). (d-e) Representative micrographs of inflammatory cells, GFAP + astrocytes (cyan) and Iba1 + myelomonocytes (red), surrounding 4G8 + Aβ plaques (yellow) in the Ent cortex of old, late-stage (d) PBS-control and (e) GA-immunized ADtg mice. (f) A representative high-magnification micrograph of Ent cortex of GA-immunized ADtg mice with an Iba1 + myelomonocytic cell (red) seen engulfing 4G8 + Aβ; white arrow tags location between channels. (g) Representative micrographs of Iba1 + myelomonocytes and CD45 + hematopoietic immune cells within PBScontrol and GA-immunized mice. (h) IHC analysis of CD45 hi area/6E10 + area ratio (n = 6 mice/group). (i) Pearson's r correlation analysis between CD45 hi hematopoietic cells and 6E10 + Aβ-plaque deposits in Ctx sections including PBS-control and GA-immunized mice (n = 12 mice). (j) Separate Pearson's r correlations between CD45 hi hematopoietic cells and 6E10 + Aβ-plaque deposits in old PBS-control and GAimmunized ADtg mice demonstrating retention of correlation with treatment. (k) Quantitative IHC analysis of GFAP + astrogliosis in total brain regions for all experimental groups (n = 6-7 mice per group). (l) Pearson's r correlation analyses between GFAP + astrogliosis and 6E10 + Aβ for total brain plaque area in PBS-control and GA-immunized ADtg mice (n = 12 mice). Bar graphs indicate mean, standard error of mean (SEM), and individual data points. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 assessed by unpaired Student's t test for two-group comparisons, and a one-way ANOVA with Tukey's post-test for three or more groups F I G U R E 5 Effects of GA immunization on synaptic density and AD-related retino-cerebral proteins in old ADtg mice. (a) Coronal section map displaying specific brain regions analyzed by IHC; regions include the cingulate cortex/retrosplenial area (Ctx), hippocampus (Hipp), and entorhinal cortex/piriform area (Ent). Magnification of 100× microscopic images covering these brain regions was analyzed for postsynaptic biomarker (PSD95). 15 consecutive z-stack images were captured with ApoTome-equipped Zeiss microscope provided highresolution images of synaptic puncta. (b) Quantitative analysis of PSD95 + synaptic area assessed in Hipp, Ent, and combined brain regions in all experimental groups (n = 6-7 mice/group). (c) Pearson's r correlation analysis between total brain PSD95 + synaptic area and astrocytic marker glutamine synthetase (GS) + area per cell (n = 18 mice). (d) Collective Pearson's r correlation of 6E10 + Aβ-plaque burden against PSD95 + post-synaptic density (black), GS + area/cell (brown), and GFAP + astrogliosis (green) in the Ent (n = 10-12 mice). (e) Representative fluorescent images (40×) of coronal brain sections from Ent stained for astrocytes (GFAP, red), post-synaptic protein (PSD95, green), and cell nuclei (DAPI, blue) from PBS-control (top) and GA-immunized (bottom) old ADtg mice. (f) Representative, high-magnification images (100×) of a GA-immunized ADtg mouse brain showing density of PSD95 density. (g-h) Mass spectrometry analysis of total normalized peak protein area (TNPPA) of significantly changed synaptic proteins, including (g) synaptophysin and (h) synapsin-2, between all experimental groups (n = 4-6 mice per group). (i) Heat map displaying relative fold change of significantly changed proteins in mass spectrometry analysis. Highlighted are AD-related amyloid-associated markers (amyloid-β A4 protein-APP/Aβ, clusterin-CLN, lysosomal-associated membrane protein 1 and 2-LAMP1/2) and inflammatory markers (glutamine synthetase-GS, intercellular adhesion molecule 1-ICAM1, h-2 class I histocompatibility antigen-HA11) that were significantly up-or down-regulated in brain and retinal tissues of PBS-control ADtg mice versus naïve WT mice and/or GA-immunized versus PBS-control ADtg mice (n = 4-6 mice per group). Bar graphs indicate mean, standard error of mean (SEM), and individual data points. *p < 0.05, **p < 0.01, ***p < 0.001, by one-way ANOVA with Tukey's post-test for three or more experimental groups. Mass spectrometry analysis by unpaired Student's t test. Pearson's r correlation analysis was used to determine statistical association TA B L E 1 Quantitative mass spectrometry analysis of preselected proteins from brain and retinal tissue

