CT1812 biomarker signature from a meta-analysis of CSF proteomic findings from two Phase 2 clinical trials in Alzheimer's disease

Abstract

INTRODUCTION

CT1812 is in clinical development for the treatment of Alzheimer's disease (AD). Cerebrospinal fluid (CSF) exploratory proteomics was employed to identify pharmacodynamic biomarkers of CT1812 in mild to moderate AD from two independent clinical trials.

METHODS

Unbiased analysis of tandem-mass tag mass spectrometry (TMT-MS) quantitative proteomics, pathway analysis and correlation analyses with volumetric magnetic resonance imaging (vMRI) were performed for the SPARC cohort (NCT03493282). Comparative analyses and a meta-analysis with the interim SHINE cohort (NCT03507790; SHINE-A) followed by network analysis (weighted gene co-expression network analysis [WGCNA]) were used to understand the biological impact of CT1812.

RESULTS

CT1812 pharmacodynamic biomarkers and biological pathways were identified that replicate across two clinical cohorts. The meta-analysis revealed novel candidate biomarkers linked to S2R biology and AD, and network analysis revealed treatment-associated networks driven by S2R. 

DISCUSSION

Early clinical validation of CT1812 candidate biomarkers replicating in independent cohorts strengthens the understanding of the biological impact of CT1812 in patients with AD, and supports CT1812's synaptoprotective mechanism of action and its continued clinical development.

Highlights

• This exploratory proteomics study identified candidate biomarkers of CT1812 in SPARC (NCT03493282)
• Comparative analyses identified biomarkers replicating across trials/cohorts
• Two independent Ph2 trial cohorts (SPARC and interim SHINE [NCT03507790; SHINE-A]) were used in a meta-analysis
• Amyloid beta (Aβ) & synaptic biology impacted by CT1812 and volumetric magnetic resonance imaging (vMRI) treatment-related correlates emerge
• Network analyses revealed sigma-2 receptor (S2R)-interacting proteins that may be “drivers” of changes

1 BACKGROUND

Alzheimer's disease (AD) is the most prevalent neurodegenerative disease worldwide and affects more than 6.5 million people in the United States alone. Unfortunately, disease-modifying therapeutic options available for patients are limited. Approved immunotherapy approaches targeting amyloid beta (Aβ) slow disease progression by about 25% over 18 months, but the disease continues to progress despite lowering of amyloid plaque to amyloid positron emission tomography (PET) Centiloid-normal levels.^{1, 2} As such, there remains substantial need to further slow the disease through the development of novel therapeutics beyond or in combination with existing therapies.

CT1812 is a first-in-class sigma-2 receptor (S2R) modulator with a novel, disease-modifying, synaptoprotective mechanism of action for AD. S2R modulators have the potential to act directly at the synapse by blocking Aβ oligomer binding and oligomer-related synaptotoxicity while restoring disrupted neuronal functioning, which a body of preclinical and early clinical data supports.^3-5 The S2R has a number of known protein–protein interactions and impacts the oligomer-binding receptor complex (Figure S1). Targeting S2R is a unique approach that has the potential to be utilized alone or in complement with currently available therapies.^{3, 5} To advance understanding of AD pathophysiology and to support the development of therapeutic agents, research into biomarkers has expanded rapidly in recent years. This development has been facilitated by academic research, the pharmaceutical/ biotechnology industry, disease-focused foundations, and collaborative efforts such as the Alzheimer's Disease Neuroimaging Initiative (ADNI), the subject of this special issue. With a primary goal of developing and validating imaging, cerebrospinal fluid (CSF), or genetic biomarkers for AD, the ADNI was initially designed as a study to gather data and standardize analyses across multiple sites.⁶ Twenty years later, the field of AD biomarkers has advanced sufficiently to have supported accelerated U.S. Food and Drug Administration (FDA) approval of the first disease-modifying therapy for AD, aducanumab (Aduhelm), based largely upon Aβ plaque biomarker evidence of target engagement,⁷ followed by the first fully approved disease-modifying therapy for AD, lecanemab.² These breakthroughs helped pave the way for future drug development in AD, with the newest FDA draft guidance further emphasizing the role of biomarkers as supportive data for drug approvals.⁸

The identification of novel AD biomarkers through large-scale discovery proteomics has accelerated the field dramatically. Unbiased discovery proteomics using tandem-mass tag (TMT) mass spectrometry (MS) can measure levels of more than 2000 proteins from a single CSF sample and enable the assessment of longitudinal changes in abundance across all proteins detected.⁹ This technique enables the procurement of rich data sets from which novel biomarkers linked to several key pathophysiological processes perturbed in the disease^{9, 10} and modulated by interventional therapeutics can be discovered¹¹ and potentially used to predict treatment response.¹²

SPARC (“Synaptic Protection for Alzheimer's Recovery of Cognition;” COG0105; NCT03493282) is a completed single-center, randomized, double-blind, placebo-controlled Phase 2 clinical trial of CT1812 in adults with mild to moderate AD.¹³ SHINE (“Synaptic Health and Improvement of Neurological Fuction with Elayta;” COG0201; NCT03507790) is a completed randomized, double-blind, placebo-controlled Phase 2 clinical trial of CT1812 in adults with mild to moderate AD. This article includes the first report of CSF biomarker proteomic findings from the SPARC cohort, in which participants were dosed with CT1812 or placebo for 6 months, and from a meta-analysis comprising SPARC and the first 24 participants in SHINE (SHINE-A), also dosed with CT1812 or placebo for 6 months. In addition, SHINE-A proteomic data sets were used to corroborate proteomic findings from SPARC to identify biomarkers that replicate across independent clinical cohorts treated with CT1812.

In this exploratory biomarker study we analyzed CSF from patients with AD who were treated with CT1812 or placebo for 6 months. Using an unbiased proteomics approach, in the SPARC cohort we identify CT1812 biomarkers that enrich pathways related to synapses, amyloid biology, and immune response, thereby recapitulating biological processes identified in the interim SHINE cohort.¹⁴ From the SPARC CT1812-treated participants, a subset of TMT-MS–detected CSF proteins were found to be significantly correlated with change in brain composite region volumetric magnetic resonance imaging (vMRI), which showed a trend toward tissue preservation in a composite brain region in participants treated with CT1812.¹³ A meta-analysis of the CSF proteomes from SPARC and SHINE-A cohorts was performed to increase analytic power. This analysis identified CT1812-impacted biomarkers—including AD biomarkers clusterin (CLU) and spondin 1 (SPON1)—that replicate in analyses of the independent cohorts, and identify new candidate biomarkers. Biomarkers related to CT1812's receptor, S2R, as well as co-receptors were also impacted. Weighted gene co-expression network analysis (WGCNA) of the meta-analysis differential expression (change from baseline, CT1812 vs placebo) identified networks of proteins that correlate with CT1812 treatment, each of which enriches biological pathways related to synapses, amyloid biology, lipid homeostasis, or immune response. These results notably highlight key S2R components as hub proteins that may reflect networks altered by target engagement of CT1812 through S2R.

2 METHODS

2.1 Cohorts used for proteomic biomarker assessments

Two cohorts from two independent trials were used in proteomic analyses: SPARC and SHINE-A, each analyzed independently and also analyzed with cohorts combined (SPARC + SHINE-A) for a meta-analysis. Other findings including safety from each trial/cohort, SPARC (NCT03493282)¹³ and SHINE-A (NCT03507790)¹⁴ have been reported previously. For comparison of placebo and drug treatment arms within each cohort and across both cohorts, participant baseline demographics and characteristics from each trial, including sex, race, body mass index (BMI), apolipoprotein E (APOE) genotype, and baseline Mini-Mental Status Examination (MMSE), are listed in Table S1.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed the literature pertaining to Alzheimer's disease (AD), biomarkers for AD and clinical development for AD, proteomics analyses and mass spectrometry (MS) techniques, and mechanistic findings related to proteins of interest that we have identified in our biomarker research. The use of recently improved proteomics techniques facilitates unbiased, quantitative assessment of AD patient cerebrospinal fluid (CSF) to enable biomarker nomination. 
2. Interpretation: The exploratory proteomic biomarker findings of the clinical trial SPARC (NCT03493282) in conjunction with the subsequent comparative analyses using an independent clinical cohort, interim SHINE (NCT03507790; SHINE-A), provide early clinical validation of biomarkers impacted by CT1812. The meta-analysis combining both cohorts increases the analytical power of our findings.
3. Future directions: Our results support candidate biomarker nomination for CT1812 and the continued development of CT1812. Future, larger trials may further solidify the evidence supporting CT1812's impact in AD and mechanism of action.

SPARC (COG0105; NCT03493282) was a single-center, randomized, double-blind, placebo-controlled parallel group Phase 2 clinical trial of CT1812 in adults with mild to moderate AD.¹³ Men and women 50–85 years of age with mild to moderate dementia due to AD (MMSE score 18–26) were screened for eligibility. AD pathogenesis was confirmed by amyloid PET imaging or by CSF biomarkers measured at the screening visit. Participants (N = 23) were randomized 1:1:1 to receive placebo or 100 or 300 mg CT1812, orally, once daily for 6 months, followed by an optional double-blind extension treatment period of another 6 months. For biomarker assessments, lumbar punctures to collect CSF were conducted at baseline and at the end of the study (6 months). Timepoint-matched biofluids were available from N = 18 participants.

