The LRRK2 kinase substrates RAB8a and RAB10 contribute complementary but distinct disease-relevant phenotypes in human neurons

Summary

Mutations in the LRRK2 gene cause familial Parkinson’s disease presenting with pleomorphic neuropathology that can involve α-synuclein or tau accumulation. LRRK2 mutations are thought to converge upon a pathogenic increase in LRRK2 kinase activity. A subset of small RAB GTPases has been identified as LRRK2 substrates, with LRRK2-dependent phosphorylation resulting in RAB inactivation. We used CRISPR-Cas9 genome editing to generate a novel series of isogenic iPSC lines deficient in the two most well-validated LRRK2 substrates, RAB8a and RAB10, from deeply phenotyped healthy control lines. Thorough characterization of NGN2-induced neurons revealed opposing effects of RAB8a and RAB10 deficiency on lysosomal pH and Golgi organization, with isolated effects of RAB8a and RAB10 ablation on α-synuclein and tau, respectively. Our data demonstrate largely antagonistic effects of genetic RAB8a or RAB10 inactivation, which provide discrete insight into the pathologic features of their biochemical inactivation by pathogenic LRRK2 mutation in human disease.

Graphical abstract

Highlights

• •RAB8a and RAB10 deficiency induce divergent lysosomal and Golgi defects
• •RAB8a and RAB10 KO neurons show isolated effects on α-synuclein and tau, respectively
• •Inactivation of different RAB GTPases can induce distinct disease-relevant phenotypes

Here, LaVoie and colleagues characterize isogenic iPSC-derived neurons deficient in the two most well-characterized LRRK2 substrates, RAB8a and RAB10. Data revealed independent Parkinson’s disease-relevant phenotypes, with frequent opposing effects associated with RAB8a and RAB10 KO. While proteostasis was affected in both cases, dominant effects were notably seen in for α-synuclein and tau in RAB8a and RAB10 KO neurons, respectively.

Introduction

Missense mutations in the LRRK2 gene are the most common cause of inherited Parkinson’s disease (PD), with this locus further linked to idiopathic PD risk (Barrett et al., 2008; Bonifati, 2006; Paisán-Ruíz et al., 2004). LRRK2 is a large protein with functional GTPase and kinase domains within which PD-linked mutations largely cluster (Kluss et al., 2019). LRRK2-PD can manifest with pure nigral degeneration or classic Lewy body pathology typical of idiopathic PD, but surprisingly a slight majority of cases show deposition of tau into neurofibrillary tangles (Henderson et al., 2019; Herbst et al., 2022; Ujiie et al., 2012; Zimprich et al., 2004). How LRRK2 pathobiology might provoke distinct and independent neuropathological features is one of the greatest unanswered questions in the field.

A subset of ∼14 RAB GTPases are among the most well-established LRRK2 substrates. LRRK2-dependent phosphorylation at a highly conserved threonine residue in their GTPase switch domain results in a net inhibition of RAB activity (Pfeffer, 2018; Steger et al., 2016). RAB GTPases regulate intracellular membrane and protein trafficking and confer membrane identity, organelle morphology, and organelle protein content (Pfeffer, 2018; Stenmark, 2009). RAB8a and RAB10 are the most well-validated LRRK2 substrates to date. RAB8a has been linked to neurodegeneration via its interactions with α-synuclein and PINK1 (Vieweg et al., 2020; Yin et al., 2014), in addition to phosphorylation by LRRK2 (McFarland et al., 2018; Steger et al., 2016; Vieweg et al., 2020). Both RAB8a and RAB10 have been implicated in cellular responses to lysosomal injury (Bonet-Ponce et al., 2020; Eguchi et al., 2018; Mamais et al., 2021). Interestingly, opposing roles have been proposed for RAB8a and RAB10 in normal ciliogenesis (Dhekne et al., 2018; Madero-Pérez et al., 2018), but such antagonistic properties have not been explored in other contexts. RAB mutations have been linked to diseases including cancer, Charcot-Marie-Tooth disease type 2B, and Warburg Micro syndrome (Banworth and Li, 2018). Thus, their inactivation by mutant LRRK2 is well-positioned to broadly impact cellular proteostasis and may contribute to neurodegenerative disease.