0.0287
Data table includes analysis of PBS-control, GA-immunized, and WT naïve experimental groups (n = 5-7 mice/group). Protein peak areas were normalized to the total protein peak area of the respective sample and subjected to one-sample t tests to compare relative protein peak areas between the respective sample groups. t test p-value smaller than 0.05 and fold change ±1.2 was highlighted as differentially expressed proteins. Ξ-Average of total normalized protein peak area (×10 5 ).
Further, to explore effects of GA immunization on retinal and brain tissues from old, late-stage ADtg mice, a sensitive and quantitative MS proteomic analysis was applied (Figure 5g-i and Table 1; the complete MS datasets of identified proteins are included in Table   S1 for brain and Table S2 for

| Cerebral and retinal AD-related protein profiles in response to immunomodulation
Next, proteins associated with amyloid, processing, clearance, and inflammation that were similarly and significantly down-or up-regulated in brain and retinal tissues from old mice either in disease or following GA immunization, as determined by MS analysis, are highlighted in Figure 5i. Specifically, parallels between brain and reti-  Table 1). Brain MS data revealed elevated GS concentrations in GA-immunized versus PBS-control ADtg mice (51% increase and p = 0.0390; Table 1), which were remarkably normalized to levels observed in the WT mice. This response in GS astrocytic function following GA was mirrored in the retina, demonstrates another linkage between these two tissues (27% increase and p = 0.0001; Table 1). Compared to WT, the immune markers intercellular adhesion molecule 1 (ICAM1), involved in T cell-APC adhesion, and the mouse MHC molecule H-2 class 1 histocompatibility complex (HA11) were increased in ADtg mouse brains (129% increase and p = 0.0026, 59% increase and p = 0.0086, respectively) and retina (20% increase and p = 0.0122, 27% increase and p = 0.0101, respectively; Figure 5i heatmaps and Table 1). However, the effects of GA did not reach statistical significance in both tissues.
In the context of amyloid processing and clearance, our quantitative MS analysis of cerebral and retinal APP/Aβ further confirmed substantial increases with disease, albeit six times greater in the brain versus the retina (20-fold vs. 3.5-fold increases, respectively, p < 0.0001; Figure 5i heatmaps and Table 1). Importantly, retinal APP/Aβ protein levels in ADtg mice were significantly reduced by GA treatment (11% and p = 0.0396; Table 1). The Aβ-related clusterin chaperone protein (CLU) showed substantial increases in cerebral concentration in old ADtg vs WT mice (~5-fold increase and p = 0.0237) and to a lesser but significant extent in the retina (20% increase and p = 0.0204). Finally, two lysosomal-associated membrane proteins 1 and 2 (LAMP1/2) were similarly but to a lesser extent affected by disease and GA immunization in the retina versus the brain. LAMP1 and LAMP2 were significantly elevated in ADtg versus WT mouse brains (54% increase and p = 0.0102, 34% increase and p = 0.0287, respectively), but did not reach statistical significance in the retina (p = 0.4068 and p = 0.0965, respectively).

