Inhibition of Moesin and CD44 in stem cell‐derived neurons affects the pathological genetic signature associated with Alzheimer's disease

Abstract

INTRODUCTION

Post mortem proteomic analysis of Alzheimer's disease (AD) brain tissue has identified novel target genes and proteins for potential therapeutic development. Moesin (MSN) and CD44 were identified as candidate targets. Using human induced pluripotent stem cells (hiPSCs)‐derived neurons, we assayed how reducing gene expression of MSN and CD44 affected amyloid beta secretion, tau phosphorylation, and transcriptional state.

METHODS

Knockdown of MSN and CD44 in hiPSC‐derived neurons was performed using short hairpin RNA (shRNA). Amyloid precursor protein processing and intracellular tau phosphorylation was measured using chemiluminescent ELISA assays. Global gene expression was analyzed by bulk RNA sequencing (RNA‐seq).

RESULTS

Knockdown of CD44 and MSN increased secretion of amyloid beta and soluble APP beta and differentially altered tau generation and phosphorylation. RNA‐seq revealed diverse effects of CD44 and MSN knockdown and the reversal of some expression shifts found in late‐onset AD.

DISCUSSION

These findings connect reduced CD44 and MSN expression to AD‐related signatures in neurons and inform their role in the disease.

Highlights

• CD44 and MSN genes are upregulated in AD.

• CD44 and MSN gene expression was knocked down in hiPSC neurons.

• CD44 and MSN KD increased Aβ secretion and altered pTau generation.

• CD44 and MSN KD reversed several transcriptomic signatures of AD brains.

1. BACKGROUND

Alzheimer's disease (AD) is characterized by neurodegeneration that leads to decreased cognitive function and dementia. The hallmark accumulation of extracellular amyloid beta (Aβ) and intracellular hyperphosphorylated tau (pTau) tangles are implicated in the progressive loss of tissue in the brains of AD patients. ¹ Despite being the most prevalent neurodegenerative disease, there remains an unmet need for more effective AD treatments. ² , ³ , ⁴ Recently approved disease‐modifying therapies that target the build‐up of Aβ successfully slow the disease, but with prominent safety and cost effectiveness concerns. ⁵ When considering the aging population, combined with the mostly undiagnosed earliest prodromal and pre‐symptomatic stages, AD's prevalence and impact on society calls for new approaches to drug discovery.

The Target Enablement to Accelerate Therapy Development for Alzheimer's Disease (TREAT‐AD) program is a National Institute on Aging (NIA)‐funded consortium that aims to develop resources to enable therapeutic development around targets associated with AD to advance the development of safer and more effective treatments. ⁶ The TREAT‐AD program has integrated transcriptomic and proteomic profiles from post mortem samples collected from AD patients to determine gene and protein targets of interest altered in the brains of AD patients. ⁷ Moesin (MSN) and CD44 were prioritized from these efforts and nominated as potential therapeutic targets. MSN and CD44 were also identified as co‐hub proteins in a proteomic co‐expression module (module M4) strongly correlated with clinical phenotypes and enriched for proteins associated with astrocytes and microglia. ⁸

MSN has been implicated in various cell functions, including maintenance of cell structure, mitosis, and actin stabilization. ⁹ , ¹⁰ The moesin protein belongs to the ezrin/radixin/moesin (ERM) family, often expressed in both glial and neuronal cell types and organizes specialized cell membranes. ¹¹ Overstabilization of actin may cause an accumulation of phosphorylated tau, as is seen in the pathogenesis of AD, whereby proteins involved with this process could serve as potential new drug targets for AD. ¹² , ¹³ In drosophila, increased tau levels induced MSN activation, and the cell cycle phase of mitosis triggered cell death rather than proliferation. ⁹ MSN has been found to be expressed by microglia around Aβ plaques and small interfering RNA of MSN in microglia in vitro reduces phagocytosis, an actin‐based mechanism. ¹⁴ , ¹⁵ Moesin protein deficiency in microglia leads to over‐pruning due to blocked interferon signaling, which regulates immune homeostasis. ¹⁶ The TREAT‐AD risk assessment found that MSN protein and RNA were upregulated in post mortem AD brain tissue relative to control individuals. ⁷ , ¹⁷

CD44 is a cofactor of MSN and has been found to be associated with Aβ toxicity. ¹⁸ The CD44 protein is a glycoprotein associated with cell–cell interaction, cell adhesion, and migration. ¹⁹ CD44 is highly expressed in glia and upregulated in post mortem AD samples. ⁷ CD44 is upregulated in astrocytes close in proximity to Aβ plaques in 5xFAD mice and may influence Aβ pathology in these cell types. ²⁰ The CD44V6 and CD44V10 splice variants of CD44 are both expressed in neurons, and knockdown (KD) of CD44V10 protects neurons from Aβ toxicity. ²¹ However, soluble CD44 protein has also been shown to induce tau phosphorylation in neurons. ²² The TREAT‐AD risk assessment found that CD44 RNA and protein levels were both higher in post mortem AD brains relative to control donors and CD44 was enriched in the cerebrospinal fluid of asymptomatic AD patients, suggesting that it could be an important early biomarker of the disease. ⁷ , ²⁰ , ²³

Studies examining the function of both CD44 and MSN largely focus on microglia and astrocytes, rather than neurons, making it unclear what downstream effects these genes may have in neuronal cells. ¹⁸ In this study, we reduced the expression of both CD44 and MSN in human induced pluripotent stem cells (hiPSCs)‐derived excitatory neurons using a short hairpin RNA (shRNA) knockdown approach and found that Aβ levels, soluble APP beta (sAPPβ) levels, and total‐tau (t‐tau) expression increased following KD. We also assessed the neuronal transcriptome after KD and found a strong similarity in the genes and functions affected by this perturbation, specifically in the context of synaptic vesicle and endolysosomal trafficking within the hiPSC‐derived neurons. Finally, KD of both CD44 and MSN induced transcriptomic profiles that were negatively correlated with post mortem transcriptomic signatures from AD brains, further informing how CD44 and MSN gene expression is linked to AD.

