CD33M inhibits microglial phagocytosis, migration and proliferation, but the Alzheimer’s disease‐protective variant CD33m stimulates phagocytosis and proliferation, and inhibits adhesion

CD33 is a Siglec (sialic acid‐binding immunoglobulin‐type lectin) receptor on microglia. Human CD33 can be alternatively spliced into two isoforms: the long isoform (CD33M) and a shorter isoform (CD33m) that lacks the sialic acid‐binding site. CD33m appears to protect against Alzheimer's disease; however, it remains unclear how. To investigate potential mechanisms by which CD33m may confer protection, we expressed the CD33m and CD33M isoforms of human CD33 in mouse BV‐2 and human CHME3 microglial cells and assessed microglia functions. In the BV‐2 cells, CD33M inhibited microglial phagocytosis of beads, synapses, debris and dead cells, while CD33m increased phagocytosis of beads, debris and cells. RNAi knockdown of the endogenous mouse CD33 increased phagocytosis and prevented CD33m's (but not CD33M’s) effect on phagocytosis. CD33M increased cell attachment but inhibited cell proliferation, while CD33m did the opposite. We also found that CD33M inhibited cell migration. In human CHME3 cells, CD33M increased cell attachment, but inhibited phagocytosis, proliferation and migration, whereas CD33m did the opposite. We conclude that CD33M inhibits microglial phagocytosis, inhibits migration and increases adhesion, while CD33m increases phagocytosis, proliferation and inhibits adhesion. Thus, CD33m might protect against Alzheimer's disease by increasing microglial proliferation, movement and phagocytosis of debris and dead cells.