Here, we used CRISPR-Cas9 genome editing to ablate RAB8a or RAB10 expression in iPSCs derived from deeply phenotyped (clinically and genetically) wild-type healthy control human subjects (Bennett et al., 2018; Lagomarsino et al., 2021). Isogenic clones from two independent control lines were differentiated into cortical neurons by NEUROGENIN-2 (NGN2) expression to produce induced neurons (iNs). We report a significant decrease in lysosomal number across both RAB8a knockout (KO) and RAB10 KO neurons compared with their isogenic controls. However, RAB8a KO neurons exhibited mild lysosome acidification, whereas RAB10 KO lysosomes were mildly alkaline. Altered glycosylation of LAMP1 was noted in both KO models, prompting an investigation into Golgi integrity that revealed a compressed, or clustered Golgi distribution in RAB8a but a more dispersed Golgi in RAB10 KO iPSCs. Interestingly, LAMP1 retention at the Golgi was only observed in RAB8a KO cells.

LRRK2-PD most commonly manifests with either classic Lewy body inclusions composed of α-synuclein, or neurofibrillary tangles composed of hyper-phosphorylated tau, prompting us to look at the consequence of RAB8a or RAB10 deficiency on these two proteins. Consistent with other opposing phenotypes linked to these two LRRK2 substrates, we found features of α-synuclein homeostasis altered in RAB8a but not RAB10 KO neurons, with selective effects on tau in RAB10 KO neurons. These data identify distinct cellular pathways that can contribute to the unique pleiomorphic pathology associated with LRRK2-PD and may provide a framework to begin to understand why co-morbid α-synuclein and tau pathology is not commonly reported (Henderson et al., 2019; Herbst et al., 2022; Ujiie et al., 2012; Zimprich et al., 2004).

Results

Divergent lysosomal alterations in RAB8a KO and RAB10 KO human neurons

To establish neuronal models of RAB8a and RAB10 deficiency we used CRISPR-Cas9 gene editing to generate homozygous-null lines of the two genes in two independent wild-type (WT) healthy control iPSC lines. Successful KO of RAB8a and RAB10 were initially tested by immunoblotting on different clones from the two control lines, BR01 and BR33 (Figure S1). Sanger sequencing validated two homozygous-null RAB8a clones on the BR33 background while the BR01 RAB8a KO clones were heterozygous null and thus were not used for further experiments. In turn, homozygous-null RAB10 clones were validated on one of each BR01 and BR33 background lines and used for subsequent experiments (Figure S1). Cells were differentiated into mature cortical glutamatergic neurons by forced expression of NGN2 via lentiviral gene delivery, according to published protocols (Sanyal et al., 2020; Zhang et al., 2013). Neurite and presynaptic phenotyping were performed to characterize differentiation and synaptic density across the different lines. MAP2 staining revealed no difference in longest neurite or total neurite length across the different RAB KO and WT control lines (Figures S2A–S2C). Furthermore, while a trend of a divergence was noted between the density of the presynaptic marker Bassoon in RAB8a KO and RAB10 KO lines, there was no significant difference compared with WT controls (Figures S2D and S2E). Initially, we set out to determine whether loss of these RAB GTPases affects lysosome numbers, by LAMP1 and LysoTracker staining (Figure 1). LAMP1 staining indicated a drop in lysosome numbers in both RAB8a KO and RAB10 KO neuron models (Figure 1A). High-content imaging was used to assay lysosomal integrity readouts across the four RAB KO lines and compared with the common WT control line BR33. High-content imaging of the LysoTracker deep red cell dye showed a significant decrease in RAB8a KO iNs (∼60%) in the total number of lysosomes (Figures 1B and 1C), while the average lysosomal area of individual lysosomes remained unaffected (Figure 1D). Lysosomal homeostasis was investigated further by assaying lysosomal pH using LysoSensor, a ratiometric pH-sensitive dye. RAB8a KO neurons exhibited significantly more acidic lysosomal lumen compared with isogenic controls (Figure 1E). In a similar line of experiments, we observed a significant decrease in lysosomal numbers in RAB10 KO compared with WT neurons, while lysosomal size was largely unaffected, as in RAB8a KO neurons (Figures 1F–1H). In contrast, we observed alkalinization of lysosomal pH in RAB10 KO, suggesting a divergent effect to RAB8a deficiency (Figure 1I).

Figure 1.