RESEARCH IN CONTEXT

1. Systematic review: The authors parsed the subject literature and the open‐source TREAT‐AD database for information on the gene targets of interest. Published research has identified the structure and interactions of CD44 and MSN and connected them to AD pathogenesis, but downstream effects on AD‐related outputs has not been investigated as thoroughly.

2. Interpretation: The findings reported in this research article provide insight into the suitability of CD44 and MSN gene expression reduction as a possible avenue of therapeutic development.

3. Future directions: This manuscript informs the roles of CD44 and MSN in human excitatory cortical neurons. Future studies aimed at developing a more complete understanding of how these genes may influence the AD brain include examining expression KD in glial cell types and their effects on AD phenotypes, studying CD44 and MSN in human cell models of AD, and further testing in animal models of AD.

2. METHODS

2.1. Plasmid engineering

Published shRNA sequences were identified for CD44 and MSN KD (Table 1). Short hairpin RNA oligonucleotide sequences were commercially generated (Millipore Sigma). shRNA sequences were cloned into the lentiviral pSICOR plasmid (Addgene: 11579) using restriction enzyme digestion and ligation. ²⁴ Plasmids were digested at the HpaI‐XhoI restriction enzyme site, and shRNA sequences were ligated into the pSicoR plasmid. Competent Escherichia coli bacteria were transformed with the ligated product and expanded. Plasmid DNA was collected via miniprep (Thermo Fisher Scientific) and shRNA insertion was confirmed with Sanger sequencing (Genewiz). A non‐targeting scrambled shRNA was used as a control (Origene). Following sequence confirmation, the shRNA‐containing lentivirus vector was packaged into third‐generation lentiviral particles at the Viral Vector Core of the Emory Center for Neurodegenerative Disease.

TABLE 1.

shRNA sequences.

Gene target	shRNA sequence	Source
MSN	3’UTR GCAGCGCATTGACGAATTTGA TCAAATTCGTCAATGCGCTGC	Lee, 2010
CD44	3’ UTR GACAGAAAGCCAAGTGGACTCAACGGAGA TCTCCGTTGAGTCCACTTGGCTTTCTGTC	Lee, 2017

2.2. Neuron differentiation and transduction

Two male cell lines were used for this experiment: CVI, a cell line derived from the diploid sequence of an individual, ²⁵ , ²⁶ , ²⁷ a derivative of CVB (GM25430, Coriell), and WTC11 (GM25256, Coriell). The CVI line is a heterozygous APOE ε4/3 carrier, while the WTC11 line is homozygous APOE ε3/3 genotype. All experiments were run on both cell lines in parallel.

Neurons were differentiated from hiPSCs  using previously described methods ²⁸ (Figure 1A). In brief, hiPSCs were differentiated into neural progenitor cells (NPCs) using dual‐SMAD inhibition. Human iPSCs were plated on Geltrex (Thermo Fisher Scientific) coated culture plates and maintained with mTesR plus medium (StemCell). On day 1 of neural induction, stem cell culture medium was changed to basal neural maintenance medium (BNMM) and supplemented with SMAD inhibitors (1:1 DMEM/F12 + glutamine media/neurobasal media, 0.5% N2 supplement, 1% B27 supplement, 0.5% GlutaMax, 0.5% insulin‐transferrin‐selenium, 0.5% NEAA, 0.2% β‐mercaptoethanol, 10 µM SB‐431542, and 0.5 µM LDN‐193189 [Biogems, Westlake Village, CA, USA]). Cells were fed daily for 7 days, then were dissociated and split with Versene (Gibco) on day 8. Cells were fed daily with BNMM without SMAD inhibitors until day 13, when medium was changed to BNMM supplemented with 20 ng/mL fibroblast growth factor (FGF) (R&D Systems). Cells were fed daily until approximately day 23, when cells were sorted using fluorescence‐activated cell sorting flow cytometry. Cells that were positive for CD24 and CD184 and negative for CD44 and CD271 ²⁹ were isolated as NPCs. NPCs were maintained in BNMM supplemented with 20 ng/mL FGF. NPCs were expanded and passaged on Geltrex‐coated culture dishes for neuron differentiation. Neuron differentiation was initiated by switching medium to BNMM supplemented with 0.2 ug/mL glial‐derived neurotrophic factor (GDNF) (PeproTech), 0.2 ug/mL brain‐derived neurotrophic factor (BDNF) (PeproTech), and 500 uM cell‐permeant cyclic adenosine monophosphate (db‐cAMP) (Sigma Aldrich). Neuron differentiation proceeded for 3 weeks, with biweekly medium changes. Following 3 weeks of differentiation, neuronal cultures enriched for mature cortical neurons utilizing a magnetic bead negative selection of surface proteins CD44, CD184, and CD271. Neurons were passaged and plated on fresh basement membrane‐coated culture plates for experimental use. For Aβ, pTau, RT‐qPCR, and cytotoxicity assays, 250,000 neurons were plated on Matrigel‐coated 96‐well culture plates. For RNA‐seq experiments, neurons were passaged to six‐well culture plates at a density of 400,000 neurons/cm.

