Unbiased genome-wide genetic screens are a powerful approach to study the functions of membrane trafficking genes in cultured mammalian cells (Wang et al., 2023; Gulbranson et al., 2019; Gulbranson et al., 2017). Previously, we conducted a genome-wide CRISPR screen to identify new regulators of surface protein homeostasis using a surface protein reporter (GLUT-SPR) based on the glucose transporter GLUT4 (Wang et al., 2023). Mouse preadipocytes expressing the GLUT-SPR reporter were mutagenized using the genome-wide GeCKO v2 CRISPR library (Ran et al., 2013). In the genome-wide CRISPR screen, we used fluorescence-activated cell sorting (FACS) to isolate mutant preadipocytes with reduced surface levels of the GLUT-SPR reporter (Wang et al., 2023). The screen recovered known regulators of cargo exocytosis and enabled us to identify previously uncharacterized exocytic regulators, including Reps1 and Ralbp1 (Wang et al., 2023). Importantly, the new exocytic regulators identified in this GLUT-SPR-based CRISPR screen broadly regulate the surface proteostasis of membrane proteins (Wang et al., 2023).

To gain new insights into the molecular functions of Retromer and Commander, we re-analyzed the data from the CRISPR screen to examine genes encoding Retromer and Commander subunits. The CRISPR library used in the screen contained guide RNAs (gRNAs) targeting the genes encoding the subunits of the Retromer and Commander complexes (Figure 1A and B). The GLUT-SPR-based CRISPR screen isolated Vps35 and Vps29 (Figure 1C), which encode subunits of the Retromer complex, consistent with the critical role of Retromer in the endosomal recycling of GLUT4 (Gulbranson et al., 2017; Pan et al., 2017). Genes encoding Vps26 were not recovered in the screen due to redundancy (Vps26a and Vps26b). Recovery of the Retromer genes also suggests that the CRISPR screen has the sensitivity and specificity to systematically examine the functional roles of endosomal recycling genes. Genes encoding the unique subunits of the Retriever complex – Vps35l and Vps26c – were not isolated in the CRISPR screen (Figure 1C). Similarly, virtually none of the genes encoding the CCC complex were recovered as significant hits in the CRISPR screen, except for Commd3, which encodes the COMMD3 subunit of the Commander complex (Figure 1A–C; Healy et al., 2023). These data indicate that the Commander complex is dispensable for the endosomal recycling of GLUT-SPR.

To further characterize the functional roles of Retromer- and Commander-encoding genes in cell physiology, next we examined whether these genes are essential for cell viability or growth by re-analyzing data from an unbiased genome-wide essentiality screen. In the essentiality screen, mouse preadipocytes mutagenized by the CRISPR GeCKO v2 library were continuously passaged for two weeks without any selection (Wang et al., 2023). The abundance of gRNAs in this cell population was sequenced and compared with that in the original CRISPR gRNA library. If a gene is essential for cell viability or growth, its corresponding gRNAs would be depleted after the two-week cell passage. We found that none of the genes encoding the Retromer and Commander complexes are essential in mouse preadipocytes (Figure 1D). Thus, the disruption of either the Retromer or Commander-dependent endosomal retrieval pathways does not impair the viability or proliferation of these cells.

Together, these global genetic analyses suggest that COMMD3 plays an unrecognized role in GLUT-SPR trafficking, independent of its canonical function within the Commander holo-complex.

To validate the role of COMMD3 in the surface homeostasis of GLUT-SPR, we deleted the Commd3 gene in mouse preadipocytes (Figure 2A). Indeed, surface levels of GLUT-SPR were strongly reduced in Commd3 KO preadipocytes, which were either cultured under standard conditions or stimulated with insulin to mimic a physiological fed state (Figure 2B). These data confirm the findings from the CRISPR screen (Figure 1C). To examine whether COMMD3 regulates the surface homeostasis of GLUT-SPR in another cell type, the fibroblast-like preadipocytes were differentiated into mature adipocytes, which are morphologically and functionally distinct from preadipocytes (Ahfeldt et al., 2012). Using flow cytometry, we observed that surface GLUT-SPR levels were markedly reduced in Commd3 KO adipocytes (Figure 2C), indicating that the role of COMMD3 in surface protein homeostasis is not restricted to a single cell type. Similar findings were observed in adipocytes using confocal imaging (Figure 2D). Surface levels of GLUT-SPR were substantially decreased in Commd3 KO cells, concomitant with elevated intracellular accumulation of the reporter (Figure 2D). These results demonstrated a critical role of COMMD3 in the endosomal trafficking of GLUT-SPR.