SHINE (COG0201; NCT03507790) was an international, multi-center, randomized, double-blind, placebo-controlled parallel group 36-week Phase 2 clinical trial of CT1812 in adults with mild to moderate AD.¹⁴ SHINE was designed to enroll ≈144 participants; men and women 50–85 years of age with mild to moderate dementia due to AD (MMSE score 18–26) were screened for eligibility. AD pathogenesis was confirmed by amyloid PET imaging or by CSF biomarkers measured at the screening visit. Participants were randomized 1:1:1 to receive placebo or 100 or 300 mg CT1812, orally, once daily for 6 months. To assess biomarkers, lumbar punctures to collect CSF were conducted at baseline and at the end of the study (6 months). An interim exploratory analysis was previously conducted¹⁴ on the first 24 participants evaluated as part A of the study (SHINE-A). Timepoint-matched biofluids were available from N = 20 participants.

All biomarker analyses carried out for this study were exploratory and for the purpose of elucidating pharmacodynamic changes of CT1812. All analyses are using pooled CT1812 drug-treatment groups. Only participants who were actively taking their treatment, as indicated by bioanalysis of drug exposure levels at end of study (herein referred to as treatment-compliant participants), were included in the analysis (SPARC N = 17; SHINE-A N = 18).

2.2 Proteomic biomarker measurements

2.2.1 SPARC

TMT-MS proteomics measurements were performed on CSF samples from treatment-compliant SPARC participants (N = 17 participants: placebo, n = 6; 100 mg CT1812, n = 5; 300 mg CT1812, n = 6) for whom both baseline and end-of-study samples were collected (N = 34 samples; N = 17 × 2 visits [baseline, end of study]). Samples were processed for TMT-MS, which was performed as described previously¹⁴ and as described in brief in the Supplemental Methods. Using TMT-MS, 2760 proteins were detected reliably across all SPARC CSF samples using the 50% detection exclusion criterion, and unbiased analyses were performed as described previously¹⁵ and as described in the analysis pipeline below.

2.2.2 SHINE-A

TMT-MS proteomics measurements for SHINE-A were previously performed and analyzed.¹⁴ CSF samples from treatment-compliant SHINE-A participants (N = 18 participants: placebo, n = 7; 100 mg CT1812, n = 4; 300 mg CT1812, n = 7) for whom both baseline and end-of-study samples were collected (N = 36 samples: placebo, n = 18 × 2 visits [baseline, end of study]) were processed and analyzed using TMT-MS followed by unbiased quantification.¹⁴ Using TMT-MS, 2161 proteins were detected reliably across all SHINE-A CSF samples using the 50% detection exclusion criterion. TMT-MS differential abundance was performed to assess treatment effects (CT1812 vs placebo) via analysis of variance (ANOVA) following determination of the log-transformed (log2(ratio)) change from baseline for each participant in which 122 proteins were found to be altered.¹⁴ Uniprot identifiers for all differentially abundant proteins meeting each statistical threshold assessed were visualized on volcano plots.

2.3 Benchmarking of proteomes from SPARC AD cohort to SHINE-A cohort and to AD reference population

To determine whether the SPARC AD participant sample CSF proteome resembled the SHINE-A participant sample CSF proteome¹⁴ multidimensional scaling (MDS) plots were generated to illustrate inter-sample variability across cohorts on baseline samples. To determine whether SPARC and SHINE-A AD cohorts were typical of AD, baseline CSF samples from each study and from a previously well-characterized AD cohort¹⁶ were compared. Pooled AD biomarker-positive (AD) and age-matched, non-demented biomarker-negative (control) CSF reference standards were generated from the Goizueta Alzheimer's Disease Research Center (ADRC) at Emory.^{14, 16} Each pool comprised CSF from multiple individuals (37 healthy controls and 46 patients with AD) combined into one pool of control benchmark CSF and one pool of AD benchmark CSF. One technical replicate from each pool (AD and control) was placed on each (9) of the TMT-MS plexes (4 SPARC CSF plexes and 5 SHINE-A CSF plexes), and the abundances of AD-related proteins were compared. Batch-to-batch variation was very low.

2.4 Identification of pharmacodynamic biomarkers of CT1812

2.4.1 SPARC differential abundance

The log2 ratio change from baseline for each participant was first calculated. Pharmacodynamic biomarkers of CT1812 were determined via calculating TMT-MS differential abundance (log2 fold change, CT1812 vs placebo) and assessed via ANOVA. Differentially expressed proteins were identified based on the unadjusted p-value threshold of p ≤ 0.05. This analysis was also performed for AD versus control CSF.

2.4.2 Core AD biomarkers and validation of TMT-MS in SPARC proteomic analyses

Validation of TMT-MS for the SPARC analysis was performed as described previously for SHINE-A,¹⁴ wherein key AD core proteins related to AD pathology were measured from SPARC participant CSF samples using clinically validated assays.¹⁷ Neurogranin, neurofilament light (NfL), and YKL-40 (or chitinase-3-like protein 1 [CHI3L1]) were measured by enzyme-linked immunosorbent assay (ELISA); synaptotagmin (Syt1) by immunoprecipitation-liquid chromatography-parallel reaction monitoring MS (IP-LC-PRM-MS), and total tau (t-tau) by Lumipulse (herein referred to as qNrgn, qNfl, qYKL-40, qSyt1, qTau).

2.4.3 SPARC/SHINE-A meta-analysis differential abundance

A meta-analysis was performed to increase the statistical power to detect biomarkers that may be significantly affected by CT1812. The SPARC and previously described SHINE-A¹⁴ data sets were combined for analysis (N = 35 treatment-compliant participants, n = 70 CSF samples). To remove batch effects and minimize the potential for introducing artifacts, the combined data set was normalized via Tunable Approach for Median Polish of Ratio (TAMPOR),¹² and the raw abundance measurements from Proteome Discoverer were collated and searched. The combined search resulted in 2102 proteins that passed the 50% missingness detection exclusion criterion. Differential abundance to assess treatment effects was assessed as described for SPARC.

2.5 Proteomic analysis pipeline for SPARC and meta-analysis

2.5.1 Brain mapping

The identified differentially expressed proteins were mapped to one of 44 previously generated AD brain co-expression network modules built from 516 brain samples with healthy individuals and asymptomatic and symptomatic AD patients to understand how CSF biomarkers altered by CT1812 might impact brain networks disrupted in AD¹⁰ using previously described methods.¹⁸

2.5.2 Pathway analyses

Proteins meeting the statistical threshold for each comparison were assessed using STRING (version 12)¹⁹ and Metacore (version indicated in each figure legend). For STRING, only Uniprot identifiers with annotated protein names were used for pathway analysis. For Metacore, both protein names and Uniprot identifiers were used for analyses. The STRING protein–protein interaction (PPI) network for each condition was exported and all pathway terms (e.g., for Kyoto Encyclopedia of Genes and Genomes [KEGG], Gene Ontology [GO] Biological Processes, and Reactome) were ranked according to strength or false discovery rate (FDR) for interpretation. For visualization of the PPI network, low, medium, and high confidence was used as appropriate (confidence thresholds indicated for each figure) for visualization purposes. Unconnected nodes were excluded from the PPI maps.

2.6 SPARC proteome correlation analysis with volumetric MRI

T1-weighted MRI was performed at baseline, 12 weeks, and 24 weeks to define regions of interest (ROIs) and to assess brain volumetric changes, as described previously.¹³ To elucidate the effects of CT1812 on imaging outcomes, a composite ROI of AD-affected brain regions was defined and change from baseline volume (mL) was assessed.¹³ Pearson correlation analysis was performed only in CT1812-treated participants across all proteins detected via TMT-MS against the quantitative values of brain volume. Proteins meeting a threshold of r ≥ |0.5| and p ≤ 0.05 were considered for STRING analyses. Correlated proteins that were also differentially abundant in participants treated with CT1812 versus placebo (log2 fold change; p ≤ 0.1) were analyzed using STRING.

2.7 Comparative analysis of two independent cohorts: SPARC and SHINE-A

To determine whether biomarkers replicate from SPARC as part of a clinical validation process, a comparative analysis was performed using the SHINE-A proteomic data set to assess biomarker replication across trials. Proteins identified from each differential abundance analysis of SPARC herein and SHINE-A¹⁴ were compared to each other. Uniprot protein identification lists were used for comparative analysis (p ≤ 0.1: 254 proteins for SHINE-A, 603 proteins for SPARC; p ≤ 0.05: 122 proteins for SHINE-A, 256 proteins for SPARC). A statistical threshold of p ≤ 0.1 was used due to the small cohort size of each trial with the rationale that if p ≤ 0.1 is achieved in both trials and the directionality of change is the same; this may be indicative of a pharmacodynamic effect of CT1812 that may achieve statistical significance of p ≤ 0.05 in a larger cohort.