RAB8a KO and RAB10 KO human neurons have altered lysosomal morphology and function

(A) Representative confocal images of RAB8a KO, RAB10 KO, and isogenic WT control neurons stained with LAMP1, MAP2, and DAPI.

(B–I) Lysosomal parameters including lysosomal count, average lysosomal area, and lysosomal pH were assessed by high-content imaging of LysoTracker red or LysoSensor staining.

(J and K) Western blot analysis of levels of glycosylated LAMP1. All lysosomal analyses were collated from three to four independent differentiations in separate 96-well plates, with more than eight wells per genotype per plate, on 21-day-old iNs. Replicates in the graphs represent wells across four independent differentiation batches (>32 per line). ^∗p < 0.05; ^∗∗p < 0.001; ^∗∗∗p < 0.0001; ^∗∗∗∗p < 0.0001; one-way ANOVA Tukey’s post hoc.

As a marker of lysosomal integrity, we quantified the levels of LAMP1 by immunoblotting. RAB8a KO and RAB10 KO neurons exhibit lower levels of LAMP1 protein, compared with the isogenic control lines, with altered migration of the glyco-protein observed independent of background (Figures 1J and 1K). Our data indicate a parallel effect of RAB8a and RAB10 deficiency in lysosomal numbers and levels of glycosylated LAMP1 protein, but a divergent effect on lysosomal pH in human neurons. We have reported altered LAMP1 glycosylation in in vivo models of LRRK2 activity as well as GCase1 heterozygous-null human neurons previously (Kluss et al., 2020; Sanyal et al., 2020). RAB8a and RAB10 are involved in vesicle trafficking between the trans-Golgi network (TGN), a central organelle for protein glycosylation, and the plasma membrane. Dysregulation of LAMP1 glycosylation in our RAB GTPase KO cell models suggested impairment in TGN-related processes, prompting us to assay Golgi integrity. The morphology of the cis-Golgi network (CGN; GM130) and TGN (TGN46) network, as well as the distribution of individual Golgi stacks and colocalization with LAMP1, were assayed by confocal microscopy. After 3D rendering of z stack images (Imaris platform; Bitplane), CGN and TGN stack size, sphericity, ellipticity, relative proximity to each other, and proximity to the nucleus were measured in WT BR33 and the isogenic null RAB8a KO and RAB10 KO iPSCs (Figures 2 and S3). RAB8a KO iPSCs exhibited a clustered Golgi morphology with significantly reduced (−16%) mean distance between CGN stacks compared with WT cells, while an opposing phenotype was observed in RAB10 KO cells with increased (+8%) mean intra-stack distance reflecting a more dispersed morphology (Figures 2A and S3). Both RAB GTPase null models exhibited a decrease in total Golgi volume per cell and closer proximity to the nucleus of CGN and TGN stacks (Figures 2C–2F and S3). Given the impact of RAB8a and RAB10 KO on lysosomal pH, we considered LAMP1 localization. Analysis of LAMP1 colocalization with TGN revealed an increase in LAMP1 association with the Golgi in RAB8a KO cells but not RAB10 KO cells (Figure 2G), while both KO models showed a significant decrease in lysosomal numbers (Figure 2H). Taken together, these data support an overall model whereby inactivation of RAB8a or RAB10 drives opposing effects on Golgi and lysosome function.

Figure 2.

RAB8a KO and RAB10 KO cells exhibit alterations in Golgi distribution

RAB8a KO, RAB10 KO, and isogenic WT control iPSCs were fixed and stained for cis-Golgi (GM130), trans-Golgi (TGN46), and LAMP1 (A and D). z stack confocal images were 3D reconstructed in Imaris (Bitplane) and the mean distance between CGN stacks (B), distance of CGN to nucleus (C), TGN volume (E), TGN distance to nucleus (F), colocalization between TGN46 and LAMP1 (G), and number of lysosomes per cell were analyzed (H). (B, C, E, and F represent individual Golgi stacks data points across N > 50 cells over three independent experiments; G and H represent per-cell data). ^∗∗p < 0.001; ^∗∗∗p < 0.0001; ^∗∗∗∗p < 0.0001; one-way ANOVA Tukey’s post hoc.