To further characterize the function of COMMD3 in the endosomal trafficking of GLUT-SPR, we stained for EEA1, an early endosome marker (Mishra et al., 2010), and examined the morphology of EEA1-positive organelles. We observed that the morphology of EEA1-positive endosomes was significantly altered in Commd3 KO cells (Figure 3A). The size of the EEA1-positive endosomes markedly increased in Commd3 KO cells (Figure 3A and B), reminiscent of the endosome enlargement observed in Retromer-deficient cells (Neuman et al., 2021). We also observed elevated GLUT-SPR accumulation in EEA1-positive endosomes in Commd3 KO cells (Figure 3C). Next, we used Structured Illumination Microscopy (SIM) to visualize the localization of GLUT-SPR and Rab5, another marker of the early endosome (Langemeyer et al., 2018). We found that a portion of GLUT-SPR resided in Rab5-positive compartments and its co-localization with Rab5 increased in Commd3 KO cells (Figure 3D and E).

Together, these data demonstrate that deletion of Commd3 leads to the enlargement of early endosomes and accumulation of GLUT-SPR in these compartments, consistent with a trafficking defect occurring at the endosomal recycling step.

Next, we sought to confirm the Commander-independent function of COMMD3 using comparative targeted mutations. We deleted genes encoding other Commander subunits and compared the phenotypes of the KO cells with that of Commd3 KO cells. The COMMD proteins are found in three subcomplexes prior to forming the heterodecameric COMMD complex: subcomplex A (COMMD1-4-6-8), subcomplex B (COMMD2-3-4-8), and subcomplex C (COMMD5-7-9-10) (Healy et al., 2023). Here, we selected COMMD1, COMMD3, and COMMD5 as representative subunits of the three subcomplexes. We measured surface levels of integrin alpha-6 (ITGA6), a known Commander-dependent cargo (McNally et al., 2017), in cells lacking one of the COMMD proteins. As expected, surface levels of ITGA6 were downregulated in all these KO cell lines (Figure 4A), confirming the canonical role of COMMD3 within the Commander holo-complex.

Next, we examined the trafficking of other cargo proteins. We found that surface levels of GLUT-SPR were slightly increased in the Commd1 KO cells, in contrast to the strong reduction observed in Commd3 KO cells (Figure 4B). Surface levels of GLUT-SPR were only slightly affected in Commd5 KO cells (Figure 4B). Surface levels of the transferrin receptor (TfR), a Commander-independent cargo (Puthenveedu et al., 2010), were markedly decreased in Commd3 KO cells but remained intact in Commd1 or Commd5 KO populations (Figure 4C). Total TfR levels also decreased in Commd3 KO cells (Figure 4—figure supplement 1), consistent with the notion that unretrieved cargo is routed to the lysosome for degradation (Healy et al., 2023; Simonetti and Cullen, 2019; Cullen and Steinberg, 2018). Surface levels of lysosomal-associated membrane protein 1 (LAMP1) were also strongly decreased in Commd3 KO cells but were not substantially changed in Commd1 or Commd5 KO cells (Figure 4D). Together, comparative analysis of these diverse membrane proteins clearly demonstrates that COMMD3 regulates a group of cargo proteins independent of the Commander holo-complex, in addition to its canonical function within the Commander complex.

To further characterize the Commander-independent function of COMMD3, we examined the KO phenotypes of other Commander subunits. We observed that the expression of CCDC93 and VPS35L was substantially decreased in Commd3 KO cells, and VPS35L expression was diminished in Ccdc93 KO cells. These findings are consistent with the notion that stability of subunits within a protein complex is interdependent (Wang et al., 2023; Gulbranson et al., 2019). Interestingly, we observed that COMMD3 expression was not reduced but upregulated in Ccdc93 or Vps35l KO cells (Figure 5A, Figure 5—figure supplement 1). Thus, unlike other Commander subunits, COMMD3 persists when other subunits of the Commander complex are depleted, further supporting the ability of COMMD3 to regulate endosomal trafficking independent of the Commander holo-complex.