2.8 Weighted gene co-expression network analysis

To further explore the mechanism of action of CT1812, WGCNA was performed using R Bioconductor (version 1.69) to create a weighted protein co-expression network from all proteome change from baseline values from all participants in SPARC and SHINE-A used in the meta-analysis. The log2 abundance change from baseline data for n = 2102 proteins from n = 70 case–sample matrix was used to generate the network. The included data had undergone reporter purity, batch effect, and other covariate corrections. The co-expression network was created using a bi-weight mid-correlation (bicor) measure: a soft threshold power of 13 to ensure optimal compliance with scale-free topology requirements by balancing a tradeoff between cleaning up spurious correlations due to chance (particularly important when total samples in the network are low) and maintaining sensitivity of the clustering to still be able to pick up correlations in as much of the data as possible.

The WGCNA::blockwiseModules() function was used with the following settings for the consensus network: soft threshold power = 13.0, deepSplit = 2, minimum module size of 15, merge cut height of 0.07, mean topological overlap matrix (TOM) denominator, a signed network with partitioning about medoids (PAM) respecting the dendrogram and a reassignment threshold of p ≤ 0.05, with clustering completed within a single block. Construction of the network yielded 21 modules consisting of 16 or more proteins. To avoid aberrant assignments to modules, a post hoc cleanup procedure was applied by first setting a module eigengene (kME) threshold of 0.30 for module membership followed by reassignment of (1) any protein previously unassigned to a module and (2) proteins with intramodular kME >0.10 but below the maximum kME of the protein's correlation to any other modules. This cleanup and reassignment step was done iteratively to ensure that each protein was assigned to the modules corresponding to the protein's maximum kME after reassignment. Then, MEs and the signed kME table were recalculated with the WGCNA::moduleEigengenes() and WGCNA::signedKME() functions, respectively. Finally, the kME table individual protein reassignment process was repeated if additional corrections could be made, up to a total of 30 iterations. For the consensus network, this required nine iterations until resolution, which increased the module size of the smallest module (M21) in the network to 29, and decreased gray (unassigned) protein count for the network from 569 (27%) to 149 (7%).

Module eigenproteins were calculated and used for various analyses as the most representative abundance values for a module equivalent to the module's first principal component. Module–trait correlation (Pearson's correlation coefficient (r)) analyses were performed using WGCNA::networkScreening() function to assess the relationship of each module to treatment group (CT1812 or placebo), participant demographic traits (sex, BMI, race, APOE ε4 status), neurogranin,^{17, 20} and baseline MMSE score.

2.9 Gene ontology and cell-type marker enrichment analyses

To characterize WGCNA networks with co-expressed proteins based on gene ontology (GO) annotation, we used GO Elite (version 1.2.5) as published previously,²¹ with pruned output visualized using an in-house R script. Cell-type enrichment was also investigated as published previously from a culmination of two data sets (the Sharma et al. list and proteins from the Zhang et al. list).^{21, 22} In brief, Fisher exact tests were performed to assess module enrichment for five cell types: endothelial cells, microglia, astrocytes, neurons, and oligodendrocytes followed by correction for FDR (Benjamini–Hochberg).

3 RESULTS

3.1 Identification and characterization of CT1812 biomarkers in SPARC participant CSF

In validation of the TMT-MS method as quantitative, a high congruency was observed between the levels of core AD biomarkers detected using TMT-MS and using independent quantitative methods,²⁰ depicted via the Pearson correlation coefficient (r) heatmap (CHI3L1/YKL40, r = 0.82; NEFL (neurofilament light), r = 0.73; NRGN (neurogranin), r = 0.91; MAPT (microtubule associated protein tau), r = 0.82; SYT1, r = 0.81; p ≤ 0.001) (Figure S2A); scatterplots of correlations for each individual sample of validated assay-measured value and corresponding MS-measured value are shown (Figure S2B).

Differentially abundant proteins (CT1812 vs placebo log2 fold change; p ≤ 0.05) identified in the SPARC AD patient CSF are illustrated via volcano plot (Figure 1A; 87 increased and 169 decreased, 256 total). Select proteins that were altered in CT1812 versus placebo include those involved in amyloid biology or lipid binding/lipid handling, such as apolipoproteins apolipoprotein C-IV (Apo-CIV [APOC4]) and apolipoprotein J (apoJ; clusterin [CLU]), SPON1, low-density lipoprotein receptor-related protein 1 (LRP1), and Niemann-Pick protein type C1 (NPC1), an S2R-interacting protein.²³ Other impacted proteins include the neurodegeneration-associated G protein-coupled receptor 37 (GPR37), carboxypeptidase E (CPE), the macrophage-related interferon-gamma receptor 1 (IFNGR1) and complement C1q tumor necrosis factor-related (C1QTNF4), the AD biomarker tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein beta (YWHAB), tyrosine-protein kinase receptor (TIE1), immunoglobulin lambda variable 1-44 (IGLV1-44), and arginase-1 (ARG1) (Figure 1A).

FIGURE 1

SPARC CSF differential expression analysis identified significantly abundant proteins in CSF from CT1812 versus placebo. (A) For cerebrospinal fluid (CSF) proteomic analysis of the SPARC (NCT03493282) cohort (N = 17 participants), differential abundance of change from baseline levels of protein measured via tandem mass tag mass spectrometry (TMT-MS) was assessed via analysis of variance (ANOVA) (CT1812 vs placebo; p ≤ 0.05). Log2 abundance and significance for each protein detected are shown on the volcano plot (each colored dot corresponds to a protein; significantly decreased proteins are shown in green, significantly increased in red). A total of 256 proteins (169 decreased, 87 increased) were found to be significantly differentially abundant. Alzheimer's disease (AD) priority biomarkers are shown in yellow. Proteins of interest are labeled in royal blue: CLU, clusterin; CPE, carboxypeptidase E; GPR37, G protein–coupled receptor 37; LRP1, low-density lipoprotein receptor-related protein 1; NPC1, Niemann–Pick disease type C1; IGLV1-44, immunoglobulin lambda variable 1-44; SPON1, spondin 1; YWHAB, tyrosine 3-monooxygenase/tryptophan 5-monooxygenase activation protein beta; INFGR1, interferon-gamma receptor; C1QTNF4, C1q tumor necrosis factor-related protein 4; TIE1, tyrosine kinase with immunoglobulin-like and EGF-like domains 1; APOC4, apolipoprotein C4; ARG1, arginase 1. (B) The significantly (p ≤ 0.05) differentially abundant proteins were mapped to previously established AD co-expression modules,¹⁰ and the top 15 modules most represented are listed. Modules of interest are highlighted in royal blue in the legend. (C–D) Pathway analyses were performed using the 256 significant differentially abundant proteins in CT1812 versus placebo (p ≤ 0.05). (C) STRING (v12.0) analysis identified significant protein–protein interactions (p = 7.3e⁻¹²; confidence threshold 0.60 for visualization purposes). Proteins and clusters of interest indicated in royal blue. (D) Metacore pathway analysis (v23.4.71500) identified significantly enriched pathways (p ≤ 0.05), which include immune response and amyloid-related processes. Pathways of interest are indicated in royal blue. (E) Significantly differentially abundant proteins in SPARC and SHINE-A (interim SHINE; NCT03507790) were compared and 55 common proteins were altered (p ≤ 0.1) across both cohorts; 6 of these were significant at a threshold of p ≤ 0.05 (Table S2).

The 256 differentially abundant proteins were mapped to previously determined AD co-expression brain network modules¹⁰ (Figure 1B; top 15 modules of 35 shown). The CT1812-altered proteins were the most highly represented brain network modules (1) “M4 Synapse/ Neuron,” (2) “M27 Extracellular Matrix,” (3) “M26 Complement/Acute Phase,” (4) “M7 MAPK/Metabolism,” and (5) “M3 Oligo/Myelination.”

STRING analysis was used to visualize the interconnectivity of the 256 proteins (p ≤ 0.05) via a PPI map (Figure 1C). The PPI map demonstrates a high amount of interconnectivity (PPI network significance: p = 7.3e⁻¹²), with lines illustrating known relationships between proteins. Pathways relating to the immune system, vesicles, and signal receptor binding were prominent in the network. Metacore pathway analysis of the 256 significantly altered proteins verified these results from STRING: pathways impacted by CT1812 were related to protein folding (“Protein folding and maturation: angiotensin system maturation” and “Protein folding and maturation: Regulation of amyloid precursor protein processing”) and immune response (“Antigen presentation by MHC (major histocompatibility complex) class I: cross-presentation,” “Antigen presentation by MHC class II,” “Antigen presentation by MHC class I, classical pathway,” and “IL-11 (interleukin 11) signalling via JAK/STAT (Janus kinase/signal transducers and activators of transcripion)”) (Figure 1D). Metacore additionally identified enrichment in “Signal transduction: Angiotensin II/AGTR1 signaling via NOTCH, Beta-catenin and NF-kB (nuclear factor kB) pathways” (Figure 1D).