| DISCUSS ION
This study provides the first evaluation of amyloidogenic and non-  (Long et al., 2013;Mesulam et al., 2014;Minkova et al., 2017;Wahlund et al., 1993). In addition, AD patients have been reported to display reductions in left hemisphere glucose metabolism (Lehmann et al., 2013;Toga & Thompson, 2003;Weise et al., 2018), which has been associated with functional declines in verbal fluency (Weise et al., 2018). Other structural and molecular measurements also indicated lateralization between the two hemispheres, including a number of studies showing left-dominant accumulation of tauopathy, which could even predict disease onset (Mesulam et al., 2014;Minkova et al., 2017;Tetzloff et al., 2018;Wachinger et al., 2016). Most interestingly, in patients with righthand dominance there was higher left-side brain tauopathy, while in patients with left-hand dominance and known right-hemisphere language dominance, tau burden lateralization shifted to the right hemisphere (Mesulam et al., 2014). Hence, it is postulated here that our findings of side-dominant susceptibility of Aβ pathogenesis could be an outcome of differences in eye dominance, neuronal firing and circuit activity, and metabolism, eventually affecting Aβ production, processing, propagation, or clearance processes.
Nevertheless, despite these disparities, there was a close linear correlation in levels of all Aβ alloforms between the left and right retinae. In addition to having higher Aβ peptide concentrations, the OS retina appears to better predict as compared to the OD retina its cerebral counterpart. If this disparity holds true in human patients, this may have implications in research methodologies and even organ acquisition. Additionally, results from prior investigations on retinal and brain pathology, averaging data from both eyes and/or brain hemispheres, may be skewed. Future studies are warranted to establish retinal and brain Aβ-burden lateralization and investigate the possible explanation of such lateralization in order to better understand processes regulating CNS Aβ accumulation and removal.