The upregulation of COMMD3 in Commander-deficient cells, which likely reflects a compensatory mechanism, offers another strategy to test the Commander-independent function of COMMD3. We reasoned that if a cargo is dependent on COMMD3 but not on the Commander complex, its surface levels could be increased in cells with elevated COMMD3. Indeed, we observed that surface levels of TfR, a cargo dependent on Commd3 but not on the Commander complex (Figure 4C; Puthenveedu et al., 2010), were markedly increased in Ccdc93 or Vps35l KO cells (Figure 5B). Conversely, surface TfR levels were diminished in Commd3 KO cells and were fully restored by expression of a Commd3 rescue gene (Figure 5B). Total TfR expression was also reduced in Commd3 KO cells and was rescued by the Commd3 rescue gene (Figure 7D, Figure 4—figure supplement 1A). KO of both Commd3 and Ccdc93 resulted in decreased surface levels of TfR (Figure 5C), confirming that the elevated surface TfR observed in Ccdc93 KO cells was caused by COMMD3 upregulation. Consistent with this notion, overexpression of Commd3 in WT cells markedly elevated TfR surface levels (Figure 5D). Altogether, these results demonstrate that COMMD3 is upregulated in Retriever- or CCDC93-deficient cells and enhances the endosomal retrieval of COMMD3-dependent cargoes, further supporting a Commander-independent role of COMMD3 in endosomal recycling.

Next, we sought to determine the molecular mechanism by which COMMD3 regulates endosomal trafficking in a Commander-independent manner. The COMMD3 protein is comprised of two autonomously folded domains – NTD and CTD (Figure 6A). The C-terminal COMMD domains of COMMD proteins are well-conserved and mediate the formation of the heterodecameric COMMD complex within the Commander holo-complex (Boesch et al., 2024; Maine and Burstein, 2007). The NTDs of COMMD proteins share a similar structure but are more divergent in sequence (Laulumaa et al., 2024; Healy et al., 2018). The function of the COMMD NTDs is unknown. Next, we examined whether the Commander-independent function of COMMD3 relies on its NTD. We expressed the COMMD3 NTD in Commd3 KO cells and examined whether the KO phenotype could be rescued. Indeed, expression of the COMMD3 NTD fully restored the surface levels of TfR in Commd3 KO cells while the CTD did not (Figure 6B, Figure 6—figure supplement 1), indicating that the NTD is sufficient for this function.

To gain molecular insights into the NTD of COMMD3, we determined the interactomes of full-length (FL) COMMD3, NTD, and C-terminal domain (CTD) (Figure 6C and D). FL COMMD3, NTD, and CTD were individually expressed and isolated through co-immunoprecipitation (co-IP), and the associated proteins within these samples were identified using mass spectrometry. Through this proteomic analysis, we identified a group of proteins present in the interactomes of FL COMMD3 and NTD but absent from the CTD interactome (Figure 6E, Supplementary file 1). Among these proteins, ARF1 emerged as a strong candidate because it acts as a membrane trafficking regulator known to function on endosomes (Figure 6F; Nakai et al., 2013; Stockhammer et al., 2024). The ARF family of proteins is small GTPases involved in the recruitment of coat proteins, activation of membrane lipid-modifying enzymes, and interaction with the cytoskeleton (Stockhammer et al., 2024; D’Souza-Schorey and Chavrier, 2006). Using co-IP, we observed that the NTD of COMMD3 interacted with ARF1 (Figure 6G), confirming the results of the proteomic experiments. Together, these data demonstrate that the NTD of COMMD3 mediates its Commander-independent function and interacts with ARF1.

To further characterize the interaction between COMMD3 and ARF1, we used SIM to visualize their subcellular localization. We observed significant co-localization between ARF1 and the NTD of COMMD3 (Figure 7A and B), consistent with their interactions detected in proteomic and co-IP experiments (Figure 6). We noted that, in the co-IP assays using 293T cells, ARF1 expression levels were significantly elevated when co-expressed with the NTD of COMMD3 (Figure 6G). A similar ARF1-stabilizing effect was also observed in HeLa cells (Figure 7C). These findings raised the possibility that COMMD3 uses its NTD to bind and stabilize ARF1. To test this model, we examined endogenous ARF1 levels in Commd3 KO cells. Interestingly, we found that ARF1 levels were markedly reduced in Commd3 KO cells and were fully restored upon introduction of a COMMD3 rescue gene (Figure 7D). To further investigate the functional link between COMMD3 and ARF1, we overexpressed ARF1 in Commd3 KO cells. Strikingly, ARF1 overexpression fully restored the surface levels of TfR in Commd3 KO cells (Figure 7E). These data demonstrate that COMMD3 regulates endosomal trafficking through binding and stabilizing ARF1.