3.2 CT1812 impacted CSF proteins associated with a favorable change in vMRI

In participants treated with CT1812, a total of 177 TMT-MS–detected CSF proteins were found to be significantly correlated with vMRI for the composite of brain regions¹³ (r ≥ |0.50| and p ≤ 0.05). STRING PPI map demonstrates high interconnectivity (p < 1.0e⁻¹⁶), with AD biomarkers amyloid beta precursor protein (APP), amyloid beta precursor-like protein 2 (APLP2), and CLU among the connected proteins (Figure S3A). Other highly interconnected hub proteins included immune response protein complement factor H (CFH) and synaptic protein neurexin 2 (NRXN2) (Figure S3A). Of note, 45 correlates were differentially abundant (p ≤ 0.1) with CT1812 (Figure S3B), and STRING mapping illustrates the interconnectivity (PPI score p < 1.0e⁻¹⁶) for all connected proteins with APP, APLP2, and NRXN1 as highly-connected nodes. STRING GO Biological Pathways showed enrichment of identified correlates to several synaptic processes and functions (Figure S3C).

3.3 Replication across SPARC and SHINE-A cohorts

For early clinical validation of biomarkers identified in SPARC, the SHINE-A cohort was used to assess replication of findings across independent trials. Given the relatively small size of each cohort, a less stringent p-value (p ≤ 0.1) was first applied to identify pharmacodynamic effects that might move similarly in both trials but not meet statistical significance (p ≤ 0.05). Using this criterion (p ≤ 0.1), 55 proteins were identified as altered in both trials (Figure 1E, Venn diagram). Upon closer examination to assess effect size (log2 fold change) and directionality of change (up, down) (Table S2), 51 of the 55 proteins were found to be altered in the same direction in CSF in CT1812-treated participants across both cohorts, with the observation that effect sizes were also similar. Only 4 of 55 proteins showed movement in opposing directions across the two cohorts. Five of the 55 proteins achieved statistical significance (p ≤ 0.05) in both cohorts: GPR37, CLU, CPE, and SPON1 showed a decrease, IGLV1-44 an increase, and NPC1 moved in opposing directions across cohorts in CSF from CT1812-treated participants (Table S2).

3.4 Characterization of cohorts used in meta-analysis

Proteomic analysis of the combined SPARC and SHINE-A cohorts (baseline and end of study) identified 2102 detected proteins common to SPARC and SHINE-A (Figure 2A); these were used for downstream analyses. At a gross level, multidimensional scaling MDS to illustrate inter-sample variability shows sufficient overlap of the baseline proteomes from the two cohorts (Figure 2B). The SPARC and SHINE-A trial participants had similar baseline characteristics (Table S1). Cohorts were well-matched (SPARC vs SHINE-A) in age (70.0 ± 8.8 vs 71.7 ± 7.7 years), sex (12 M/ 11F vs 9 M/ 15F), race (96% White vs 95.8% White), BMI (27.1 ± 4.5 vs 26.2 ± 5.0), and baseline MMSE score (22.6 ± 1.98 vs 21.1 ± 3.57). The SPARC cohort had a higher APOE ε4 carrier percentage than did SHINE-A (78% carrier vs 58.3% carrier) (Table S1).

FIGURE 2

SHINE-A/SPARC meta-analysis cohort characterization. (A) Cerebrospinal fluid (CSF) proteomic analysis from each cohort (SPARC [NCT03493282], N = 17 participants; SHINE-A [interim SHINE; NCT03507790], N = 18 participants) was combined for a total of N = 35 participants. (B) Multidimensional scaling (MDS) plots of baseline inter-sample variance are shown for both cohorts. (C) Abundances of Alzheimer's disease (AD)-related proteins were compared between healthy control (CT) and AD control reference CSF proteomes (from the Emory Alzheimer’s Disease Research Center [ADRC]) and the SHINE-A and SPARC baseline CSF proteomes. AD and control are pooled samples run as technical replicates; SPARC and SHINE are biological replicates from individual patients. Selected proteins compared: NRGN, neurogranin; COL6A1, collagen type VI alpha chain; MAPT, microtubule-associated protein tau; CLU, clusterin; TF, transferrin; and HLA-DRB1, human leukocyte antigen class II histocompatibility D related beta chain. Data are presented as mean log2 abundance (line within box) ± standard error of the mean (SEM) (box height) and ± SD (error bars).

To benchmark the cohorts at baseline against a known AD cohort, baseline proteomes were compared to those of pooled Emory control (healthy) individuals and pooled Emory AD patient reference population CSF samples. Baseline levels of proteins disrupted in AD versus control (NRGN, collagen type VI alpha chain [COL6A1], MAPT, CLU, transferrin [TF], and human leukocyte antigen class II histocompatibility D related beta chain [HLA-DRB1]) were plotted, and proteomes of SPARC and SHINE cohorts were compared to each other and to those of pooled Emory control (healthy) individuals and pooled Emory AD patient reference population CSF samples (Figure 2C). The means of the SPARC and SHINE-A baseline CSF biomarker abundances for each protein were congruent with those from the benchmark AD biomarker–positive pool, and distinct from the healthy control AD biomarker–negative pool (Figure 2C).

3.5 Meta-analysis of SPARC and SHINE-A cohort proteomes supports a role of CT1812 at synapses and normalization of key AD biomarkers

Differentially abundant proteins (CT1812 vs placebo log2 fold change; p ≤ 0.05) identified in the SPARC/SHINE-A meta-analysis are illustrated via volcano plot (Figure 3A; 137 increased and 165 decreased, 302 total). Select proteins that were altered in CT1812 versus placebo include AD biomarkers (CLU, APP, APLP2, and sphingosin-1-phosphate phosphatase 1 [SPP1]) and synaptic proteins (synaptotagmin-7 [SYT7], NRXN1 and NRXN2, and integrin beta-2 [ITGB2]). Other proteins of interest affected included CPE, SPON1, GPR37, multiple epidermal growth factor-like domains (MEGF8), S2R-interacting prion protein (PRNP;²⁴), TF, IGLV1-44, YWHAB, and polyribonucleotide 5′-hydroxyl-kinase (CLP1) (Figure 3A; see discussion).

FIGURE 3

SHINE-A/SPARC meta-analysis brain network mapping and pathway analyses support role of CT1812 at synapses and immune function. As in Figure 2, cerebrospinal fluid (CSF) proteomic analysis from each cohort (SPARC [NCT03493282], N = 17 participants; SHINE-A [interim SHINE; NCT03507790], N = 18 participants) was combined for a total of N = 35 participants. (A) Volcano plot illustrates that 302 proteins were significantly altered (p ≤ 0.05) in CT1812 versus placebo CSF (165 downregulated, red; 137 upregulated, green; Alzheimer's disease (AD) priority biomarkers, yellow). Proteins of interest are indicated in royal blue font: APP, amyloid precursor protein; SYT7, synaptotagmin 7; PRNP, prion protein; ITGB2, beta 2 integrin; SPP1, secreted phosphoprotein 1; MEGF8, multiple epidermal growth factor-like domains 8; NRXN2, neurexin 2; APLP2, amyloid precursor-like protein 2; TF, transferrin; CLP1, cleavage factor polyribonucleotide kinase subunit 1. (B) Differentially abundant proteins (CT1812 vs placebo, p ≤ 0.05) were mapped to previously generated AD brain co-expression network modules¹⁰ to understand how CSF biomarkers altered by CT1812 might impact brain networks disrupted in AD. Top 15 modules represented are indicated. Modules of interest are labeled in royal blue in the legend. (C) STRING protein–protein interaction (PPI) map of the significantly (p ≤ 0.05) differentially abundant proteins (confidence threshold 0.600 for visualization purposes, with proteins of interest circled in royal blue). Proteins color coded according to gene ontology (GO) terms of interest and clusters of interest are labeled in royal blue. (D) Pathway analysis using Metacore (v22.4.71100) mapping shows synapse, inflammatory, and amyloid pathways to be most significantly altered with CT1812 treatment when compared to placebo-treated controls (p ≤ 0.05; pathways of interest highlighted in royal blue). (E) Abundances of proteins altered in SHINE-A and SPARC CT1812 and placebo were compared to Emory AD-control (CT) reference standards. Data are presented as mean log2 abundance (line within box) ± standard error of the mean (SEM) (box height) and ± standard deviation (SD) (error bars), with p-value indicated for each protein (CT1812 vs placebo). AD and control are pooled samples run as technical replicates; SPARC and SHINE are biological replicates from individual patients. Proteins shown: TF, transferrin; HTRA1, HtrA serine peptidase 1; CLU, clusterin; SEMA4B, semaphorin 4B; SPON1, spondin 1; CALB2, calbindin 2; APOL1, apolipoprotein L1; OLFML1, olfactomedin-like 1.

The 302 differentially abundant proteins identified in the meta-analysis were mapped to AD co-expression brain network modules (Figure 3B; top 15 of 33 modules are shown). The CT1812-altered proteins were most highly represented brain network modules (1) “M40 Antigen Binding,” (2) “M4 Synapse/Neuron,” (3) “M26 Complement/Acute Phase,” (4) “M42 Matrisome,” and (5) “M1 Synapse/Neuron” modules.