FIGURE 1.

Experimental design and validation of CD44 and MSN KD. (A) Schematic of hiPSC differentiation and shRNA transduction for experiments. Timeline created in BioRender. (B) Expression levels of CD44 and MSN after shRNA transduction in CVI and WTC11 iPSC‐derived neurons shows successful KD of CD44 and MSN. Fold change is calculated in comparison to the control scramble sample (represented by dotted line at 1.0) and tested with two independent differentiations (circle and square data points) and two primer sets (open and closed symbols). Each data point reflects the average of three technical replicates. Error bars indicate the standard deviation. One‐sample t‐tests were used to test for significance: **p < 0.01, ****p < 0.0001.

Following a 48‐h recovery after neuron enrichment and passaging, neuron cultures were transduced with shRNA‐containing lentivirus at a multiplicity of infection of 5 for 3 days. Lentivirus‐containing medium was then replaced with clean maintenance medium. The vector plasmid contained fluorescent marker green fluorescent protein (GFP); transduction of all cultures was validated by GFP fluorescence using a brightfield fluorescent microscope. After 3 days, materials were harvested for assays.

2.3. Secreted Aβ and sAPPα/sAPPβ assays

All cells were plated at 250,000 cells per well. Following successful transduction, neuronal culture medium was fully exchanged and conditioned for 72 h and collected. Secreted Aβ was measured using MSD Aβ V‐PLEX ELISA (Meso Scale Discovery [MSD]). sAPPα and sAPPβ were measured using MSD sAPPα/sAPPβ T‐PLEX ELISA (MSD). Two technical replicates were measured per neuronal culture. Results were normalized to total protein levels determined using the Bicinchoninic Acid protein quantification method.

2.4. Tau assays

Protein lysate for the measurement of intracellular t‐tau and pTau was collected from neurons 3 days after shRNA transduction. Medium was removed and collected for Aβ assay (see above), and neurons were washed once with phosphate buffered saline without calcium or magnesium then lysed for pTau analysis for 15 min on ice using complete MSD lysis buffer. Total‐tau and pTau (Thr231) levels were assayed using MSD Phospho (Thr231)/t‐tau ELISA (MSD). Two technical replicates were measured per neuronal culture. Results were normalized to total protein levels.

2.5. Gene expression assays

Harvesting of RNA was performed using the TRIzol standard protocol (Invitrogen). RNA was purified and cDNA was made as described previously. ³⁰ In brief, cDNA was synthesized using iScript cDNA synthesis kit (BioRad) and concentration measured using Nanodrop. RT‐qPCR was conducted on Real‐Time PCR System (BioRad) using Platinum SYBR Green qPCR SuperMix‐UDG (Invitrogen) reagent, and primers for CD44 and MSN detection were ordered from IDT (Table 2). Gene expression KD was calculated for each shRNA sample compared and calculated by fold change to a scramble shRNA condition and normalized to RPL27.

TABLE 2.

Primers used for CD44 and MSN detection in qPCR.

Target	Forward (5’ ‐ 3’)	Reverse (5’ ‐ 3’)
CD44 1	ATCTTGGCTCCCTCTTGGC	TTCTGCCCACACCTTCTTCG
CD44 2	TCAATGCTTCAGCTCCACCT	TTTCTGGACATAGCGGGTGC
MSN 1	CGCTAGTGAGGGACCCAATC	TCACACGCACACTGATCGTT
MSN 2	CCATGCCCAAAACGATCAGTG	CAGCCAGGTGGAGAAACCTT

2.6. Lactate dehydrogenase assay

Lactate dehydrogenase (LDH) levels in conditioned medium were measured in 72 h of transduction to measure the amount of cell death in each sample using a LDH assay kit following the manufacturer's specifications (Promega). Two technical replicates were measured per neuronal culture.

2.7. Bulk RNA‐seq analysis

For RNA‐seq experiments, CVI and WTC11 neurons were differentiated and transduced as described previously. RNA was harvested using TRIzol (Invitrogen), according to the manufacturer's specifications. RNA was quantified and RNA‐seq processing was performed by Genewiz (Azenta Life Sciences) with the Standard RNA‐seq service. Ribosomal RNA was removed using Poly‐A selection. Sequencing was performed using the Illumina platform. RNA samples were sequenced at a depth of 50 million reads per sample. Sequence reads were trimmed using Trimmomatic version 0.36, ³¹ and trimmed reads were mapped to the human reference genome GRCh38 using the STAR aligner version 2.5.2b, ³² where BAM files were generated. Unique gene hit counts were calculated using the Subread package version 1.5.2. ³³

The KD data from these two targets were part of a larger TREAT‐AD iNeuron target KD study (syn53755339). We used the full study to accurately model the sample variance and power differential expression analysis and then extracted the results for CD44 and MSN. Differential expression was assessed using the DESeq2 R package, ³⁴ and expression counts were modeled as a linear combination of shRNA (i.e., CD44, MSN, control), cell line (i.e., CVI or WTC11), and experimental batch. Log2 fold change shrinkage was applied with the ashr R package (version 2.2‐63). ³⁵