While two previous reports have analyzed ocular Aβ burden by
ELISA in several transgenic murine AD models (Alexandrov et al., 2011;Parthasarathy et al., 2015;Schultz et al., 2020), they did not include analysis of paired retinas and brains, lateralization assessment, or they did not separate the neurosensory retina from other ocular tissues. Our previous investigations identified for the first time the presence of Aβ plaques and characterized their morphology in retinas from AD patients and early-stage cases Koronyo-Hamaoui et al., 2011;La Morgia et al., 2016;Shi et al., 2020). Moreover, our preliminary data detected a correlation between paired retinal Aβ 42 -containing plaques and cerebral plaque burden in these patients . In terms of retinal and brain response to therapy, reports in 12-and 14-month-old murine models of AD showed similar reductions in retinal and brain Aβ plaques following immunization with dendritic cells loaded with MOG45D  or with anti-Aβ antibodies (Hwang et al., 2014;Liu et al., 2009). However, this was never precisely quantified by biochemical methods, nor measured specifically in old mice following GA immunization. Importantly, the feasibility to detect in vivo individual Aβ plaque appearance and clearance dynamics in real time via a noninvasive retinal curcumin optical imaging method (Koronyo et al., 2012), and with aging (Sidiqi et al., 2020), encourages utilization of retinal amyloid imaging to gauge response to therapy with high spatial resolution. One limitation of this study is the lack of live retinal/brain imaging data to correlate with the histopathological findings. Our results rely on histological and biochemical analyses, revealing new phenomena that support the need to further correlate between live retinal/brain amyloid imaging and AD-related pathological biomarkers. Future investigation using live retinal imaging is warranted to identify such AD biomarkers, especially to facilitate evaluation of response to therapy.
The current studies led us to reveal the following fundamental unknowns regarding the potential connection between retinal and brain pathology in ADtg mice: (1) While the levels of retinal Aβ 38 , the more benign alloform, weakly corresponded to its left-posterior brain counterpart, the levels of retinal Aβ 40 and Aβ 42 , the synaptotoxic and AD-pathognomonic alloforms (Raskatov, 2019), tightly predicted the levels of their brain counterparts, especially following Although one limitation of this study is the lack of cognitive testing, which was due to enhanced frailty and attrition rate among these old ADtg mice, our results of pre-and post-synaptic preservation (e.g., PSD95, synaptophysin, synapsin-2), which are typically predictive of cognitive function (Ferreira et al., 2015;Shankar et al., 2008), may indeed reflect a functional preservation.
Interestingly, the strength of the association displayed here between the OS and OD retina in levels of pathognomonic Aβ 42 was very similar to those observed between the retinae (individually and together) and their cerebral counterparts. Moreover, we found in our old ADtg mouse cohorts that the ratio between Aβ 42 and Aβ 40 , which is another key measure of AD neuropathology (Rembach et al., 2014;Schindler et al., 2019), was tightly regulated and similar across brain and retinal tissues. Importantly, the only imbalance in Aβ 42/40 ratio that was found in the OD retina of old ADtg mice was restored to levels equal to other measured CNS tissues by GA. The Aβ 42/40 ratio is thought to more accurately depict pathological burden and has been indicated in elevated protein toxicity when compared to single alloform levels alone (Kuperstein et al., 2010). This phenomenon may manifest as an OD-dominant lateralization in retinal and left-brain degenerative pathologies, which should be studied in future investigations. This result may also entail that despite differences in levels, the relative rates of Aβ 42 and Aβ 40 accumulations, affected by production and clearance, either in the retina or brain, are tightly connected, and potentially gives insight into another aspect of GA therapeutic impact.
In this study, we found that the concentrations of each Aβ alloform are substantially lower in the retina than in the brain, which is in accordance with previously published results in animal models and human donors (Alexandrov et al., 2011;Grimaldi et al., 2018;Koronyo et al., 2017;Koronyo-Hamaoui et al., 2011;Schultz et al., 2020;Shi et al., 2020). Of note, it is believed that the concentrations of Aβ 42 deemed detrimental or toxic to brain cells and can lead to LTP deficits are dose-dependent (Raskatov, 2019). Whether and how retinal concentrations translate to retinal cell toxicity and pathology warrant future investigation. Nonetheless, growing histological evidence, both in humans and animal models, shows that retinal Aβ depositions are tightly associated with local neuronal dysfunction (ERG), RNFL loss, RGC degeneration, and general tissue atrophy (Asanad et al., 2019;Grimaldi et al., 2018;Hart et al., 2016;Huang et al., 2017;Ju et al., 2013;Koronyo et al., 2012Koronyo et al., , 2017La Morgia et al., 2016;Lei et al., 2017;Ning et al., 2008). Future studies should establish the detrimental retinal Aβ concentrations, assemblies, and topographical locations in AD.