| INTRODUC TI ON
Microglia are brain-resident macrophages and can phagocytose pathogens, debris, protein aggregates, apoptotic cells, neurons and synapses (Galloway et al., 2019;Vilalta & Brown, 2018;Wolf et al., 2017). Microglial phagocytosis is mediated and regulated by several different receptors, including CD33. Human CD33 (also known as Siglec-3) is a transmembrane receptor with an extracellular domain that binds sialic acid residues of glycoproteins on the cell surface, and uses an intracellular ITIM (immunoreceptor tyrosine-based inhibition motif) domain and ITIM-like domain to inhibit immune/ inflammatory functions of cells (Estus et al., 2019;Paul et al., 2000).
In such cells, extracellular cross-linking using antibodies induces tyrosine phosphorylation of the ITIM and ITIM-like domains of CD33, which in turn recruits and activates the tyrosine phosphatases SHP-1 and SHP-2, resulting in inhibition of tyrosine phosphorylation signalling in the cells (Paul et al., 2000;Taylor et al., 1999;Ulyanova et al., 1999).
Human CD33 and mouse CD33 are homologous in their extracellular domains, but have limited homology in their intracellular domains, such that mouse CD33 has an ITIM-like domain, but lacks the ITIM domain present on human CD33 (Brinkman- Van der Linden et al., 2003). Griciuc et al. (2013) reported that both human and mouse CD33 inhibited phagocytosis of amyloid beta (Aβ). Whereas Bhattacherjee et al. (2019) reported that expression of human CD33 inhibited monocyte and microglial phagocytosis (of beads, dextran, myelin and Aβ), but expression of mouse CD33 had no effect on phagocytosis. These results appear inconsistent, so it remains unclear whether mouse CD33 has a similar function to human CD33.
Human CD33 can be alternatively spliced into two major isoforms known as CD33M (also known as FL-CD33), which encodes the full-length protein (containing all seven exons), whereas CD33m (also known as D2-CD33 or CD33ΔV-Ig), the shorter form, lacks the sialic acid-binding domain coded by exon 2 (Hernández-Case lles et al., 2006). Mouse CD33 can also be alternatively spliced into two isoforms but this occurs in the cytoplasmic region and therefore the isoforms are distinct from those in human CD33 (Tchilian et al., 1994). GWAS studies have identified a single-nucleotide polymorphism (SNP, rs3865444A) with decreased risk of Alzheimer's disease (AD) (Bradshaw et al., 2013;Hollingworth et al., 2011;Malik et al., 2013;Naj et al., 2011). Subsequently it was found that rs3865444 (in the promotor of CD33) is in linkage equilibrium with rs12459419 (which is located in the intron close to the start of exon 2 of CD33), and this may be the disease-modifying site by increasing the splicing out of exon 2, thus decreasing the ratio of CD33M to CD33m (Estus et al., 2019). However, the underlying biology of why CD33m and CD33M confer differing susceptibility to developing Alzheimer's disease remains unclear.
One hypothesis is that CD33M inhibits microglial uptake of Aβ, while CD33m is non-functional and therefore does not inhibit phagocytosis of Aβ, resulting in increased Aβ uptake, reduced amyloid plaque load and disease (Bradshaw et al., 2013;Griciuc et al., 2013). Griciuc et al. (2013) found increased microglial expression of CD33M in Alzheimer's brains, and a positive correlation between microglial expression of CD33M and plaque load, suggesting that CD33M inhibits microglial phagocytosis of plaques. Consistent with this, CD33 knockout mice crossed with an amyloid model (APP/ PS1) had reduced plaque load. Furthermore, Griciuc et al. (2013) showed that microglia from CD33 knockout mice took up more Aβ, indicating that endogenous mouse CD33 inhibited microglial uptake of Aβ. They also reported that over-expression of human CD33M in BV-2 microglial decreased Aβ uptake, while expression of human CD33m had no effect on Aβ uptake (Griciuc et al., 2013). Bradshaw et al., (2013) compared humans with the risk and protective alleles of CD33, and found that the risk allele for AD was associated with: increased surface expression of CD33M on monocytes, decreased Aβ uptake by monocytes and increased fibrillar Aβ in both live brain and post-mortem brains. The post-mortem brains had increased neuritic pathology and activated microglia (Bradshaw et al., 2013). Overall, these findings and those of Griciuc et al. (2013) are consistent with the hypothesis that the protective allele increases expression of CD33m relative to CD33M, and that CD33M inhibits microglial phagocytosis of Aβ, while CD33m is nonfunctional, potentially explaining why CD33M expression is detrimental in AD, whereas CD33m is protective.
The above model assumes that CD33m has lost function compared to CD33M. However, recently a variant of the CD33 gene was found resulting in no CD33 expression and no change in AD risk.
This finding is not consistent with CD33m being protective simply by loss of CD33M function, and therefore suggests the possibility that CD33m might gain some unknown function relative to CD33M to protect against AD (Estus et al., 2019;Papageorgiou et al., 2019).
Here, we sought to elucidate the roles of CD33 in microglial function and AD by determining: (1) whether CD33M and CD33m have differential effects on microglial phagocytosis of a variety of different phagocytic targets, including different forms of Aβ, (2) whether CD33 variants affect microglial migration, attachment and proliferation, (3) whether CD33m has a loss-of-function or gain-offunction effect on microglia, (4) whether mouse and human CD33 have different effects on microglia functions, and finally (5) whether human CD33 has different effects on mouse and human cells. To address these aims, we expressed human CD33M and CD33m in mouse and human microglial cell lines (BV-2 and CHME3 respectively) and measured microglial functions, such as phagocytosis, migration, attachment and proliferation.

| Ethical statement
Explicit institutional ethical approval was not required for this study. The ARRIVE guidelines were not applied since no animals are involved.

| Study design
This study was not pre-registered and no randomisation, blinding or sample calculation was performed. The study was exploratory and no exclusion criteria were pre-determined. For all experiments, unless stated otherwise, at least n = 3 independent cell culture preparations were used.

| RNA extraction and qPCR
Whole RNA was extracted using Monarch total RNA Miniprep kit (NEB, T2010). cDNA was synthesised using 1 µg of whole RNA, Superscript II Transcriptase kit (Invitrogen, 18064022) and random hexamer primers (ThermoFisher, SO142). qPCR was performed using SYBR Green JumpStart Taq ReadyMix (Sigma, S4438). Amplification was performed using Rotor-gene Q cycler (Qiagen). Data were analysed using rotor-gene software and data expressed as comparative concentration which was normal-