RAB8a KO human neurons accumulate insoluble α-synuclein

Accumulation of highly insoluble α-synuclein is a pathological hallmark of postmortem PD tissue and indicative of Lewy body pathology. We, and others, have shown accumulation of α-synuclein in different neuronal models of genetic PD, including LRRK2 mutant primary mouse neurons and GBA1 heterozygous-null human neurons (Mazzulli et al., 2016; Sanyal et al., 2020; Schapansky et al., 2018). To assess α-synuclein metabolism in RAB8a and RAB10 null cells, we sequentially extracted total cellular proteins from iNs in isotonic buffer containing 1% (w/v) NP40 and determined the levels of detergent-soluble and insoluble α-synuclein (dissolved in 2% [w/v] SDS). We observed an accumulation of insoluble α-synuclein in RAB8a KO neurons compared with isogenic controls while no difference was observed in RAB10 KO neurons (Figures 3A–3C), suggesting RAB8a-specific effects on the capacity of these neurons to degrade α-synuclein. Here, we used β-actin as a loading control and validated our data by normalizing on β3-tubulin as well, as a neuronal-specific loading control (Figure S3).

Figure 3.

RAB8a KO human neurons accumulate insoluble α-synuclein and secrete α-synuclein in media

RAB8a KO, RAB10 KO, and WT neurons were collected at D21 and extracted in NP-40 containing lysis buffer (soluble fraction), followed by SDS resuspension of the pellet (insoluble). Densitometry analysis represents α-synuclein levels normalized to actin (A–C: ^∗p < 0.05; ^∗∗p < 0.001; one-way ANOVA Tukey’s post hoc, n = 3 differentiations).

(D and E) Medium was collected at D21 and analyzed by dot-blot for total and oligomeric α-synuclein. Protein yields of neuronal lysates were used to normalize for variability in cell density and analyze equivalent amounts of conditioned media (^∗∗∗p < 0.0001; ^∗∗∗∗p < 0.0001; one-way ANOVA Tukey’s post hoc; n = 4 differentiations).

One of the prominent models of how pathology may be propagated in PD supports the release of oligomeric α-synuclein from cells and direct cell-to-cell transfer (Volpicelli-Daley et al., 2014). Previously, we have shown that GBA1 heterozygous-null neurons secrete higher levels of oligomeric α-synuclein in the media compared with isogenic controls (Sanyal et al., 2020), suggesting that lysosomal dyshomeostasis has a direct effect on α-synuclein secretion in human neurons. Here, we asked whether the increase in insoluble α-synuclein observed in RAB8a KO neurons is accompanied by changes in α-synuclein release, by immunoblotting of conditioned media. RAB8a KO and RAB10 KO neurons secreted higher levels of α-synuclein in the media compared with isogenic controls, while a trend toward an increase was observed in oligomeric α-synuclein in RAB8a KO cells (Figures 3D and 3E). These data suggest that RAB8a and RAB10 deficiency may induce insufficient degradation and augmented secretion of α-synuclein species in neuronal models.

RAB10 KO human neurons show altered tau levels and phosphorylation

Tau pathology in the form of intracellular inclusions of hyperphosphorylated and aggregated tau protein are a common occurrence in LRRK2 PD brain, and studies in cell models have highlighted a mechanistic interplay between tau and RAB GTPase pathways downstream of LRRK2 activity (Guerreiro et al., 2016; Sobu et al., 2021). To characterize the effects of RAB8a and RAB10 deficiency on the biochemical properties of tau, we extracted proteins from mature cortical neurons and compared the levels of Ser202/Thr205-phosphorylated tau between the KO lines and isogenic controls. Previous studies have reported detection of distinct phospho-Tau bands, between ∼54 kDa and 68 kDa (Falcon et al., 2015; Jackson et al., 2016), by immunoblot analysis of human tissue. In fact, a strong correlation with tau aggregation propensity has been shown for pTau species running close to 62 kDa but not the lower molecular weight pTau band (∼54 kDa) (Jackson et al., 2016). Here, we probed with a commercial pS202, pT205 tau (AT8) antibody and detected three bands approximately at 54 kDa, 62 kDa, and 70 kDa (Figure 4A). A significant increase in phosphorylated tau (∼62 kDa) was observed in RAB10 KO neurons compared with WT or RAB8a KO neurons (Figures 4A and 4B). We also assessed tau phosphorylation at residues S396/S404 (PHF-1) that have been linked to Alzheimer’s disease (AD) and other Tauopathies and observed no difference in phosphorylation across lines (Figure 4C). We observed an overall decrease in total tau levels in RAB10 KO neurons compared with WT controls and tested whether this reflects a shift toward insolubility. Here, we found a similar trend of lower tau levels in the detergent-insoluble fraction (Figures 4D and 4E). These data demonstrate a divergence between pathways regulating disease-relevant post-translational modifications of tau downstream of the two RAB GTPases.