Gene set enrichment‐based functional characterization of computed transcriptomic log2 fold change differences was performed with the gseGO function from the clusterProfiler R package (version 4.14.6) ³⁶ against the human org.Hs.eg.db annotation database (version 3.20.0; Bioconductor). ³⁷ , ³⁸ , ³⁹ Functional over‐representation testing of overlapping gene sets was performed using the enrichGO function from the clusterProfiler R package (version 4.14.6) ³⁶ against the human org.Hs.eg.db annotation database (version 3.20.0). ³⁷ Significant terms from gene set enrichment analysis (GSEA) and over‐representation analyses were mapped onto the TREAT‐AD biological domains (syn25428992.v10). ⁷

Transcriptomic correlation between KD datasets from iNeurons and late‐onset AD (LOAD) data from Accelerating Medicines Partnership for Alzheimer's Disease (AMP‐AD) were computed as follows. Genes were partitioned into biological domains and subdomains based on annotation in GO terms within each domain (syn25428992.v10). ⁷ For each biological domain or subdomain gene set we computed the Pearson correlation between the computed log2 fold change following target gene KD and the fold change from our meta‐analysis of AMP‐AD transcriptomes. ⁷ The significance of each correlation was adjusted for multiple hypothesis testing with Bejamini‐Hochberg correction. ⁴⁰

2.8. Statistical analysis

Statistical analysis of cellular assays was performed using GraphPad Prism. Figures were made using GraphPad Prism and R statistical software. Aβ and pTau measurements were performed on two independent neuronal differentiations and bulk RNA‐seq analysis was performed on one neuronal differentiation.

3. RESULTS

3.1. Validation of shRNA KD of CD44 and MSN expression

A schematic of the hiPSC differentiation and CD44/MSN KD is presented in Figure 1A. To validate shRNA KD of CD44 and MSN expression in iPSC‐derived neurons, quantitative PCR was performed using RNA harvested from transduced neurons. CD44 and MSN show significant KD in both cell lines compared to scrambled control (Figure 1B,C). KD in the CVI cell line resulted in a 95.3% reduction in CD44 expression and a 92.1% reduction in MSN expression (Figure 1B). While significant, KD was less effective in WTC11, with a 61% reduction in CD44 expression and a 70.2% reduced expression in MSN compared with scramble controls (Figure 1C). Lentiviral delivery of CD44 and MSN KD constructs did not affect neuronal viability, as measured by a LDH cytotoxicity screen of conditioned medium (Figure S1). Only CD44 KD in the WTC11 cell background significantly affected viability, with a reduction in cytotoxicity being measured following CD44 expression KD compared to control shRNA (Figure S1B).

3.2. CD44 and MSN KD increases Aβ secretion

To measure the effects of CD44 and MSN KD on amyloidogenic APP processing, Aβ1‐38, 40, and 42 levels were measured using ELISA. CD44 expression KD significantly increased Aβ1‐40 and Aβ1‐42 in neurons of both cell lines (Figure 2A,B). Reducing MSN expression resulted in similar changes, with increased secretion of Aβ1‐40 and Aβ1‐42 (Figure 2A,B). Interestingly, KD of CD44 or MSN results in an increase in the Aβ1‐42/Aβ1‐40 ratio in CVI background neurons but not in WTC11 background neurons (Figure 2C). The significantly greater Aβ1‐42/Aβ1‐40 ratio in the CVI line indicates a shift toward more pathogenic processing of APP. The similar ratio to control shRNA in WTC11 background neurons suggests that overall levels of APP processing are increased, shown by the increased concentrations of Aβ1‐40 and Aβ1‐42, but there is no shift toward a pathogenic increase of Aβ1‐42 (Figure 2C). Aβ1‐38 levels showed trends similar to those of Aβ1‐40 and Aβ1‐42 (Figure S2A). Consistent with an increase in amyloidogenic processing of APP, we observed an increase of soluble APPβ as compared to soluble APPα fragments in the medium from neurons with CD44 and MSN KD (Figure S2B–D). To determine whether the changes in Aβ were due to effects of CD44 or MSN KD on APP or presenilin (PSEN)1/2 expression, we interrogated our bulk RNA‐seq data. Analysis of APP and PSEN1/2 transcripts in the RNA‐seq samples revealed a significant increase in APP expression in both cell lines with CD44 KD. MSN KD showed a trend, but non‐significant, increase in APP expression (Figure S2E). PSEN1 expression was not consistently different from either CD44 or MSN KD, and although PSEN2 expression tended to increase, this was not significant (Figure S2E). This suggests that the increases observed in sAPPβ and Aβ levels with CD44 and MSN KD may be partially explained by increases in APP and gamma‐secretase expression.

FIGURE 2.

CD44 and MSN KD increases secreted Aβ peptides. (A) Aβ1‐40 concentrations of each shRNA neuron sample in CVI and WTC11 cells normalized to total protein. (B) Aβ1‐42 concentrations of shCD44 and shMSN in CVI and WTC11 cells normalized to total protein. (C) Overall ratio of Aβ40:42 peptides after CD44 and MSN KD. Statistical significance calculated using one‐way ANOVA test. For Aβ1‐40: CVI shCD44 p = < 0.0001, shMSN p = 0.0801; WTC11 shCD44 p = 0.0134 shMSN p = 0.0182. For Aβ1‐42: CVI shCD44 p = < 0.0001, shMSN p = 0.0193; WTC11 shCD44 p = 0.0094. shMSN p = 0.0238. For Aβ42:40 ratio: CVI shCD44: p = 0.0155 shMSN: p = 0.0076 WTC11shCD44: p = 0.5526 shMSN: p = 0.3112. Each dot represents the average of two technical replicates of one biological replicate, with two separate differentiations (circle and square data points).