| Phagocytosis assays
For pH-Rodo (ThermoFisher Scientific, P36600) stained isolated synapses, neuronal debris and Alexa Fluor 555-labelled amyloid beta (Anaspec, ANA60480-01). Cells were seeded in triplicate 24 hr prior to target addition at 20,000 cells per well in DMEM supplemented with 0.5% heat-inactivated FBS and 1% penicillin/ streptomycin. Preparation of monomeric, oligomeric and fibrillar amyloid beta was conducted as described previously (Neniskyte et al. 2011). Isolated synapses were prepared as described by Dunkley et al (2008). Neuronal debris were obtained by scraping cells from live neuronal-glial culture (85% neurons), cells were passed through a 0.4 × 13-mm needle 10 times, then fluorescently labelled with TAMRA (Nomura et al., 2017). For uptake into mouse BV-2 cells, we used rat neuronal debris because rat and mouse are genetically similar, and more neurons can be obtained from rats (consistent with the 3Rs). Whereas for human CHME3 cells, we made neuronal debris from human SH-SY5Y cells (RRID:CVCL_0019). SH-SY5Y cells were cultured to confluency, then harvested using trypsin and resuspended in 1X PBS. Cells were passed through 0.4 × 13-mm needle 10 times, then fluorescently labelled with TAMRA. Protein concentration of both rat and human neuronal debris was determined using Pierce BCA assay (ThermoFisher Scientific, 23227).
For uptake of fluorescent sky blue five micron carboxylated beads (Spherotech, FP-0570-2) and pH-Rodo-labelled PC12 (RRID:CVCL_0481, ECACC) cells were seeded, in duplicate, 24 hr prior to target addition at 50,000 cells per well in DMEM supplemented with 0.5% heat-inactivated FBS and 1% penicillin/ streptomycin. For 0.005% w/v fluorescent sky blue five micron carboxylated beads and 300,000 live or dead pH-Rodo-labelled PC12 incubation time was 3 hr with BV-2 cells. For 16 µg of isolated synapses, 30 µg of rat neuronal debris and 2 µM of amyloid beta incubation time was 1 hr with BV-2 cells. Finally, for 60 µg of human SH-SY5Y neuronal debris, incubation time was 3 hr with CHME3 cells.
After incubation, cells were washed once with 1X PBS to remove non-phagocytosed cargo and then harvested using trypsin, pelleted and finally resuspended in 1X PBS for flow cytometry (Accuri C6, BD systems). To check the contribution of phagocytosis to apparent uptake, we inhibited phagocytosis with 10-µM cytochalasin D for 1 hr before adding phagocytic targets ( Figures   S3-S7).

| Sialidase treatment and assessment of desialylation
For cells treated with sialidase enzyme, 5% exogenous neuraminidase enzyme (Sigma, N3001-4UN) was added to cells for 1 hr before adding phagocytic target. To assess extent of desialylation BV-2 cells were stained with FITC-conjugated peanut agglutinin (PNA), which binds to the penultimate sugars of desialylated glycans (galactose followed by N-acetylglucosamine residues).
Sialylation blocks PNA binding and therefore increased binding indicates desialylation.

| Flow cytometry
For flow cytometry, an Accuri C6 flow cytometer was used (BD systems). At least 5,000 events were analysed for each treatment replicate. FlowJo (version 10) was used for creating raw histograms and flow plots. Fc blocking antibodies were not used. Forward and side scatter was used to distinguish cells from unphagocytosed targets, i.e., beads, debris and synapses, by gating on cells in the forward and side scatter plots. Within this scatter gate, a fluorescence gate was set to identify cells that were over a threshold of fluorescence. For phagocytic targets, this fluorescence gate was set so that for cells incubated in the absence of any fluorescent targets (i.e. absence of fluorescent beads, debris, synapses, Aβ or targets cells), 99% of the cells were below this gated fluorescence and 1% were above this gate. So, for cells incubated in the presence of fluorescent targets, the % of cells with fluorescence greater than the gated fluorescence were used as the measure the % of cells that had phagocytosed the targets. For detection of GFP (green fluorescent protein), PE (phycoerythrin) and pH-rodo, a 488-nm excitation laser was used, and fluorescence emission was measured using the following channels/ detectors: FL-1/FITC, FL-2/PE, FL-3/PerCP-A respectively. For detection of sky blue fluorescent beads, a 640-nm excitation laser and FL-4/APC-A detector were used. For quantification of CD33 protein on the surface of cells, forward and side scatter was used to identify cells, and the FL-2/PE fluorescence of IgG control-PE treated cells was used to gate the fluorescence of anti-CD33-PE-treated cells.