Recent studies in human patients and animal models have demonstrated that amyloid deposition in vessels reduced blood flow and was associated with various other BBB and blood-retina barrier (BRB) dysfunctions, PDGFRβ down-regulation, as well as pericyte loss (Deane et al., 2009;Kimbrough et al., 2015;Nation et al., 2019;Ramanathan et al., 2015;Schultz et al., 2018;Shi et al., 2020;Sweeney et al., 2018;Zlokovic, 2011). In addition, astrocytes and their projections, called astrocytic end feet, extend to the walls of blood vessels and are considered a pivotal component of the neurovascular unit (Liu et al., 2018;Verkhratsky & Nedergaard, 2018). Our data indicate that immunomodulation was able to substantially curb reactive astrocytosis even at such an advanced disease stage, conceivably having a positive effect on BBB integrity and Aβ clearance. Notably, our measurements of astrocytic phenotype directly and strongly correlated with severity of vascular and parenchymal Aβ pathology. Future studies should look more closely into vascular dysfunction in AD and specifically assess the integrity of BBB/BRB-related biomarkers including astrocytic end feet (e.g., aquaporin-4), to investigate potential roles in BBB/BRB and lymphatic/glymphatic Aβ clearance processes.
The combined evidence for curbing cerebral Aβ-plaque burden led us to undertake in-depth analysis of plaque subtypes to determine which type is primarily targeted by this immunomodulatory approach. Our morphological assessment of Ent Aβ deposits suggested that GA immunization in old mice was most effective in reducing the size and number of large and dense-core plaques. These plaque subtypes were previously implicated in triggering neuroinflammation and increased neurotoxicity (Busche et al., 2008;Koffie et al., 2009;Li et al., 2020;Rasmussen et al., 2017;Selkoe, 2008;Shankar et al., 2008;Wang et al., 2016). In this regard, peripherally derived myelomonocytes were shown to more effectively recognize and phagocytose plaques comprised of fibrillar Aβ forms versus soluble oligomers El Khoury et al., 2007;Koronyo et al., 2015;Lebson et al., 2010;Michaud et al., 2013;Rentsendorj et al., 2018;Zuroff et al., 2017).
Nonetheless, GA activation of macrophages also induced a more effective extracellular degradation and clearance of soluble and oligomeric Aβ 42 forms, thereby protecting neurite structures and synaptic density (Koronyo et al., 2015;Li et al., 2020). This effect on oligomers was not specifically assessed here. Our histological observations, however, highlighted that in brains of GA-immunized mice, there was a spatial organization of astrocytes and microglia/macrophages surrounding Aβ plaques that may enhance the physical barrier, better protect neuronal network, and improve Aβ clearance. The organized appearance of GFAP + astrocytes is most probably due to less activated astrocytes in between plaques. Future studies of the possible effects of GA immunotherapy on brain and retinal oligomeric Aβ burden and associated gliosis in old ADtg mice are warranted.
Among the novel molecular mechanisms of GA immunization identified here, we found that levels of brain and retinal glutamine synthetase were normalized to levels comparable to those found in healthy WT mice via immunohistochemistry and mass spectrometry.
Hence, GA-induced restoration of GS physiological levels may explain reduced excitotoxicity and preserved synapses (Ortinski et al., 2010;Son et al., 2019;Tani et al., 2014). In accordance with our MS data, dysfunctions in glutamate recycling processes were implicated in AD and drastic declines in the excitatory amino acid transporter 1 (EAAT1) of astrocytes within human tissue and astrocytic cultures exposed to Aβ oligomers have been reported (Huang et al., 2018;Zoia et al., 2004). Interestingly, GS immunoreactive area adjacent to plaques tightly associated with post-synaptic density, suggesting that restoration of GS expression to homeostatic WT levels in Aβburdened brain regions could prevent synaptic loss. The re-establishment of resting astrocyte morphology by GA identified by GFAP and GS histological IHC patterns and MS analysis further supports therapeutic efficacy at advanced disease stage.
Our quantitative MS analysis also showed that both retinal and cerebral tissues in old ADtg mice displayed significant up-regulation of several inflammatory-related proteins such as ICAM1, an adhesion molecule found on antigen-presenting cells and pericytes which facilitates transport of innate immune cells across the BBB (Martens et al., 2020;Proebstl et al., 2012). These findings are in line with previous studies reporting up-regulation of cerebral ICAM1 in a number of AD transgenic models (Ferretti et al., 2016) and, moreover, correlations with Aβ and tau pathology in human AD patients . Increased h-2 class I histocompatibility antigen (HA11) mouse MHC molecule further implies heightened immune signaling and involvement in retinas and brains of old AD model mice.
A number of markers for amyloid production and cellular processing were markedly increased in the old, late-stage ADtg mice.
Clusterin (CLU; also named Apolipoprotein J) and amyloid-β A4 protein, encompassing pre-processed APP and the various Aβ alloforms, showed extensive increases in diseased brains and retinas, corroborating our MSD and IHC analyses of Aβ. Consistent with previous results, the brain displayed a substantial increase in magnitude changes as compared to the retina Koronyo-Hamaoui et al., 2011). Finally, the lysosomal-associated membrane protein 1/2 (LAMP1/2) molecules have also been implicated in the degradation of Aβ fibrils (Barrachina et al., 2006;Gurney et al., 2018) and were found to be up-regulated in brain and retinal tissues from old ADtg mice. In response to GA immunomodulation, we revealed a preferential down-regulation of LAMP1 in the brain and LAMP2 in the retina. Overall, GA restored these molecules to levels similar to those observed in the healthy WT mice.
The results from this study substantiate the versatile therapeutic to synaptic loss and subsequent cognitive decline (Jankowsky et al., 2004;Meyer-Luehmann et al., 2008). This study was performed according to regulations of the Cedars-Sinai Medical Center Institutional Animal Care and Use Committee under an approved protocol.