Figure 4.

RAB10 KO increases phospho-Tau species in human neurons

Human cortical neurons were collected on D21 and phosphorylated (AT8: pS202/T205 Tau and PHF-1: pS396/S404) and total soluble and insoluble Tau levels were assessed in RAB8a KO and RAB10 KO cells compared with two isogenic control lines (A and B). The ratio of each pTau band (54 kDa, 62 kDa, and 70Da) out of the total pTau signal was plotted (62 kDa groups: one-way ANOVA Tukey post hoc, BR01RAB10KO vs. BR01 or BR33: ^∗∗∗∗p < 0.0001, BR33 RAB10 KO vs. BR01 or BR33: ^∗p < 0.05, F (5, 12) = 29.83). Densitometry analysis represents PHF-1/tTau (C), soluble tTau normalized to actin (D) and detergent-insoluble tTau over actin (E) (n = 3 differentiations, one-way ANOVA Tukey’s post hoc). Schematic of RAB8a KO and RAB10 KO cellular phenotypes and proposed model (F) (created with BioRender.com).

Discussion

LRRK2 is a multi-domain kinase genetically linked to multiple neurologic diseases, which most commonly present with the pathologic accumulation of either α-synuclein or tau protein. A subset of the RAB GTPase protein family are bona fide LRRK2 substrates with PD-linked mutations increasing their phosphorylation (Liu et al., 2018; Steger et al., 2016), but how this contributes to disease is unknown. Given the converging evidence of endolysosomal dysfunction in PD (Kluss et al., 2019; Mamais et al., 2023), we focused attention on the two most well-validated substrates of LRRK2, RAB8a and RAB10, and their roles in endolysosomal biology and the proteostasis of α-synuclein and tau.

We and others have shown that PD-linked mutations in LRRK2 impair lysosomal acidification and reduce basal autophagic flux (Schapansky et al., 2018; Mamais et al., 2021; Eguchi et al., 2018; Henry et al., 2015; Obergasteiger et al., 2020). However, the contribution of individual LRRK2 substrates to these effects remained unexplored. We used CRISPR-Cas9 gene editing to ablate RAB8a and RAB10 expression in iPSCs and differentiated these into NGN2-induced neurons. While we observed a decrease in lysosomal number in both RAB KO neurons, luminal lysosome pH was reduced in RAB8a KO iNs and increased in RAB10 KO iNs, compared with control.

RAB8a and RAB10 can both localize to endosomal compartments and the TGN and regulate the dynamic interface between late Golgi membranes and endolysosomes (Pfeffer, 2018), with new evidence of a presynaptic role for RAB10 in neurons (Singh et al., 2023). Glycosylation of LAMP1 is necessary for its transport from the endoplasmic reticulum via the Golgi to the lysosome and its role in lysosomal integrity (Schwake et al., 2013). Irregular LAMP1 glycosylation has been linked to PD and the tauopathy, Niemann-Pick disease (Cawley et al., 2020; Fernandes et al., 2016). Here, we observed irregularities in LAMP1 glycosylation in RAB8a and RAB10 KO neurons supporting impairment in LAMP1 post-Golgi trafficking. We observed a decrease in total Golgi volume in both RAB GTPase KO cell models while only RAB8a KO cells showed LAMP1 retention in the Golgi. Opposing effects were noted, with RAB8a KO Golgi being more compacted and RAB10 KO Golgi more dispersed than WT. We surmise that both deviations from WT morphology mutually impair glycosylation at the Golgi and post-Golgi trafficking. There is a precedent for RAB8a and RAB10 to hold opposing actions, as Dhekne and colleagues (2018) reported that RAB8a stimulates cilia formation while RAB10 inhibits this process. Thus, this antagonistic relationship between RAB8a and RAB10 with respect to lysosomal pH and Golgi morphology may further represent a canonical feature of their biology.