STRING analysis was used to visualize the interconnectivity of the 302 proteins (p ≤ 0.05) via the PPI map (Figure 3C). The PPI map demonstrates high interconnectivity (PPI significance, p ≤ 1.0e⁻¹⁶) with key AD proteins APP, APOE and CLU, Aβ oligomer receptor protein PRNP, and synaptic proteins NRXN1 and NRXN2 central hubs within the network. Functional enrichment terms “vesicle,” “immune system,” and “synapse” were relevant to a large proportion of nodes in the network (green-, blue-, and red-colored nodes). Similarly, Metacore pathway analysis revealed that the most significantly (p ≤ 0.05) impacted pathways included amyloid-related biological processes (“gamma-secretase proteolytic targets” and “gamma-secretase regulation of neuronal cell development and function”), immune response (“lectin-induced complement pathway,” “classical complement pathway,” “alternative complement pathway,” and “alternative complement cascade disruption in age-related macular degeneration”), and trafficking-related processes (“dynein—dynactin motor complex in axonal transport in neurons” and “aberrant lipid trafficking and metabolism in age-related macular degeneration pathogenesis”) (Figure 3D).

Leveraging the Emory AD versus control reference standards, 144 proteins that were significantly disrupted in AD versus control reference standards (p ≤ 0.05) were also differentially affected by CT1812 in the meta-analysis (p ≤ 0.05). AD-relevant proteins normalized toward healthy control levels by CT1812 are plotted to illustrate the differential abundance in CT1812 versus placebo-treated SPARC and SHINE-A participants (p ≤ 0.05; Figure 3E) relative to AD versus controls. Among the proteins normalized by CT1812 treatment were synaptic signaling proteins calbindin 2 (CALB2) and semaphorin 4B (SEMA4B), lipid homeostasis protein apolipoprotein L1 (APOL1), and several AD or Aβ-related proteins that were found previously to be normalized by CT1812 in SHINE-A (SPON1, HTRA1, CLU).¹⁴ The meta-analysis further identified that AD genetic risk factor olfactomedin-like 3 (OLFML3) was normalized by CT1812 treatment, a finding not detected previously in the SHINE-A analysis.

3.6 Comparative analysis between SPARC and SHINE-A and meta-analysis of CSF proteomics combining both cohorts identifies robust pharmacodynamic biomarkers of CT1812

Comparative analysis was performed to identify overlapping impacted proteins between the two trials and the meta-analysis. A total of 51 proteins with unique Uniprot identifiers were observed as commonly altered (using p ≤ 0.1) in SPARC, SHINE-A and in the meta-analysis (combined SPARC and SHINE-A) (Figure 4A; Table S2). Among the 51 overlapping proteins across trials and analyses, protein–protein interactions were identified using STRING analysis (Figure 4B). APP and PRNP were major hub proteins in the protein–protein interaction map (Figure 4B). Of interest, 46 of 123 proteins found to be significant (p ≤ 0.05) in the meta-analysis that were not significant (p ≤ 0.05) in the independent proteomic analyses of SHINE-A and SPARC cohorts were trending (p ≤ 0.1) in both cohorts. All 46 were found to be altered in the same direction (Table S3). Notably, 13 of these 46 biomarkers did not achieve statistical significance (p ≤ 0.05) in either SPARC or SHINE-A and thus represent novel candidate biomarkers of CT1812 identified in the higher-powered meta-analysis (Table S3).

FIGURE 4

Comparative analysis of SPARC, SHINE-A, and multi-consensus analyses support role of CT1812 at synapses. (A) A comparative analysis assessing the overlap of biomarkers identified from independent cohorts and analyses was performed, identifying 51 proteins in common (p ≤ 0.1). To generate the Venn diagram, Uniprot identifiers (Gene = 0 included) were used. (B) STRING analysis was performed on the 51 proteins (unidentified proteins removed) and showed a significant network interaction (medium confidence threshold 0.400; disconnected proteins excluded from map) and high connectivity with amyloid precursor protein (APP) and prion protein (PRNP), circled in royal blue along with other proteins of interest. (C) STRING analysis of these proteins generated Biological Process Gene Ontology (GO) terms illustrating several enriched pathways related to synaptic function and amyloid biology. (D) When the differential expression significance threshold was increased to p ≤ 0.05, five biomarkers were found to be replicated (statistical significance and directionality) across the cohorts and analyses (GPR37, G-protein coupled receptor 37; CLU, clusterin; CPE, carboxypeptidase E; SPON1, spondin 1; and IGLV1-44, Immunoglobulin Lambda Variable 1-44).

Pathway analysis (Figure 4C) using STRING of the 51 overlapping proteins (p ≤ 0.1 criterion) identified several significant (p ≤ 0.05) biological process GO terms related to neuronal processes and synaptic function, including “neuroligin clustering involved in postsynaptic membrane assembly,” “sympathetic neuron projection extension and guidance,” “gephyrin clustering involved in postsynaptic density assembly,” “negative regulation of dendritic spine maintenance,” “postsynaptic density protein 95 clustering,” and “neuronal signal transduction.” Among the top rank order of significant pathways were “positive regulation of tau-protein kinase activity” and several processes related to amyloid biology (“positive regulation of amyloid fibril formation,” “negative regulation of amyloid beta formation,” “positive regulation of amyloid precursor protein catabolic processes,” and “regulation of amyloid precursor protein catabolic process.”)

To identify the most robust candidate biomarkers of CT1812, the 51 overlapping proteins were assessed for statistical significance of p ≤ 0.05. Five proteins met the p ≤ 0.05 criterion across independent cohorts and analyses (Figure 4D). These proteins include IGLV1-44, which was increased with CT1812 treatment, and four additional proteins (GPR37, CLU, CPE, and SPON1), which were decreased with CT1812 treatment (Figure 4D).

3.7 Network analysis of SPARC/SHINE-A meta-analysis identifies groups of proteins significantly associated with CT1812 treatment

The 2102 differentially abundant proteins identified in the meta-analysis of change from baseline CT1812 versus placebo were used to build a protein co-expression network using WGCNA. This network consisted of 21 modules or groups of proteins related to one another by their co-expression (Figure 5A), which were then assessed for correlation with patient traits (Figure 5A). Of the 21 modules, seven were significantly correlated with CT1812 treatment (p ≤ 0.05), with five modules negatively correlated (Greenyellow, r = −0.45; Purple, r = −0.40; Red, r = −0.38; Black, r = −0.37; and Tan, r = −0.35) and two modules positively correlated (Lightcyan, r = 0.37 and Blue, r = 0.37) with CT1812 (Figure 5B). Purple and Lightcyan modules were additionally correlated with CT1812 dose (i.e., 0 [for placebo], 100 mg, or 300 mg), with Purple negatively correlated (r = −0.34) and Lightcyan positively correlated (r = 0.35). Lightcyan also correlated with CSF CT1812 levels (r = 0.44) (Figure 5A). Other traits were assessed for correlation with CT1812-associated modules, and we found that Greenyellow and Black modules correlated negatively with NRGN change from baseline (Greenyellow r = −0.43 and Black, r = −0.39) (Figure 5A). None of the CT1812-associated modules were correlated with participant sex, APOE ε4 status, BMI, or baseline MMSE score (Figure 5A).

FIGURE 5

Network analysis identifies seven modules significantly associated with CT1812 treatment. Proteomic change from baseline data from Figure 2A was used for weighted gene co-expression network analysis (WGCNA) to identify networks (modules) of proteins that are co-expressed longitudinally. (A) Hierarchical clustering dendrogram showing clustering of proteins into color modules (21 networks). Modules were mapped to previously defined cell types,^{22, 56} indicated below module names. Significant enrichment of a cell type (p ≤ 0.05) is indicated with asterisk. Heatmap of module association with traits listed along the y-axis with r values (top) and p-values (bottom) indicated for each module-trait association within the heatmap (red: positive correlation; blue: negative correlation). Significantly correlated (p ≤ 0.05) module–trait associations are indicated with black boxes. (A–B) Seven of 21 modules were correlated significantly with treatment (p ≤ 0.05, CT1812 vs placebo): “Greenyellow,” “Purple,” “Red,” “Lightcyan,” “Black,” “Blue,” and “Tan.”

3.8 Networks correlated with CT1812 treatment are associated with synapses and amyloid biology

To elucidate potential CT1812 mechanisms, we analyzed module protein membership, identifying the proteins with the strongest module membership (“hub” proteins, Figures 6A,D and 7A,D; largest nodes/proteins in center of module wheel) and other associated proteins (smaller nodes). The Tan module hub proteins included S2R-interacting protein PRNP (Figure 6A). Other hub proteins of interest include AD-associated biomarker VGF (nerve growth factor inducible) and CT1812 candidate biomarker GPR37 (Figure 6A). GO analysis of the Tan module showed significant enrichment in neuronal biological processes (“Learning or Memory,” “Synaptic Signaling,” and “Regulation of Synaptic Plasticity”) and vesicular cellular components (“Transport Vesicle Membrane,” “Transport Vesicle,” “Dendritic Tree,” and “Dendrite”) (Figure 6B), which demonstrate that this group of proteins that are significantly correlated with CT1812 are involved in neuronal signaling and vesicular trafficking. Metacore pathway analysis identified significant enrichment in pathways related to amyloid biology (p ≤ 0.05; “Gamma-secretase proteolytic targets,” “Gamma-secretase regulation of neuronal cell development and function,” and “Protein folding and maturation: Regulation of amyloid precursor protein processing”). Other Metacore-identified pathways related to neuronal signaling were also significantly enriched (p ≤ 0.05): “Post-translational processing of neuroendocrine peptides,” “Role of CDK5 (cyclin dependent kinase 5) in the nervous system,” and “NOTCH signaling in the nervous system;” Figure 6C).