Finally, we sought to determine whether ARF1 binding is required for the Commander-independent function of COMMD3. AlphaFold3 predicted a high-confidence structure of the ARF1-GTP:COMMD3-NTD heterodimeric complex (Figure 8A and B). According to this structural model, the α1 helix of COMMD3-NTD binds to the switch 1 of ARF1, while the α3 and α4 helices of COMMD3-NTD interact with the switch 2 of ARF1 (Figure 8A and B, Source data 1). The switches 1 and 2 of ARF1 are highly conserved regions that are regulated by GTP binding and interact with effectors involved in endosomal recycling (Goldberg, 1998; Sauvageau et al., 2017). The binding of COMMD3 to ARF1 is mainly mediated by hydrophobic interactions and hydrogen bonds, including hydrogen bonds formed by K60 and H61 of COMMD3 with Y81 of ARF1, K60 of COMMD3 with R79 and Q83 of ARF1, H63 of COMMD3 with H80 of ARF1 (Figure 8A and B, Source data 1). AphaFold3 did not predict high-confidence structural models between COMMD3-NTD and ARF1-GDP or apo-ARF1, suggesting that COMMD3 selectively recognizes the GTP-bound form of ARF1. This conclusion is consistent with the observation that COMMD3 strongly stabilizes GTP-bound ARF1 (Figures 6G and 7C).

Altogether, these results further support the conclusion that COMMD3 regulates endosomal trafficking through binding and stabilizing ARF1 (Figure 9).

Next, we introduced point mutations into the NTD of COMMD3 based on the structural model, aiming to disrupt its interaction with ARF1 (Figure 8C). The COMMD3-NTD mutant was expressed at a similar level as the WT protein in Commd3 KO cells (Figure 8D). We observed that the COMMD3-NTD mutant failed to restore surface levels of TfR, whereas WT COMMD3-NTD fully rescued the KO phenotype (Figure 8E). In agreement with this finding, WT COMMD3-NTD, but not the mutant, restored ARF1 expression in Commd3 KO cells (Figure 8F). Altogether, these results further support the conclusion that COMMD3 regulates endosomal trafficking through binding and stabilizing ARF1.

HeLa and 293T cells were obtained from ATCC and had been authenticated by the vendor. Mouse preadipocytes were immortalized cells derived from mouse adipose tissue and were validated using expression markers such as GLUT4, as well as functional assays including differentiation into adipocytes and insulin responsiveness. All cell lines were routinely tested for mycoplasma contamination by the StemTech core facility at the University of Colorado Boulder. The cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with 10% FB Essence (FBE, VWR, #10803–034) and penicillin/streptomycin (Thermo Scientific, #15140122). The cells were maintained in a humidified incubator at 37 ℃ with 5% CO₂. To differentiate preadipocytes into mature adipocytes, preadipocytes were grown to ~95% confluence before a differentiation cocktail was added at the following final concentrations: 5 µg/mL insulin (Sigma, #I0516), 1 nM Triiodo-L-thyronine (T3, Sigma, #T2877), 125 µM indomethacin (Sigma, #I-7378), 5 µM dexamethasone (Sigma, #D1756), and 0.5 mM 3-isobutyl-1-methylxanthine (IBMX, Sigma, #I5879). After two days, the cells were switched to DMEM supplemented with 10% FBE, 5 µg/mL insulin, and 1 nM T3. After another two days, fresh DMEM media supplemented with 10% FBE and 1 nM T3 were added to the cells. Differentiated adipocytes were usually analyzed six days after addition of the differentiation cocktail.

Genome-wide CRISPR screens of GLUT-SPR surface homeostasis and gene essentiality were described previously (Wang et al., 2023). To individually ablate a candidate gene, gRNAs targeting the gene were chosen via the CRISPick algorithm (https://portals.broadinstitute.org/gppx/crispick/public) to maximize KO efficiency and minimize off-target effects. The upstream guide was cloned into the pLenti-CRISPR-v2 vector (Addgene, #52961) and the downstream guide was cloned into a modified version of the pLentiGuide-Puro vector (Addgene, #52963), in which the puromycin selection marker was replaced with a hygromycin selection marker.