| Attachment/xCELLigence assay
For attachment assays, E-plates (ACEA biosciences, 00300600900) 96-well plates were used. First, the plate was coated with 10 µg/ml human fibronectin (Sigma, F0895) for at least 1 hr at room temperature (18℃). The plate was then blocked with 3% BSA for 1 hr at room temperature (18℃). The plate was then washed three times with 1X PBS, followed by addition of cells (15,000 per well). Real-time cell impedance measurements were collected using xCELLigence software (ACEA biosciences) for at least 4 hr.

| Migration assay
Cells were seeded at 20,000 per well in DMEM supplemented with 0.5% heat-inactivated FBS and 1% penicillin/streptomycin in the top compartment of trans-well apparatus (Coring, 3422). Cells were agitated slowly for 2 min and then allowed to settle in incubator for at least 15 min before adding treatments to the bottom compartment.
Treatments were as follows: 100 µM ATP and DMEM supplemented with 0.5% heat-inactivated FBS and 1% penicillin/streptomycin. DMEM supplemented with 0.5% FBS was used as a control so there would be no concentration gradient between the top and bottom compartment of the trans-well. BV-2 cells were incubated for 2 hr, while CHME3 cells were incubated for 6 hr, both at 37℃, 5% CO 2 .
Cells were fixed with 4% paraformaldehyde (PFA) and stained with 4′,6-diamidino-2-phenylindole (DAPI), then imaged using an EVOS microscope system (Thermofisher). The number of cells present on the bottom of the membrane (and therefore had migrated through the trans-well) was determined by the number of DAPI-positive nuclei.

| Proliferation assays
For both BV-2 and CHME3s, cells were seeded at 100,000 per ml (final volume of 10 ml) in a T-75 flask and cultured in DMEM supplemented with 10% heat-inactivated FBS and 1% penicillin/streptomycin for 48 hr. Cells were then harvested with trypsin and counted using a haemocytometer. To ensure BV-2 cells were not proliferating during the 24-hr period between seeding and phagocytosis assays, cells were plated at 100,000 per ml in 24-well plate in DMEM supplemented with 0.5% heat-inactivated FBS and 1% penicillin/ streptomycin for 24 hr. They were then harvested with trypsin and counted using a haemocytometer.

| Statistical analysis
The statistical analysis was performed on GraphPad prism (version 8.0) data shown represented as a mean of at least n = 3 independ-

| Knockdown of endogenous mouse CD33 or desialylation of cells increases phagocytosis by BV-2 cells
As expression of human CD33M in mouse BV-2 cells inhibited phagocytosis, we tested whether knockdown of the endogenous mouse synapses (CD33 no sialidase 13 ± 3, after sialidase treatment 10 ± 2, Figure 2c). This suggests that the effect of desialylation on phagocytosis could be partly mediated via CD33. However, desialylation still increased phagocytosis of beads and decreased phagocytosis of synapses after CD33 knockdown, so desialylation appears to increase phagocytosis at least partly by CD33-independent means.

| Knockdown of endogenous mouse CD33 prevents human CD33m increasing phagocytosis in BV-2 cells
One possible means by which human CD33m expression might increase phagocytosis is by disrupting the function of endogenous mouse CD33 (e.g. by dimerising with it). To test this possibility, we compared the effect of human CD33 expression on phagocytosis in the presence or absence of endogenous mouse CD33 knockdown using RNAi. We chose to examine the effect of knockdown and expression on phagocytosis of neuronal debris as this may be the most physiological relevant target of phagocytosis, and had a robust response to CD33 expression ( Figure 1b). As expected, in BV-2 cells treated with a control, non-targeting RNAi, CD33M and K7R-CD33M reduced phagocytosis, and CD33m increased phagocytosis. Phagocytosis is expressed as mean % of cells phagocytosing debris ± SEM were: control 12 ± 2, CD33M 8 ± 1, CD33m 18 ± 2, K7R-CD33M 6 ± 1 (Figure 3a). However, in cells treated with RNAi to knockdown endogenous mouse CD33, CD33m did not change phagocytosis, while CD33M and K7R-CD33M still reduced phagocytosis: control 28 ± 5, CD33M 18 ± 3, CD33m 29 ± 5, K7R-CD33M 11 ± 1 (Figure 3b). This is consistent with human CD33m increasing phagocytosis by disrupting the function of endogenous mouse CD33.