3.3. Intracellular tau expression changes with target gene KD

To measure intracellular tau and pTau levels in KD samples, protein lysate was collected from neurons after CD44 or MSN KD and assayed with ELISA. CD44 KD increases t‐tau and pTau abundance in both CVI and WTC11 cell lines. While MSN KD slightly increased t‐tau, this was only significant in the WTC11 cell line (Figure 3A). Interestingly, CD44 KD significantly increased pTau in both cell lines (Figure 3B). Despite this increase in pTau, the ratio of pTau to t‐tau was either decreased or unchanged with KD of CD44 and MSN in both cell lines, likely due to the increases in t‐tau levels (Figure 3C).

FIGURE 3.

CD44 and MSN KD decreases the pTau to t‐tau ratio in neurons. (A) Concentration of total intracellular tau in neurons following control, CD44, and MSN KD normalized to total protein. Calculated with one‐way ANOVA test. (B) Phospho‐Tau (Thr231) units following control, CD44, and MSN KD normalized to total protein. (C) Overall ratio of pTau/t‐tau. Statistical significance calculated using one‐way ANOVA test. For t‐tau: CVI shCD44 p = 0.0003, shMSN p = 0.3409; WTC11 shCD44 p = 0.0054 shMSN p = 0.0199. For pTau CVI shCD44 p = 0.0005, shMSN p = 0.5726. For pTau/t‐tau ratio: CVI shCD44 p = 0.0405, shMSN p = 0.0003; WTC11 shCD44 p = 0.9211; shMSN p = 0.0007. Each dot represents the average of two technical replicates of one biological replicate, with two separate differentiations (circle and square data points).

3.4. CD44 and MSN KD results in similar transcriptomic changes

Ongoing multi‐omic studies of post mortem AD brains have revealed significant shifts of the transcriptomic and proteomic landscapes with the progression of disease. ⁷ To compare the effects of CD44 and MSN KD on gene expression, we performed RNA‐seq of shRNA‐treated hiPSC neurons. Gene expression was controlled for cell line and batch effects given that cell line differences explained nearly 30% of intra‐sample variance (Figure 4A,B). Our RNA‐seq data confirm expression of CD44 and MSN in both cell lines under control conditions, as well as significant reduction of each gene in the corresponding KD (Figure 4C). Given that both CD44 and MSN were identified as hub nodes in the M4 proteomic module, ⁸ we examined whether other M4 genes were affected by the KD of CD44 and MSN. We first confirmed expression of M4 genes in our dataset and found that 170 out of 183 M4 genes were reliably detected in our hiPSC‐derived neurons (Figure 4D, mean normalized counts > 25 across all samples by DESeq2). With CD44 and MSN KD, we found 22 of the 183 M4 genes with significantly altered expression levels (Figure 4E). Only four M4 genes showed similar changes in expression with both CD44 and MSN KD (Figure 4E bracket, 4F).

FIGURE 4.

RNA sequencing of neurons after CD44 and MSN KD. (A) Principal component analysis (PCA) of RNA sequencing results of CD44 KD, MSN KD, and controls. (B) Plot of top two principal components (PCs) with samples plotted. PC2 is primarily driven by cell line. (C) Expression values of CD44 and MSN in hiPSC‐derived neurons under control and KD conditions. (D) Distribution of expression levels (log2 TPM) for all expressed genes (N = 18,268, pink) versus the 170 genes from co‐expression Module 4 (M4, teal) ⁸ that were found to be reliably expressed in hiPSC neuronal transcriptomes (mean normalized counts > 25 across all samples). (E) UpSet plot of the 22 Module 4 genes found to be differentially expressed following KD of either CD44 or MSN. Each row (lower left) shows a differential expression set (up‐ or downregulated in the CD44 or MSN KD), and each column is an intersection. Most of the M4 genes that are significantly differentially expressed are affected in only one KD; the bracket indicates the four genes (shown individually in panel F) that change in the same direction in both KDs. (F) Box and whisker plot of Module 4 gene expression changes that were found to have similar shifts in both CD44 and MSN KD. BLVRB, CSTB, and DBI had significantly increased expression in both CD44 and MSN KD, while CAPS had a significant decrease in expression in both KDs.