FIGURE 6

Treatment-associated modules: Tan (M12) and Purple (M10) modules composed of S2R-interacting proteins as hub proteins. Weighted gene co-expression network analysis (WGCNA) was performed to identify modules of proteins that are co-expressed longitudinally across participants. (A–C) Tan module: (A) Proteins comprising the Tan module are shown with hub proteins (inner circle) with greatest association (kME; eigengene-based connectivity), and other constituent proteins from the module (larger size node = larger kME). Hub proteins of interest (GPR37, G protein–coupled receptor 37; PRNP, prion protein [PrP^c]; VGF, VGF nerve growth factor inducible) are indicated in royal blue circles. (D–F) Purple module: (D) Proteins comprising the Purple module are shown as described in (A). Hub proteins of interest (APLP2, amyloid beta precursor-like protein 2; NRXN2, neurexin 2; APP, amyloid precursor protein) are indicated in royal blue circles. The biology represented by each network module was determined using gene ontology (GO) analysis (B, E; enrichment shown by z score), or by Metacore (Version 24.1.71600) (C, F; p ≤ 0.05; top 10 listed, non-relevant diseases/organs removed from list). Relevant pathways of interest are indicated in blue text.

FIGURE 7

Treatment-associated modules: Black (M7) is a neuronal module associated with pre- and post-synaptic processes, and Greenyellow (M11) is associated with glycosylation and lipid metabolism pathways. Weighted gene co-expression network analysis (WGCNA) was performed to identify modules of proteins that are coexpressed longitudinally across participants. (A-C) Black module: A. Proteins comprising the Black module are shown with hub proteins (inner circle) with greatest association (kME; eigengene-based connectivity), and other constituent proteins from the module (larger size node = larger kME). Hub proteins of interest (SEMA7A, semaphorin 7A; NTNG1, netrin G1; SEMA6A, semaphorin 6A) and module constituent carboxypeptidase E (CPE) are indicated in royal blue circles. (D–F) Greenyellow module: (D) Proteins comprising the Greenyellow module are shown, as described in (A). The hub protein of interest clusterin (CLU) is indicated in a royal blue circle. The biology represented by each network module was determined using gene ontology (GO) analysis (B, E; enrichment shown by z score), or by Metacore (Version 24.1.71600) (C, F; p ≤ 0.05; top 10 listed, non-relevant diseases/organs removed from list). Relevant pathways of interest are indicated in blue text.

The CT1812 dose–associated Purple module hub proteins consisted of key proteins involved in AD biology, such as APP, APLP2, and NRXN2 (Figure 6D). GO analysis of the module members showed significant (p ≤ 0.05) enrichment in molecular function “Amyloid beta Binding” and cellular components “Endocytic Vesicle Membrane” and “Dendritic Spine” (Figure 6E). Metacore pathway analysis identified pathways related to amyloid biology were significantly enriched (p ≤ 0.05; “Protein folding and maturation: Amyloid precursor protein processing,” “Gamma-secretase proteolytic targets,” and “Gamma-secretase regulation of neuronal cell development and function” Figure 6F).

Treatment-associated Black module (r = −0.37; p = 0.029) was neuronally enriched, as assessed via cell-type enrichment (“Neuron” cell type; p = 1.6e⁻⁵). Hub proteins included synapse-related proteins netrin G1 (NTNG1), semaphorin 7A (SEMA7A), and semaphorin 6A (SEMA6A) (Figure 7A). CT1812 candidate pharmacodynamic biomarker CPE was identified in the second tier of the Black module (Figure 7A). GO analysis of module members showed significant enrichment in several synaptic biological processes (e.g., “Regulation of Dendritic Spine Development” and “Maintenance of Synapse Structure”) and pre- and post-synaptic cellular compartments (p ≤ 0.05; Figure 7B). STRING pathway analysis similarly included top-enriched STRING clusters related to semaphoring signaling, neurexin and neuroligin signaling, and synaptic membrane assembly (strength ≥1.3, FDR ≤0.05). Metacore pathway analysis of proteins comprising the Black module expanded upon pathways that were enriched other than synapse and included enrichment in “Cholesterol dysregulation in Alzheimer's disease” and “Regulation of amyloid precursor protein processing” (p ≤ 0.05; Figure 7C).

Greenyellow was the most significant module correlated with CT1812 treatment (r = −0.45; p = 0.007) and CLU (apoJ) was the second highest-ranked hub protein (Figure 7D). GO analysis of proteins comprising the Greenyellow module showed enrichment in proteostasis-related biological processes (“Protein O-Linked Glycosylation” and “Protein Folding”) (Figure 7E). STRING pathway analysis identified “Lipoprotein particle” as the top STRING network cluster enriched by Greenyellow module proteins (strength = 1.42, FDR = 0.045), as well as enriched KEGG pathway “Cholesterol metabolism” (strength = 1.3, FDR = 0.008). Metacore pathway analysis also demonstrated O- and N-glycan biosynthesis pathways as the topmost enriched by Greenyellow module proteins and includes “Negative regulation of WNT/beta-catenin signaling” among the significantly enriched pathways (Figure 7F).

Treatment-associated Lightcyan module (r = 0.37; p = 0.027) was microglia-enriched, as assessed via cell-type enrichment (“Microglia” cell type; p = 0.003). Hub proteins consisted primarily of histone and cytoskeleton-related proteins (Figure S4A), which in turn drove the GO analysis (Figure S4B) and STRING pathway analysis (Figure S4C) enrichment of several actin cytoskeleton biological processes. STRING analysis also identified enrichment in phagocytic and vesicular cellular components (“Phagocytic cup,” “Vesicle,” and “Extracellular exosome”; FDR≤0.01), as well as KEGG pathway “Phagosome” (strength 1.28; FDR 0.00029). Individual protein members of the pathway “Phagosome” included ITGB2, histocompatibility antigen DR alpha chain (HLA-DRA), and TUBB (tubulin beta chain). Additional STRING results showed significant enrichment (strength 1.49; FDR 0.0103) of the cellular component “immunological synapse.” Metacore pathway analysis showed enrichment in top pathways related to immune response and transport, including via phagocytosis (“Immune response_iC3b-induced phagocytosis via alpha-M/beta-2 integrin”; Figure S4C).

A higher-level summarization of findings from the proteomics analyses focused on candidate biomarker nomination in relation to CT1812 mechanism of action and AD is shown in Figure 8. Beyond the differential abundance and network analyses, correlation analysis with CSF CT1812 exposure levels at 6 months in the meta-analysis provided further biomarker evidence of target engagement, as proteins known to interact with S2R and S2R components were correlated with CT1812 levels: netrin receptor DCC (Uniprot P43146; r = −50.4, p = 8.49e⁻⁶), PRNP (A2A2V1; r = −0.39, p = 7.44e⁻⁴), and low-density lipoprotein receptor (LDLR) (J3KMZ9; r = 0.30, p = 1.2e⁻²). Further pharmacodynamic correlates with CT1812 exposure of AD synaptic biomarkers were observed: NRGN (Uniprot Q92686; r = −0.47, p = 4.00e⁻⁵), Syt1 (P21579; r = −0.42, p = 2.58e⁻⁴), and synaptotagmin 11 (Syt11) (Q9BT88; r = −0.34, p = 3.76e⁻³).

FIGURE 8

Biomarkers replicated across cohorts and analyses support the hypothesized CT1812 mechanism of action. CT1812 (cyan) binds the sigma-2 receptor (S2R), or TMEM97 (transmembrane protein 97; purple), which is in complex with LDLR (low-density lipoprotein receptor) and PGRMC1 (progesterone receptor membrane component 1; purple) at the cell surface. Amyloid-β (Aβ) oligomers (red) bind the oligomer receptor complex (blue), containing LilrB2 (leukocyte immunoglobulin-like receptor subfamily B member 2), NoGo, and the major component PrP^c(prion protein), which S2R (TMEM97) directly binds.²⁴ CT1812 engagement of the S2R complex induces a conformational change in the oligomer receptor complex, disengaging the oligomer from the synapse.³ Biomarkers identified in SHINE-A [interim SHINE; NCT03507790], SPARC [NCT03493282], and the meta-analysis suggest target engagement through TMEM97 (S2R)-interactors LDLR (purple box), PrP^c (blue box), Aβ (maroon box), and where S2R is localized, synapses. In addition, biomarkers related to inflammation, genetic risk factors of Alzheimer's disease (AD), and disease modification are shown. ^tIndicates normalization toward healthy control cerebrospinal fluid (CSF) levels, * indicates proteins identified as correlates with volumetric magnetic resonance imaging (vMRI), and bolded royal blue font indicates proteins identified in treatment-related weighted gene co-expression network analysis (WGCNA) modules as hub proteins.