| K7R-CD33M increases cell attachment to fibronectin, whilst CD33m decreases attachment to fibronectin in BV-2 cells
Microglial migration is in part determined by cellular adhesion to extracellular matrix, so we tested whether the CD33 variants affected microglial adhesion to fibronectin (which is a major component of brain extracellular matrix) measured by the increase in impedance as the cells adhere and spread on fibronectin-coated plates. K7R-CD33M increased adhesion to fibronectin (K7R-CD33M 194 ± 27% of control ± SEM, Figure 5), while CD33m reduced adhesion (CD33m 51 ± 9% of control ± SEM, Figure 5). CD33m reduced cell spreading on the same matrix measured as the area of the plates occupied by the cells (Figure S9). Thus, CD33M over-expression appears to increase matrix adhesion and CD33m reduces matrix adhesion.

| CD33M inhibits migration of BV-2 cells towards ATP
Phagocytosis of targets normally requires migration to those targets, so we tested whether the CD33 variants affected microglial migration through a trans-well in response to a chemotactic gradient of ATP. ATP increased migration through the trans-well by control and CD33m-expressing microglia. Results expressed as cell numbers migrated ± SEM. (ATP control: 1,095 ± 179, ATP + CD33m: 905 ± 135, Figure 6a), but did not significantly stimulate migration of cells expressing CD33M or K7R-CD33M (ATP + CD33M 568 ± 112, ATP + K7R-CD33M: 598 ± 133, Figure 6a). This indicates that CD33M inhibits microglial migration to ATP.
Because the CD33 variants affected cell proliferation measured as cell numbers after 48 hr in 10% serum, we checked whether the CD33 variants affected cell numbers in the conditions used to assay phagocytosis by plating BV-2 cells in DMEM supplemented with 0.5% serum and counting cell numbers after 24 hr. We saw no differences in cell numbers between the different lines in these conditions ( Figure S10b), indicating that any apparent differences in phagocytosis between the variants were not owing to differences in cell number.

| CD33M decreases phagocytosis and proliferation but increases adhesion to fibronectin in human CHME3 cells
To test whether the effects of expressing human CD33 variants in mouse BV-2 cells were also seen in human cells, we stably We confirmed expression of the CD33 variants at the level of both mRNA and cell surface protein CHME3 after lentiviral transfection of the CHME3 cells ( Figure S11 & S12).
We assessed proliferation of CHME3 cells expressing the different CD33 variants by seeding the same cell numbers then counting cells 48 hr later. We found that expression of CD33M and K7R-CD33M decreased the proliferation rate, whereas expression of CD33m increased the proliferation rate compared to the control cells. Results expressed as cell number per mL ± SEM (control 290 ± 12, CD33M 225 ± 10, K7R-CD33M 140 ± 12, CD33m 405 ± 17, Figure 7b). Again, these data are consistent with the findings seen in the BV-2 cells (Figure 6b). Note that CHME3 cells proliferate at a slower rate than BV-2 cells, and therefore the cell numbers were lower, but the same effect of CD33M decreasing F I G U R E 5 K7R-CD33M increases and CD33m decreases cell attachment to fibronectin. Attachment of human CD33 variant-expressing BV-2 cells to fibronectin over 30 min, measured by electrical impedance of cell layer. n = 4 independent cell culture preparations. Statistics: two-way ANOVA. * is with respect to control cells. ***p <.001, ****p <.0001 proliferation and CD33m increasing proliferation was found in the two cell lines.
To assess migration of CHME3 cells expressing the CD33 variants, we counted cells migrating through a trans-well in response to an ATP gradient, as previously for BV-2 cells. CD33M had a tendency to decrease migration, and CD33m had a tendency to increase migration, but due to the variability of the control, the only significant difference was between CD33M and CD33m. Results expressed as mean % of migrated cells ± SEM were: control 2,485 ± 560, CD33M 1,326 ± 123, CD33m 3,759 ± 434 and K7R-CD33M 1,086 ± 87 ( Figure 7c). These findings are consistent with those seen in BV-2 cells (Figure 6a), i.e. CD33m has the opposite effect to CD33M on migration.
Finally, to assess the effect of CD33 variants on cellular adhesion, we measured the rate of adhesion to fibronectin-coated plates as before, but for a longer period of time (4 hr