Due to similarities between CVI and WTC11 cell lines in Aβ and pTau generation, we merged cell lines for RNA‐seq analyses rather than looking for interaction effects between the shRNA and cell line. KD of CD44 and MSN in hiPSC‐derived neurons resulted in significant transcriptomic changes. Differentially expressed genes (DEGs) were identified in CD44 and MSN KD neurons in comparison with control shRNA‐treated neurons (Figure 5A); 2932 genes were differentially expressed with MSN KD, and 2207 genes exhibited significant changes in expression in CD44 KD samples. All DEGs of both KDs are shown in Table S1. A significant number of genes showed similar patterns of change in CD44 and MSN KD (Figure 5B). There was a statistically significant overlap in both genes with increased differential expression in both KD conditions (n = 88, p = 1.11e‐85) and genes with decreased differential expression in both KD conditions (n = 52, p = 1.83e‐28) (Figure 5B, Table S2). Functional analysis of the log2 fold change values for all genes in CD44 and MSN KD using GSEA also revealed remarkably similar patterns of enrichment (Figure 5C). Many biological domains showed similar levels of enrichment with both target KDs. Notably, the endolysosome, proteostasis, synapse, and lipid metabolism domains all showed very similar increases in gene expression in both KDs (Figure 5C). Decreased enrichment was detected in structural stabilization, cell cycle, and DNA repair domains. The only biological domain with the opposite direction of enrichment between the two KDs was the immune response domain, which showed increased expression following MSN KD and decreased expression following CD44 KD (Figure 5C). When we looked for over‐represented functions among the DEGs shared between MSN and CD44 KD, we found that cell cycle, DNA repair, and structural stabilization domain terms were enriched among the 52 commonly downregulated genes, while synapse, endolysosome, and proteostasis domain terms were enriched among the 88 commonly upregulated genes (Figure 5D). Notably, the enriched terms among the shared upregulated genes all pointed to increased expression of genes involved in vesicle and trafficking functions. Concept mapping reveals DEG shared between enriched terms (Figure 5E). For example, MCM5 and MCM6 are downregulated in both CD44 and MSN KD and have similar effects in DNA replication related processes. Likewise, both RAB27B and SYNGR2 expression is increased in both CD44 and MSN KD and both are involved with synaptic and exocytotic vesicle functions. These data demonstrate the statistical and functional similarities in transcriptomic shifts found following KD of both CD44 and MSN, reflecting the similarities we observed in Aβ secretion and pTau generation.

FIGURE 5.

CD44 and MSN KDs elicit similar transcriptomic shifts. (A) Volcano plots of DEGs identified in CD44 and MSN KDs in hiPSC neurons. (B) Bar plot of DEGs that are up‐ or downregulated with CD44 and/or MSN KD. There is significant overlap of DEGs that are up‐ and downregulated in both KDs, and much fewer DEGs that change expression in opposite directions. (C) Gene ontology (GO) domain enrichment analysis of significantly altered DEGs in CD44 and MSN KD neurons. The enriched domains of CD44 and MSN KD neurons demonstrate similar patterns of change. Points represent significantly enriched GO terms and the violin plots show the distribution of all significantly enriched terms from each domain. The midline of each violin plot represents the median NES value for significant terms. (D) Subdomain of overlapping increased and decreased enrichments of CD44 and MSN KD. CC, cell cycle, Sy, synapse, En, endolysosome, Pr, proteostasis, DR, DNA repair, SS, structural stabilization. Cell cycle subdomain: microtubule cytoskeleton organization involved in mitosis. Proteostasis subdomain: vesicle‐mediated transport to plasma membrane. Structural stabilization subdomain: microtubule polymerization or depolymerization. Points represent significantly enriched GO terms, and the violin plots show the distribution of all significantly enriched terms from each domain. The midline of each violin plot represents the median −log₁₀ adjusted p value for significant terms. (E) Concept map of DEGs and subdomains of enrichments detected in CD44 and MSN KD experiments.

3.5. CD44 and MSN KD reverses subdomain transcriptomic changes in post mortem AD brains but shows increases in genes regulating APP processing

CD44 and MSN gene expression and protein abundance increase in the progression of AD. ⁸ , ¹⁸ Next, we compared the transcriptomic changes elicited by KD of CD44 and MSN to those observed in disease. We utilized transcriptomic data generated by the AMP‐AD, ⁴¹ which provides publicly available multi‐omic datasets of AD patient cohorts. ⁴² Specifically, we utilized our integrated meta‐analysis of AMP‐AD transcriptomic datasets, described previously. ⁷ We calculated correlations of changes in genes annotated to specific AD biological domains and constituent subdomains between CD44 and MSN KD hiPSC neurons and AMP‐AD LOAD transcriptomic data and found significant correlations in multiple subdomains for each KD target (Figure 6A). Notably, similar subdomains were correlated for both of our KD targets, including for the axon regeneration, phagocytosis, and phosphatidyl inositol metabolic process subdomains. Interestingly, nearly all significant domain correlations we detected exhibited negative correlations between target KD hiPSC neurons and AD post mortem brain samples, indicating a reversal of disease‐related gene expression changes (Figure 6B). This was true of significant correlations detected in both CD44 and MSN KD, with the correlation coefficients of each KD comparison being roughly equivalent (Figure 6B). We found a larger number of significant correlations between CD44 KD and AMP‐AD data (Figure 6A), but, similarly to the overall trend, nearly all of them showed inverse relationships to disease. A closer examination of the phagocytosis, lysosome, and axon regeneration subdomains highlighted the expression patterns of the genes driving these negative correlations (Figure 6C). The axon regeneration subdomain showed the strongest negative correlation with post mortem AD brain gene expression patterns, with, for example, growth‐associated protein 43 (GAP‐43) showing strong downregulation in AD but strong upregulation in hiPSC neurons with reduced CD44 and MSN expression. We next compared the fold change of genes in the phagocytosis, lysosome, and axon regeneration subdomains between CD44 and MSN KDs (Figure 6D). We found positive correlations between each target KD, confirming the similarity of effects in reducing the expression of our two target genes. However, these analyses also help identify the shared signature between the KDs. For example, SLC11A1 is strongly downregulated in CD44 KD but largely unaffected in the MSN KD, while ATP6V1A and F2RL1 are upregulated in both CD44 and MSN KD. Overall, when comparing the transcriptomic changes in hiPSC neurons with lowered CD44 and MSN expression to changes found in post mortem AD brains, we find a reversal of expression in many disease‐associated gene sets. Finally, we observed increased expression across KD conditions of key genes involved in APP/Aβ pathways, including a consistent increase in BACE1 expression (Figure 6E), which may explain the increases we observed in Aβ and sAPPβ from CD44 and MSN KD neurons.