| GA immunization and tissue allocation
ADtg mice from Cohort 1 were subcutaneously injected with 200 μl of either PBS or 100 μg glatiramer acetate (GA; also known as Copaxone ® , TEVA Neuroscience) in PBS. Injections were administered twice a week for the first 2 weeks, then once a week for the following 6 weeks-totaling 10 injections. A week after the completion of treatment period, WT and ADtg mice reached an average age of 22.4 months; they were euthanized with perfusion using ice-cold saline supplemented with 0.5 mM EDTA, as previously described (Koronyo et al., 2015). Brains right hemisphere were collected and fixed overnight in 2.5% paraformaldehyde (Sigma-Aldrich), then cryo-protected in 30% sucrose. The entirety of the right brain hemisphere was separated for immunohistochemistry analyses. The left posterior brain hemisphere (divided at Bregma −1 mm) was snap frozen for protein analyses and stored at −80°C.

| Retinal and brain protein isolation and processing for meso scale discovery (MSD)
Frozen mice eyeballs were placed in a cold solution of 1% Protease Inhibitor in PBS whereupon the cornea and iris were separated, and lens extracted along with the vitreous, being careful not to damage the retina. Using a pair of forceps, the retina was carefully detached from the sclera and snipped at the base of the optic nerve.
The retinal weight was recorded and tissue subsequently frozen on dry ice. Posterior brain tissues were separated, weight was recorded, and frozen on dry ice. A lysis buffer containing 0.15 M NaCl, 1 mM Ethylenediaminetetraacetic acid/EDTA, 1% Triton X, and 1% Protease Inhibitor was added to each retina and brain tissue and

| CAA scoring
For amyloid burden assessment in mouse brain vasculature, sections were stained with 6E10 (1:200; #803001; BioLegend) according to a previously described standard protocol Rentsendorj et al., 2018). Various degrees of cerebral amyloid angiopathy (CAA) in animals were defined by analyzing 6E10labeled brain sections using a scale of 0-4 (0 indicates no CAA, 4 indicates severe CAA, detailed in Figure 3f; three brain sections per animal), as previously described (Wyss-Coray & Mucke, 2002). We scored three sections (~10 vessels per image) per animal spanning a 2.25 mm 2 area of the entorhinal cortex per mouse. These scores were averaged per mouse and compared between PBS-control and GA-treated groups.

| Amyloid plaque size determination
The number and area (µm 2 ) of 6E10 + Aβ plaques were determined extra-small area = <100 µm 2 , length = <15 µm, width = <10 µm (detailed in Figure 3h). We covered a total of 2.25 mm 2 area per mouse brain, and mean values were calculated as the average of 15 images, each spanning 1.5 × 105 μm 2 area. Analyzers were blinded to the mouse groups when performing all counts.

| Synaptic quantification
For synaptic analysis and to cover the entorhinal cortex/piriform area, three of the same rectangular fields (90 μm × 70 μm) under 100× oil objective lens were precisely selected in the lateral and medial blade molecular layer (ML) of the dentate gyrus (DG), the stratum lacunosum-moleculare (SLM), the stratum radium (SR), and the stratum oriens (SO) of cornu ammonis 1 (CA1) in each condition, respectively. In addition, three of the same fields were carefully chosen in layers 2 and 3 of the entorhinal cortex. Fifteen optical sections per field, nine fields per hippocampal area, six fields per entorhinal cortex/piriform area, five fields per cingulate cortex/ retrosplenial area per section, and 300 total images per brain were analyzed. Single optical section images at 0.25 μm intervals and 3.75 μm Zeiss ApoTome high-resolution scans were performed.
Synaptic puncta number and synaptic immunoreactive (IR) area were quantified using Puncta Analyzer 2, 3, and ImageJ (NIH) macro and batch process. Average synaptic area was calculated for each condition.

| Sample preparation for mass spectrometry analysis
The protein lysates from brains and retinae were subjected to detergent removal process using detergent removal spin column (Pierce™, Thermo Scientific) as per manufacturer's instructions.
Post-detergent removal, protein amounts were quantified using BCA assay (Pierce™, Thermo Scientific) as per manufacturer's instructions and an equal amount of (50 µg) proteins per samples were subjected to reduction with 10 mM dithiothreitol (DTT) followed by alkylation with 20 mM iodoacetamide (IAA) in the dark.
Finally, the reaction was quenched with excess DTT for 15 min.
Proteins were digested at 37°C overnight with trypsin at a 1:50 ratio (enzyme to protein ratio). The digests were quenched with formic acid, and peptides were desalted using self-packed SDB-RP StageTips (Empore SPE disks) and dried in vacuum centrifuge.
Peptide concentration was determined using peptide BCA assay kit as per manufacturer's instructions.