| D ISCUSS I ON
CD33 is an inhibitory receptor expressed on myeloid cells, including microglia, potentially activated by sialic acid residues (Lajaunias et al., 2005). CD33m is thought to have a reduced function relative to CD33M, because of this splice variant lacking the sialic acid-binding domain (Griciuc et al., 2013;Siddiqui et al., 2017). Consistent with this, it was reported that CD33M inhibited microglial phagocytosis of Aβ, while CD33m had no effect. This could explain why expression of CD33m, rather than CD33M, apparently protects against AD, as it has lost the capacity to inhibit microglia phagocytosis (Griciuc et al., 2013;Siddiqui et al., 2017). However, we show that expression of CD33m actually increases microglial phagocytosis of multiple   (Siddiqui et al., 2017) or CD33m binding SHP-1/2 without activating them and thereby inhibiting their activation. As we see a small proportion of CD33m present on the cell surface but a large expression of CD33m mRNA in both BV-2 and CHME3 cells (Figures S1, S2, S11 and S12), it is feasible that CD33m is more highly expressed intracellularly, although we have not tested this.
It could be argued that the various phagocytic targets we have used, including Aβ, beads, neuronal debris, isolated synapses and dead cells, are all forms of "debris". It is generally thought that extracellular debris accumulates in the brain with neurodegenerative diseases (Takahashi et al., 2007;Neumann et al., 2008;Marquez-Ropero et al. 2020). Neurodegeneration may generate debris as dead neurons and bits of neurons, but such debris might also induce neurodegeneration, for example, by activating microglia (Map his et al., 2015;Spangenberg et al., 2016). Thus, one possible means by which CD33M may predispose to AD is by blocking microglial phagocytosis of the debris accumulating during neurodegeneration (or ageing), thereby exacerbating neuroinflammation.
TREM2 is one of the main phagocytic receptors mediating microglial phagocytosis of debris (Takahashi et al., 2005), and loss-offunction mutations in TREM2 predispose to AD (Sudom et al., 2018), and this might potentially be due to reduced microglial phagocytosis of debris. CD33 could potentially antagonise phagocytic signalling by TREM2 (or CR3), as TREM2 and CR3 activate phagocytosis via inducing phosphorylation of the tyrosine-protein kinase Syk, while CD33 activates the tyrosine-protein phosphatases SHP-1 and SHP-2 (Paul et al., 2000), which may dephosphorylate Syk and Syk substrates. This may be one potential mechanism by which CD33M is inhibiting phagocytosis in our study. Consistent with this, Griciuc et al. (2019) showed that in the 5XFAD mouse model of AD, TREM2 knockout increased the pathology, while CD33 knockout reduced the pathology, but knocking out both TREM2 and CD33 resulted in the increased pathology of the TREM2 knockout, suggesting that TREM2 was acting "downstream" of CD33 to induce these effects.
These results are consistent with CD33 simply suppressing TREM2 signalling via activating SHP-1/2 (as CD33 then requires TREM2 signalling to have an effect), but they do not rule out more complex interactions between CD33 and TREM2.
Mouse and human CD33 have homologous extracellular domains, and divergent intracellular domains, but whether their functions diverge is less clear. Bhattacherjee et al., (2019) found that expression of human CD33 inhibited phagocytosis, but expression of mouse CD33 did not inhibit phagocytosis (Bhattacherjee et al., 2019). However, Gricuic et al., (2013) found that knockout of CD33 increased microglial phagocytosis of amyloid beta, and this is consistent with our finding that RNAi knockdown of mouse CD33 increased phagocytosis of beads, debris and synapses. These discrepancies between groups could be because of different experimental techniques, cells, animals, levels of CD33 expression or levels of cell sialylation.
CD33 is thought to be an inhibitory receptor potentially activated by sialic acid residues of other glycoproteins (Lajaunias et al., 2005).