We found that the levels of detergent-insoluble and secreted α-synuclein were increased in RAB8a KO, but not in RAB10 KO iNs. These new findings are consistent with previous reports from our group and others linking LRRK2 mutations to α-synuclein solubility and release, and suggest that LRRK2-RAB8a interactions are sufficient for these effects (Schapansky et al., 2018; Volpicelli-Daley et al., 2016). The distinction between RAB8a and RAB10 here is interesting, as RAB8a but not RAB10 has been linked to another PD gene (PINK1) (Vieweg et al., 2020), which presents with classic α-synuclein-rich Lewy body pathology (Gandhi et al., 2006). Furthermore, RAB8a has been found to physically associate with α-synuclein (Yin et al., 2014). Thus, these new data and the larger available literature would endorse a special relationship between the LRRK2 substrate RAB8a and α-synuclein.

While Lewy bodies are a main pathological hallmark of idiopathic PD, there is an overrepresentation of tau pathology in LRRK2 PD (Henderson et al., 2019; Herbst et al., 2022; Ujiie et al., 2012; Zimprich et al., 2004). Furthermore, a recent genome-wide association study linked genetic variation in the LRRK2 locus to progressive supranuclear palsy (PSP), a formal tauopathy (Jabbari et al., 2021). How LRRK2 might drive independent and distinct neuropathological features remains unexplored. It is noteworthy that RAB10 phosphorylation has been reported in PSP and AD brains with colocalization between pRAB10 and phosphorylated tau in neurons bearing tau-positive neurofibrillary tangles (Yan et al., 2018) and that RAB10 has been genetically linked to altered risk for AD (Andrews et al., 2019; Le Guen et al., 2021; Ridge et al., 2017; Tavana et al., 2018). Here, we report complex changes in tau in RAB10 KO but not RAB8a KO iNs. We observed a decrease in total tau and relative increase in Ser202/Thr205 phosphorylation in RAB10 KO neurons compared with isogenic controls, with neither occurring in RAB8a KO iNs. Given the multiple links between RAB10 and tau, and our new data, allelic variation affecting RAB10 function may influence the manifestation of α-synuclein or tau pathology in patients. Although altered tau phosphorylation has been reported in LRRK2 models (Bailey et al., 2013; Melrose et al., 2010; Nguyen et al., 2018; Schapansky et al., 2018), the mechanisms remain unclear, and these data indicate a likely role for RAB10.

Our study presents critical further evidence of opposing functions for RAB8a and RAB10 (Figure 4F). While many of the same organelles are implicated, RAB8a and RAB10 ablation consistently manifested distinct phenotypes within the secretory pathway. In addition, we identify novel and independent roles for RAB8a and RAB10 in modulating the homeostasis of α-synuclein and tau, respectively, the two proteins implicated in the neurological disorders linked to the RAB kinase, LRRK2. This work aligns well with independent studies involving RAB8a and RAB10 and brings novel resources and a new conceptual framework to consider the downstream consequences of LRRK2-dependent phosphorylation of individual RAB proteins.

Experimental procedures

Resource availability

Corresponding author

Further information and requests for reagents should be directed to the corresponding author Dr. Matthew J. LaVoie (mlavoie@ufl.edu).

Materials availability

Materials and additional details can be made available by the corresponding author upon request.

Data and code availability

All data generated and analyzed during this study are included in the article and supporting files.

Generation of RAB8a KO and RAB10 KO human iPSCs

CRISPR-Cas9 genome editing was performed as before (Sanyal et al., 2020). For information on single guide RNA sequences and validation of clones see supplemental experimental procedures.

Differentiation of human iPSCs

iPSCs were differentiated into cortical neurons using established protocols (Zhang et al., 2013). See supplemental experimental procedures for further information on iPSC culture, transduction, and differentiation methods.

Biochemical assays and cell imaging

Several biochemical assays, including analysis of soluble/insoluble cell lysate fractions, dot-blot, immunocytochemistry, and lysosomal characterization were performed. For further information on these assays and their analysis refer to supplemental experimental procedures.

Statistical analyses

All experiments were conducted at least three independent times for three differentiations. Error bars indicate mean ± SD. Statistical analysis was performed using GraphPad Prism software, using a one-way ANOVA with Tukey’s post hoc test.

Published: February 1, 2024