| Peptide ion library generation
To produce a peptide ion library of relevant proteins, a pool of peptides was prepared individually for brain and retina sample peptides by mixing equal amounts of respective samples. This identified a base list of proteins from which a subset was chosen that were related to previous IHC analyses. Samples were desalted with C18 Sep Pack Light Cartridges (Waters, USA) and dried down using vacuum centrifugation. Peptides were reconstituted with 5 mM ammonia solution (pH 10.5) and loaded onto an Agilent 300 Extend C18 column (2.1 mm × 150 mm, 3.5 μm, 300 Å). Using a 1,260 quaternary HPLC system, peptides were separated using a linear gradient of 5 mM ammonia solution with 90% acetonitrile (pH 10.5) starting from 3 to 30% for 55 min at a flow rate of 300 μl/min. Peptides were separated into a total of 90 fractions that were consolidated into 17 for liquid chro-

| Data independent acquisition using SWATH mass spectrometry (SWATH-MS)
Protein quantification using SWATH-MS was performed as de-

| Mass spectrometry data analysis
IDA-MS data analysis and ion library generation are as follows.
Respective IDA-MS data files for brain and retina samples were consolidated and searched with ProteinPilot (v5.0, Sciex) using the ParagonTM algorithm in thorough mode. UniProt Mus musculus proteome database was used and searched using a tolerance of 2 missed tryptic cleavages. Fixed modifications were set for carbamidomethylation of cysteine. An Unused Score cutoff was set to 1.3 (95% confidence for identification). Resultant two data files were utilized as spectral/ion library for SWATH-MS data analysis. Mass spectrometry data for both brain and retina samples were analyzed using a threshold of 1.2 fold change. Among the remaining list of proteins present in both retina and brain, specific AD proteins of interest were further highlighted (Table 1), and only if they were down-regulated or up-regulated significantly in either brain or retina of ADtg mice relative to WTs and restored either partially or fully to WT levels by GA treatment. A complete list of significantly identified proteins for brains (Table S1) and retina (Table S2) display fold change and comparisons between WT naïve and PBS control, as well as GA-immunized and PBS control.

| SWATH-MS data analysis
Both Ion library and SWATH-MS data files were imported into PeakView software 2.1 using the SWATH MicroApp 2.0 (SCIEX), and data were extracted using the following parameters: Top 6 most intense fragments of each peptide were extracted from the SWATH data sets (75 ppm mass tolerance, 5 min retention time window). Modified and shared peptides were excluded from quantification. After data processing, peptides (max 100 peptides per protein) with confidence >99% and FDR <1% (based on chromatographic feature after fragment extraction) were used for the quantitation. Cumulative protein areas from extracted ion chromatograms were exported to Excel for further analysis.
The protein peaks areas were normalized to the total protein peak area of the respective sample and subjected to one-sample t tests to compare relative protein peak areas between the respective sample groups. t test p-value smaller than 0.05 and fold change ±1.2 was highlighted as differentially expressed proteins.
Two approaches were considered for determining differential expression: ANOVA on the log-transformed normalized protein peak areas of all samples and t test pairwise comparisons of pairs of specific samples. For the analysis of variance, proteins were deemed to be differentially expressed if the ANOVA p-value was less than 0.05 and the maximum protein fold change exceeded 1.2.

| Statistical analysis
Data were analyzed using GraphPad Prism 6.01 (GraphPad Software). A two-tailed unpaired Student's t test was applied for analytic comparisons between two groups. Comparison of three or more groups was performed using one-way ANOVA with Tukey's multiple comparison post-test of paired groups. Analysis of two independent variables was performed using two-way ANOVA with Sidak's post-test. Correlation analysis was performed using Prism Pearson's tests. Results are expressed as means ± standard deviations (SDs) or means ± standard errors of the mean (SEMs) as indicated. A p-value <0.05 was considered significant, and a p-value <0.10 was regarded as a trend.