FIGURE 6.

hiPSC‐derived neuron MSN and CD44 KD transcriptomic changes negatively correlate with changes found in post mortem AD brains. (A) Biodomain correlation of CD44 and MSN KD to AMP‐AD post mortem AD transcriptomic data. Numerous subdomains have significant negative correlation to changes observed in AD brains. (B) Correlation between statistically significant domains of CD44 KD comparison to late‐onset AD (LOAD) brain tissue samples versus MSN KD compared to LOAD brain tissue samples. CD44 and MSN KD results show similar correlations with post mortem brain changes. (C) Subdomains and correlations of gene expression are similar in CD44 and MSN KD. Phagocytosis, lysosome, and axon regeneration subdomains with CD44 and MSN KD are negatively correlated with patterns of expression found in AD brains and show similar trends in gene expression changes. (D) Correlations of subdomain enriched genes in panel 6C between CD44 and MSN KD. Changes in these gene sets are positively correlated, reflecting similar changes with reduced expression of either CD44 or MSN. Blue lines with gray clouds in (B), (C), and (D) represent the regression fits and confidence intervals, respectively. (E) Individual expression values of genes involved in APP/Aβ pathways.

4. DISCUSSION

The TREAT‐AD consortium has identified CD44 and MSN as potential therapeutic targets for AD due to proteomic data showing protein levels that are correlated with neuropathological changes in AD brains. ⁸ RNA expression analysis of data collated from large cohort studies (Religous Orders Study/Memory and Aging Project, Mayo Clinic, and Mount Sinai Brain Bank) also reveal increased CD44 and MSN gene expression in AD brain tissue. ⁴³ , ⁴⁴ , ⁴⁵ Here, we used hiPSC‐derived neurons to examine the effects of reduced CD44 and MSN gene expression on AD neuropathological hallmarks, Aβ, and pTau, and transcriptomic profiles. We used two cell lines harboring apolipoprotein E (‐apoE) genotypes ε3/3 and ε3/4, which exhibit distinct phenotypic profiles, to model inter‐individual variability in Alzheimer's disease. ⁴⁶

We measured secretion of Aβ to assess how reducing CD44 and MSN changed APP amyloidogenic processing. We observed increases in all three detected Aβ species, 1‐38, 1‐40, and 1‐42, showing that reduced CD44 and MSN expression increases amyloidogenic processing of APP. There was also an increase in the Aβ1‐42/Aβ1‐40 ratio in the CVI cell line, indicating a push to a more pathogenic form of Aβ. The increase in the Aβ1‐42:Aβ1‐40 ratio further supports an overall detrimental effect on Aβ processing when CD44 and MSN expression levels are reduced in cortical neurons. Both genes are involved in Aβ clearance, ¹⁸ and reducing their expression likely hindered that process. Interestingly, KD of both targets resulted in very similar effects on Aβ secretion, with significant increases in nearly all three Aβ species detected in both cell lines with both KDs. The WTC11 cell line showed no shift in the Aβ1‐42/Aβ1‐40 ratio in contrast to the CVI cell line.

CD44 and MSN are predominantly associated with Aβ accumulation and clearance, but some effort has been made to examine their role in phosphorylation of tau. ²² For example, secreted CD44 has been found to increase tau phosphorylation. ²² We measured levels of intracellular t‐tau and phosphorylation of pTau (Thr231) as our second measure of AD neuropathological hallmarks. Here, our results were more varied than in the measurements of Aβ secretion, with the overall pattern showing increases in the levels of t‐tau and pTau in both cell lines and target KDs. Of note, the ratio of pTau to t‐tau was significantly decreased with KD of both targets in the CVI cell line and with MSN KD in the WTC11 cell line. Both CD44 and MSN are part of extracellular matrix‐related pathways and cell‐to‐cell interaction, and tau strengthens cell scaffolding. ⁴³ , ⁴⁴ Our data suggest that reducing the levels of these genes results in an increased generation of t‐tau but that phosphorylation of tau is simultaneously increased.

Large cohort studies of AD patients are providing insights into the complex changes that arise in the pathology of the disease. Such studies are key to identifying potential genes that may play a role in AD to find potential therapies. We analyzed the changes in transcriptomic state in CD44 and MSN KD hiPSC‐derived neurons to understand how changes in these target genes affect the cell states and how it compares to our understanding of the gene expression patterns found in the disease. We found large transcriptomic shifts with CD44 and MSN KD, and the observed changes were remarkably similar, with a significant overlap of DEGs from both KDs. The similarities in transcriptomic signatures between the two KDs may not be surprising, given that CD44 and MSN interact. ¹⁸ However, the expression level of CD44 did not decrease significantly when MSN was knocked down and vice versa, suggesting some effects may be independent of their interaction. The overlapping effects of CD44 and MSN KD underscore their shared role in affecting synaptic vesicle trafficking and exocytosis, which are processes that are linked with the observed increase in Aβ secretion.