We find that desialylation of microglia by adding sialidase increases phagocytosis of beads, consistent with previous findings (Allendorf et al., 2020;Nomura et al., 2017;Pluvinage et al., 2019), but decreased phagocytosis of synapses. RNAi knockdown of mouse CD33 reduced the effect of desialylation on phagocytosis, suggesting that CD33 may contribute to this effect, but the data were not robust enough to make a firm conclusion. Recently, Pluvinage et al., (2019) showed that CD22 (Siglec-2) was required for the effect of sialylation/desialylation on microglial phagocytosis (Pluvinage et al., 2019); and Allendorf et al. (2020) found that complement receptor 3 was also required (Allendorf et al., 2020). As different phagocytic receptors may be required for phagocytosis of beads and synapses, it seems reasonable that desialylation may affect these different receptors differently, resulting in increased phagocytosis of beads but decreased phagocytosis of synapses.
Microglial phagocytosis normally requires microglial migration, so we tested the effect of CD33 on migration of microglia through trans-wells in response to a chemotactic gradient of ATP. We found that expression of human CD33M in both cell types inhibited microglial migration towards ATP, whereas expression of CD33m did not inhibit migration. CD33M also increased microglial attachment to fibronectin in CHME3 cells but only with supraphysiological levels of CD33M (K7R-CD33M) in BV-2 cells. Whereas CD33m reduced both attachment to fibronectin in both cell lines and cell spreading on fibronectin in the BV-2 cells. Together, these findings suggest that over-expression of CD33M increases microglial attachment, which may reduce migration, and potentially reduce phagocytosis.
Increased adhesion has been linked to reduced migration in a number of cell types (DiMilla et al., 1991;Palecek et al., 1997;Schmidt & Friedl, 2010). Thus, it is possible that the increase in adhesion induced by CD33M might contribute to reduced migration found in the presence of ATP. Other Siglecs have been shown to inhibit migration, including Siglec-E in inhibition of neutrophil and monocyte recruitment, and CD22 in inhibition of B cell migration and adhesion (McMillan et al., 2013;Nitschke et al., 1997). CD33 has been shown to mediate adhesion of CD33-expressing cells to erythrocytes via binding to sialic acid residues on the erythrocytes (Freeman et al., 1995). It could be that over-expression of CD33M increases cellular adhesion and spreading by binding sialic acid residues on fibronectin, but we do not think this is the case, as sialidase treatment of the fibronectin to remove sialic acid residues did not significantly decrease CD33Minduced adhesion in BV-2 cells (data not shown).
Another feature of microglial activation is proliferation: microglia can proliferate in response to damage, pathogens or inflammation.
We found that CD33M inhibits proliferation, whereas CD33m stimulates proliferation in both cell lines. Increased proliferation in the presence of CD33m potentially means more microglia are present to clear debris and Aβ, and thereby potentially protecting the brain in AD. Cellular attachment can induce proliferation in some cell types (Pugacheva et al., 2006). A potential role of CD33 in proliferation has been shown as activation of CD33 by anti-CD33 antibodies reduced proliferation of leukaemic cells (Balaian et al., 2003;Vitale et al., 1999).
The microglial processes investigated in this study, proliferation, migration and phagocytosis, are all key aspects of microglia activation, and all are inhibited by CD33M. Microglia activation is a hallmark of neurodegenerative diseases such as AD. Our findings suggest a potential novel role of CD33m in increasing microglial proliferation, movement and phagocytosis of debris and Aβ, which might contribute to protecting the AD brain. However, further work is required to determine the molecular mechanisms by which CD33m and CD33M confer these differential effects on microglia.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.
How to cite this article: Ann Butler C,