4 DISCUSSION

This study provides the first exploratory CSF proteomics characterization of 6 months of CT1812 treatment in the SPARC trial cohort with mild to moderate dementia due to AD. Findings from the SPARC trial were corroborated through comparative analysis of proteomic findings from the SHINE-A cohort. Finally, the higher-powered meta-analysis combining proteomic data from the SPARC and SHINE-A cohort enabled the discovery of novel pharmacodynamic biomarkers of CT1812 that were previously undiscovered, including those not identified in the SPARC cohort analysis alone. Brain module mapping, pathway analyses, and WGCNA to identify modules correlated with CT1812 treatment were also performed and provide a systems-level examination of the pharmacodynamic effect of CT1812 in patients with AD. Together, the findings elucidate the wide-spanning pharmacodynamic effects of CT1812 in patients with mild to moderate dementia due to AD.

4.1 Biomarkers identified in the SPARC cohort

Analysis of the SPARC cohort CSF proteome revealed biomarkers of interest in AD and neurodegeneration (CLU, CPE, SPON1, APOC4, GPR37) with a more expansive list delineated by functional category in Figure 8. Notably, interactors of the S2R complex (e.g., PRPN and NPC1), and other proteins associated with target or pathway engagement of CT1812 with S2R were also identified. Brain mapping of the differentially abundant proteins to AD-relevant modules¹⁰ further revealed that the impacted genes relate to synapse biology, and STRING- and Metacore-based pathway analysis confirmed a role in amyloid biology and immune response. These findings from SPARC corroborated many results from our recent exploratory unbiased proteomic analysis of a separate independent mild to moderate AD cohort, the SHINE-A interim cohort.¹⁴ It is important to note that with the SPARC analysis we identify biomarkers that strongly correlate with brain volumetric MRI endpoints, which include CLU and APP as highly interconnected proteins. As reported previously, CT1812 treatment exhibited a trend toward slowing the rate of brain volume loss on the vMRI regional composite.¹³ We identified 45 correlates that had also been differentially expressed in CT1812- versus placebo-treated CSF (p ≤ 0.1 threshold), biomarkers of CT1812 that may also represent disease modification. Remarkably, this subset of proteins altered by CT1812 showed strong enrichment for pathways related to synapse structure and function, as identified using STRING, further supporting CT1812 mechanisms in synaptic protection.

4.2 A subset of pharmacodynamic biomarkers of CT1812 replicate across independent clinical trials

Comparative analysis between SPARC and SHINE-A identified robust candidate biomarkers of CT1812 that replicated across the two Phase 2 trials. When p-value criteria of p ≤ 0.1 and p ≤ 0.05 were applied to the proteins impacted by treatment with CT1812, a total of 55 and 6 proteins, respectively, were identified to be altered, with 51 and 5 moving in the same direction, suggesting a clear role of CT1812 in modulating their expression. Assessing impacted proteins at these two different significance thresholds allows for more effective nomination of candidate biomarkers: lower stringency avoids type II errors and allows monitoring of trends and impacted pathways/networks, whereas the higher stringency threshold refines our pool of individual candidate biomarkers. Among the biomarkers identified that replicate across both trials (p ≤ 0.05), moving in the same direction, were proteins linked to the AD phenotype: CLU, CPE,²⁵ and SPON1, among others.

4.3 Novel biomarkers discovered in the meta-analysis

Because both SPARC and SHINE-A were relatively small cohorts for this exploratory analysis, given the similarities in cohorts— same patient population (mild to moderate dementia due to AD; similar baseline characteristics: age, sex, race, BMI, baseline MMSE score; Table S1), same doses of CT1812 and treatment duration—we combined the two congruent proteomic data sets into a meta-analysis. This meta-analysis provided a higher-powered study, enabling novel candidate biomarkers to be discovered. Forty-six biomarkers that had only trended toward significance in both individual SPARC and SHINE-A analyses achieved statistical significance in the meta-analysis, bolstering the detection of an impact on biomarkers of synaptic health, as well as immune response, lipoprotein, and Aβ biology. Among these biomarkers newly revealed by the more highly powered analysis, PRNP and SYT7 were significantly altered and the AD risk factor OLFML3 was found to be normalized by CT1812 treatment. Other proteins associated with Aβ biology or AD risk factors (CLU, SPON1, HTRA1) that were identified in the SHINE-A analysis as normalized toward benchmark healthy control levels by CT1812 were similarly found be normalized in this meta-analysis.

4.4 Compelling biological links to AD and S2R biology

A lowering of CLU (also known as ApoJ) by CT1812, identified previously in SHINE-A, was corroborated by the SPARC and meta-analyses. A significant genetic risk factor for late-onset AD,^{26, 27} CLU is a multifunctional glycoprotein with a key role in lipid transport, immune modulation, and Aβ aggregation and/or clearance.^{28, 29} In addition to a role in AD, given that Aβ oligomer displacement and clearance into the CSF is a central mechanism of CT1812,^{3, 4} CLU may also be a potential biomarker of CT1812-pathway engagement.

Prion protein (PRNP; PrP^c, an Aβ oligomer receptor), also a novel discovery in the SPARC/SHINE-A meta-analysis, is a direct interactor of the S2R, TMEM97.²⁴ When CT1812 binds TMEM97, resulting in the displacement of Aβ oligomers from PrP^c, downstream changes in PrP^c-mediated signaling and a subsequent rescue of neuronal function are observed.²⁴ The ability of CT1812 to impact PRNP suggests that CT1812 can engage its target (i.e., bind to TMEM97) and successfully modulate disease-relevant pathways.³⁰ The extensive interconnectivity between the proteins identified in this meta-analysis including several that are tied to PRNP and S2R is compelling and may enable hypothesis generation on key downstream mediators of target engagement by CT1812.

APP is another Aβ-related biomarker that can bind to and be regulated by PrP^c31 and hence may be a biomarker of pathway engagement. APP levels were lower with CT1812 treatment in SHINE-A, a trend of a decrease was seen in SPARC, and significant lowering was observed in the meta-analysis. Of note, APP is a central hub of interconnected nodes in the STRING network analysis and may be a driver of downstream changes relevant to AD. SPON1, which was found to be significantly lower in CSF in participants treated with CT1812 in SHINE-A, SPARC, and the meta-analysis, binds APP, thereby inhibiting β-secretase cleavage, and is associated with dementia risk.^{32, 33} SPON1 is also member of the matrisome module in AD brain, which also includes APP and apoE, which is the module most strongly associated with AD pathology, localizes to amyloid plaques,¹⁰ and is the earliest changing protein identified in CSF in autosomal dominant AD—≈30 years prior to symptom onset.³⁴ Given both the disease relevance and the mechanism of action of CT1812 in impacting Aβ dynamics in displacing Aβ oligomers from neuronal synapses, it is encouraging that many of the consistently identified proteins (CLU, PRNP, APP, and SPON1) relate to Aβ.

As an additional finding related to APP, the abundance of LDLR-related protein (LRP1), an LDL receptor family member, in CSF was significantly decreased in CT1812 versus placebo in the SPARC trial, with a decreasing trend in the meta-analysis. LRP1 is a receptor of Aβ and plays multiple roles in amyloid regulation. LRP1, like S2R-interacting protein LDLR, can regulate cholesterol uptake through endocytosis of apoE-enriched very-low-density lipoprotein (VLDL).³⁵ Furthermore, LRP1 can modulate APP trafficking and processing (reviewed in Eggert, et al. 2017³⁶) and regulate Aβ clearance by glial cells.^{37, 38} Although the direct interaction between LRP1 and S2R is unknown, LRP1 interacts with PrP^{c39, 40} and can facilitate PrP^c-mediated Aβ oligomer uptake and toxicity.³⁹ Demonstrating the interconnectivity of the biomarkers identified, NPC1, a direct binder of the S2R TMEM97,²³ is an intracellular cholesterol transporter that also interacts with the LDLR component of the S2R complex^{41, 42} and was found to be altered in both SPARC and SHINE-A. Although we consider this finding with caution because the CT1812 effect occurred in opposite directions in each of the cohorts, further evaluation is warranted given the strong tie to TMEM97.

Other noteworthy biomarkers identified in these analyses include some that relate to secretion and vesicle trafficking, such as CPE⁴³ and GRP37.⁴⁴ Preclinical data show that CT1812 regulates neuronal vesicle trafficking deficits caused by Aβ oligomer.^{45, 46} S2R localizes to synapses,⁴⁷ as well as to subcellular compartments integral to vesicular trafficking, as demonstrated in other cell types.⁴⁸ Although CPE and GRP37 do not have direct links to S2R, changes to proteins involved in vesicular and endosomal transport processes may reflect CT1812-S2R target engagement, processes that are further supported by the CT1812-associated modules in our network analysis.

4.5 Network analysis provides a deeper systems-based perspective and supports mechanism of CT1812

To understand from a systems-level or network-based approach the pharmacodynamic effect of CT1812 in patients with AD, we leveraged WGCNA. WGCNA is a powerful systems-based analysis applied in biomarker discovery for AD,^{9, 10, 21} used to construct protein networks or “modules” in which proteins with similar co-expression patterns across samples are grouped together and are thought to represent biologically meaningful networks linked by similar biological functions or specific cell types. The highest centrally connected proteins (hub proteins) have the greatest degree of interconnectivity with other proteins in the network (have a high kME, or module eigengene value)⁴⁹ and are thought to drive the biology associated with the group of proteins in the network.