Several of the shared DEGs are known to affect APP processing and Aβ secretion (Figure 6E). BACE1, which encodes β‐secretase, is strongly upregulated after CD44 and MSN KD. GPR3, or G‐protein coupled receptor 3, is upregulated in both MSN KD and CD44 KD, and previous studies showed that overexpression of GPR3 modulated the activity of γ‐secretase and was associated with increased secretion of Aβ both in vitro and in vivo. ⁴⁷ Similarly, RAB27B is upregulated in both MSN KD and CD44 KD. Perturbations of the RAB27B mouse ortholog produces lysosomal defects and accumulation of misfolded alpha‐synuclein, ⁴⁸ and a recent study suggested that inhibition of RAB27B in iPSC‐derived neurons led to a build‐up of Aβ in the lysosomal compartment and decreased secretion to the extracellular space. ⁴⁹ The transcript for RTN4R, the reticulon 4 receptor or Nogo receptor, is also upregulated in both the MSN KD and CD44 KD, and previous work showed that the interaction of Nogo receptor proteins with APP influenced the production of Aβ. ⁵⁰ Finally, two subunits of the lysosomal V‐ATPase (ATP6V1A and ATP6V1B2) are also among the DEGs that are upregulated in both CD44 and MSN KD. Inhibition of the V‐ATPase with bafilomycin A1 results in increased secretion of Aβ, ⁵¹ and both Aβ and phosphorylated tau can disrupt V‐ATPase function neurons, in part through interaction with ATP6V1B2. ⁵² Importantly, endolysosomal dysfunction mediated by these pathogenic proteins was shown to be ameliorated by treatment with hyaluronidase, or by its metabolite hyaluronic acid (HA) and the HA receptor CD44. ⁵² Thus, the transcriptomic effects of perturbation of these two targets partially explains the phenotypic effect of increased secretion of Aβ.

Several transcriptomic AD biological domains were significantly correlated between our CD44 and MSN KD hiPSC neurons and post mortem AD brains. Nearly all of these significant correlations exhibited a negative relationship, notably in the domains of axon regeneration, lysosome function, lipid metabolism, and autophagy. The presynaptic GAP‐43 is important for synaptic plasticity, learning, and memory, and levels decrease in neurons but are increased in the cerebrospinal fluid of AD brains. ⁵³ , ⁵⁴ Similarly, the levels of GAP‐43 were significantly reduced in the post mortem brain dataset we utilized for this study. With KD of CD44 or MSN in hiPSC‐derived neurons, the levels of gene expression of GAP‐43 were significantly increased, helping drive the negative correlation we detected in the AD biodomains of axon regeneration (Figure 6C). Other AD biodomains had similar patterns; for example, ATP6V1A expression is decreased in AD brains and is necessary for lysosomal acidification, and decreased expression results in neuronal dysfunction due to lysosomal build‐up. ⁵⁵ As described earlier, CD44 and MSN KD increases the levels of expression of ATP6V1A, in contrast to the decreased levels found in AD brains (Figure 6C). The inverse relationship found between CD44 and MSN KD gene expression and post mortem AD brain gene expression suggests they may be linked to other AD‐related cellular processes and gene expression.

In the human brain, expression of CD44 and MSN are strongly expressed in microglia and astrocytes, as are other genes in the M4 module; however, they are also expressed in various subtypes of inhibitory and excitatory neurons (Figure S3A,B). It is important to understand their roles in different cell types, and in this work, we examined the role of CD44 and MSN in hiPSC‐derived excitatory neurons, which have detectable expression of many of the M4 genes (Figure 4D). MSN and CD44 KD has downstream genetic effects pertaining to membrane turnover and vesicle transportation in neurons, as seen with GPR3, RAB27B, and RTN4R upregulation, independently of glial cell clearance. Therefore, the increase in neuronal Aβ could be attributed to MSN and CD44 roles not relevant to microglia and astrocytes. Further examination into the effects reducing CD44 and MSN expression in multiple brain cell types is warranted, but the neuronal effects described here are critical for understanding the brain‐wide potential for therapies targeting these genes. It is important to understand the effects in all central nervous system cell types, as treatments not targeted to a cell type will likely have different effects in brain cell types. The proteomic study that initially nominated CD44 and MSN ⁸ was based on bulk proteomics, and small effects in neurons could be lost in the larger changes observed in glial cells. Additionally, here we tested the effects of CD44 and MSN KD in cells in unchallenged conditions, free from AD‐related stresses. The cellular and transcriptomic effects we found here would likely be different in neurons exposed to disease‐relevant conditions, such as increased levels of extracellular Aβ or pTau or in cells carrying AD risk genes. Taking these complexities into account helps clarify the potential role for CD44 and MSN as therapeutic targets.

While CD44 and MSN KD decreased the pTau/t‐tau ratio, the increased individual levels of tau and pTau in addition to increased amyloidogenic APP processing we present here diminishes the potential for CD44 and MSN as therapeutic targets. Despite this, we found a normalization of many disease‐affected gene pathways, which could prove to be beneficial if the detrimental changes to Aβ and pTau could be overcome. This reinforces the need to study multiple cell types in gene therapy, as global KD has varying effects on different cells. Together, these data demonstrate that CD44 and MSN play a role in AD and demonstrate that modulation of genetic expression in hiPSC‐derived brain cell types is an effective model to test the cellular and transcriptomic effects of potential targets for therapeutic development.

AUTHOR CONTRIBUTIONS

Study design: JEY, RB, EHB, CAW. Data acquisition: EHB, CAW, IKAP. Data analysis: GAC, JCW, GWC, EHB, CAW. Writing – original draft: EHB, CAW, GAC. Writing – reviewing and editing: JEY, GAC, EHB, CAW. Funding: AIL, HF, RB, JEY, ELZ, GWC, TREAT‐AD.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Consent was not necessary for the nature of this experiment.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

Supporting Information

ICMJE Disclosure Form