In this study, WGCNA expanded upon the differential expression analyses from the SPARC/SHINE-A meta-analysis to understand the biological relationships between significantly changed proteins altered by CT1812 treatment. Seven modules correlated with CT1812 treatment, and, notably, several of the treatment-related modules comprised S2R components and direct or indirect interacting proteins as hub molecules, supporting the notion that the hypothesized CT1812 mechanism of action, through S2R, is impacting biomarkers altered with drug treatment.

Analysis of the treatment-associated modules strongly supports our proposed CT1812 mechanism of action (schema in Figure 8), in which CT1812 targets S2R, a synapse-localized protein⁴⁷ that interacts with Aβ oligomers at synapses.^{3, 24} In our unbiased network approach, four of the seven treatment-related modules were neuronal enriched (Black, Red, Tan, Purple). GO and pathway analyses of these modules support CT1812 treatment effects on synapse function (Purple, Tan, Black) and neurotransmission (Purple, Black), consistent with a localization of SR2 to neuronal synapses²⁴ and a role in synaptoprotection (see review of Lizama et al., 2023).^{5, 45, 46} At the molecular level, S2R (TMEM97) directly binds the major constituent of the oligomer receptor, PRPN (PrP^c).²⁴ Of interest, network analysis identified PRNP as a hub protein (Tan module), supporting CT1812 engagement with its direct target (S2R) and supporting a CT1812 effect in disease-relevant biological pathways, as the Tan module was strongly associated with neuronal processes. In our proposed mechanism of action, CT1812 engages S2R (TMEM97), enabling a conformational change disrupting direct interaction of PrP^c with Aβ oligomers. Consistent with this model, treatment-related modules were identified that contained hub proteins that are directly or indirectly involved in amyloid biology. The Purple module hub proteins included APP and its paralogue APLP2,⁵⁰ which are both known to regulate PRNP, and PRNP is known to regulate APP and APLP2.³¹ Furthermore, CLU (hub protein in Greenyellow) is a lipoprotein that can bind Aβ and is involved in Aβ aggregation, clearance, and transport mechanisms.⁵¹ Finally, the neuronal Black module contained as a hub protein reticulon 4 receptor (RTN4R, also known as Nogo-66 receptor 1 or Nogo), is involved in synaptic function and another Aβ-binding receptor⁵² comprising a complex with PrP^c (schema in Figure 8). Although RTN4R itself was not changed significantly in the meta-analysis (p = 0.15, CT1812 vs placebo), its presence in a synaptic, treatment-related module and interconnectedness (kME of 0.86) with other synaptic proteins is interesting given the synaptoprotective mechanism of CT1812 through S2R and its interaction with the oligomer receptor. Taken together, these the network data analyses identify CT1812-associated networks that represent key disease-relevant biological pathways that are driven by proteins directly (PRNP) or indirectly interact with S2R.

Beyond the S2R-interacting proteins and amyloid-related processes, the treatment-associated networks were associated with vesicular trafficking processes, consistent with S2R biology and CT1812 mechanism–related processes that were elaborated previously through preclinical studies (reviewed in Lizama et al., 2023⁵). Other treatment-associated modules were glia enriched such as the Lightcyan module, in which pathway analysis highlighted immune processes such as complement and phagocytic mechanisms, which are well-studied processes that are implicated in AD pathophysiology.⁵³ Microglia play roles in regulating neurotransmission and synaptic density via phagocytic mechanisms and complement-dependent–mediated synaptic pruning.⁵⁴ Given that microglia can respond to and impact changes in synaptic activity, it is interesting to speculate whether there may be a non-cell-autonomous role of CT1812, though neuronal S2R, that may lead to an effect on microglia to further regulate synaptic activity and synaptic density. In addition, microglial phagocytosis modulates Aβ clearance, and Aβ oligomers have been shown to induce an inflammatory response resulting in microglial phagocytosis,⁵⁵ providing another possible connection between the findings in this study. Although speculative at present, future studies will determine if any of these hypotheses may be worth further study preclinically to elaborate potential mechanism underpinnings of CT1812 in AD.

4.6 Study limitations

Although we leveraged two independent Phase 2 clinical trials to increase power of the proteomic analyses to identify biomarkers of CT1812, the meta-analysis is still limited by a small number of participants and samples. It will be important to validate candidate CT1812 biomarker findings in larger cohort studies assessing the effect of CT1812 in patients with dementia.

4.7 Conclusions

• The unbiased proteomics analyses reported from SPARC, comparative analysis of SPARC with SHINE-A, and meta-analysis of proteomic data sets from the two independent cohorts identify CT1812 pharmacodynamic biomarkers that replicate across independent trials and analyses.
• Pathway analyses of networks identified amyloid biology, synaptic function, and immune response pathways associated with CT1812 treatment, further supporting a CT1812 mechanism of action.
• A subset of proteins that were correlated with brain composite vMRI in CT1812-treated SPARC participants was also differentially abundant with CT1812 treatment versus placebo and highly enriched for synaptic biological processes, corroborating preclinical evidence that CT1812 is protective of synapses.
• The meta-analysis both corroborated biomarkers and pathways identified previously and facilitated the identification of additional, novel biomarkers that were previously undetected.
• A systems-based approach using WGCNA enabled treatment-associated networks to be illuminated with S2R-interacting proteins as key hub proteins, which may indicate that changes in the network may reflect target/pathway engagement of S2R by CT1812.
• Collectively, multi-modal analyses of the SPARC/SHINE-A meta-analysis further illuminated CT1812 mechanism of action in AD, from which we identified S2R-interacting (PRNP) and amyloid-related (APP, APLP2, and CLU) proteins as drivers of the biological processes correlated with CT1812 treatment.
• These findings support the continued research into the pharmacodynamic effects of CT1812 through proteomics discovery to further the understand effects of CT1812 treatment in patients with AD.
• Replication of these findings in larger cohorts will support clinical validation of these emerging biomarkers of CT1812.

AUTHOR CONTRIBUTIONS

The analysis was conceived by Mary E. Hamby, Allan I. Levey, Nicholas T. Seyfried, and Kiran Pandey, Duc Duong conducted the TMT-MS; Kiran Pandey, Britney N. Lizama, Claire Williams, Valentina Di Caro, Mary E. Hamby, and Hilary A. North analyzed the data. Kaj Blennow and Henrik Zetterberg led canonical CSF sample measurements and analysis. Christopher H. van Dyck, Adam P.Mecca, Michael Grundman, Mary E. Hamby, and AnthonyO. Caggiano were involved in clinical trial design and analysis. All authors interpreted the data. Kiran Pandey contributed to the statistical analysis. Britney N. Lizama, Mary E. Hamby, and Hilary A. North wrote the manuscript. All authors contributed to revising/editing the manuscript.

ACKNOWLEDGMENTS

We thank the research participants for their contributions, and the staff of the Yale ADRU and the Yale PET Center for their excellent technical assistance, as well as former Cognition Therapeutics employees Evi Malagise and Lora Waybright for technical assistance in earlier analyses. This work was supported by funding from the National Institute on Aging (RF1AG057553 and 1R01AG058660-01), the Alzheimer's Drug Discovery Foundation, and by Cognition Therapeutics. H.Z. is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council supported by grants from the Swedish Research Council (#2023-00356; #2022-01018 and #2019-02397), the European Union's Horizon Europe research and innovation programme under grant agreement No. 101053962, and Swedish State Support for Clinical Research (#ALFGBG-71320).

CONFLICT OF INTEREST STATEMENT

B.N.L., C.W., V.D., A.O.C., and M.E.H. are employees and shareholders of Cognition Therapeutics, Inc. M.G. is a consultant to and shareholder in Cognition Therapeutics. H.A.N. is a consultant to Cognition Therapeutics. H.Z. has served on scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, ALZPath, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognition Therapeutics, Denali, Eisai, Nervgen, Novo Nordisk, Passage Bio, Pinteon Therapeutics, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave; has given lectures in symposia sponsored by Cellectricon, Fujirebio, Alzecure, Biogen, and Roche; and is a cofounder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside the submitted work). K.B. has served as a consultant and at advisory boards for AC Immune, Acumen, ALZPath, AriBio, BioArctic, Biogen, Eisai, Lilly, Moleac Pte. Ltd, Novartis, Ono Pharma, Prothena, Roche Diagnostics, and Siemens Healthineers; has served on data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials, and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai, and Roche Diagnostics; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside of the submitted work). D.M.D., N.T.S., A.I.L., and K.P. are co-founders, employees, consultants, and/or shareholders of EmTheraPro. C.H.vD. reports consulting fees from Eisai, Roche, Ono, and Cerevel and grants for clinical trials from Biogen, Eli Lilly, Eisai, Roche, Genentech, UCB, and Cerevel outside the submitted work. A.P.M. reports grants for clinical trials from Genentech, Eli Lilly, and Janssen Pharmaceuticals outside the submitted work. All other authors have no competing interests to declare. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All human subjects provided written informed consent.