Guide sequences targeting the mouse Commd3 gene are: CTTCGCGCTTCTCCTCCGGG and CTTGAAACAGATCGACCCAG. Guide sequences targeting the mouse Commd1 gene are: TCACGGACACTCGGGTGTCA and ACTGCTCAAACCAAAAAGCA. Guide sequences targeting the mouse Commd5 gene are: GTTGTTGAAACTCGTAGTCG and TGCCAGCGCCAACCTGTCAG. Guide sequences targeting the mouse Ccdc93 gene are: CGAAAGTACCGACGGCAGCG and GATGACCGCCATGGCAAACG. Guide sequences targeting the mouse Vps35l gene are: GGATTATGTGAACCGCATAG and GGAGGTTTGCAAGTGCATCA. Double KO cells of Commd3 and Ccdc93 were generated using one guide per gene: Commd3 – CTTCGCGCTTCTCCTCCGGG and Ccdc93 – GATGACCGCCATGGCAAACG.

CRISPR plasmids were transfected into HEK 293T cells along with pAdVAntage (Promega, #E1711), pCMV-VSVG (Addgene, #8454), and psPax2 (Addgene, #12260). The 293T cell culture media containing lentiviral particles were harvested daily for four days and centrifuged at 25,000 rpm (113,000 g) for 1.5 hr at 4 °C using a Beckman SW28 rotor. Viral pellets were resuspended in PBS and used to infect target cells. After lentiviral infection, cells were selected using 3.5 µg/mL puromycin (Sigma, #3101118) for 2 days, followed by selection using 500 µg/mL hygromycin B (Thermo, #10687010) for another 2 days. All KO cells used in this work were pooled KO cell populations.

The codon-optimized human COMMD3 gene with a 3xFLAG-encoding sequence or an mCherry-3xFlag-encoding sequence was subcloned into the SHC003BSD-GFPD vector (Addgene, #133301). This COMMD3 gene was not targeted by gRNAs used in the KO experiments. The plasmid expressing the NTD (amino acids 1–124) of human COMMD3 was generated in a similar way. For the IP experiments, DNA fragments encoding FL and truncated human COMMD3 were subcloned into the SHC003BSD-GFPD vector. FL COMMD3 (amino acids 1–195) and NTD were fused to mCherry and 3xFlag tags at their C-termini. The CTD (amino acids 125–195) was fused to an mCherry tag at its N-terminus and a 3xFlag tag at its C-terminus. The human ARF1 gene was subcloned into the SHC003BSD-GFPD vector with an HA-encoding sequence at the 3’ end. The constructs were transfected into 293T cells to produce lentiviral particles using a similar procedure as CRISPR lentiviral production. The lentiviruses were used to infect target cells, followed by selection using 10 µg/mL blasticidin (Thermo Fisher Scientific, #BP2647).

Cells grown on cell culture plates were washed with ice-cold KRH buffer and blocked at 4 °C with KRH buffer supplemented with 5% FBE. Subsequently, the cells were labeled for 1 hr with KRH buffer containing 2% FBE and the following antibodies: anti-HA (BioLegend, #901501, RRID:AB_2565006), APC-conjugated anti-LAMP1 (BioLegend, #328619, RRID:AB_1279055), anti-ITGA6 (Invitrogen, #14-0495-82, RRID:AB_891480), APC-conjugated anti-mouse secondary antibodies (eBioscience, #17-4015-82, RRID:AB_2573205), APC-conjugated anti-rat antibodies (Invitrogen, #A10540, RRID:AB_10562535), and APC-conjugated anti-TfR/CD71 antibodies (BioLegend, #334108, RRID:AB_10915138). Following antibody labeling, cells were washed twice with KRH buffer containing 5% FBE and once with PBS. Cells were dissociated using Accutase before resuspension in PBS buffer containing 5% FBE. Cells were analyzed in triplicate on a CyAN ADP analyzer (Beckman Coulter).

Cells grown on glass coverslips were washed with PBS and fixed using 4% PFA in PBS. Cells were then permeabilized using 0.1% Tween-20 and blocked with PBS buffer containing 5% FBE. Permeabilization was omitted when surface proteins were stained. Cells were labeled for 1 hr with 2% FBE in PBS using the following antibodies: anti-HA, anti-FLAG M2 (Sigma-Aldrich, #F1804, RRID:AB_262044), Alexa Fluor 647-conjugated anti-mouse IgG (Invitrogen, #A32733, RRID:AB_2633282), and Alexa Fluor 568-conjugated anti-rabbit IgG (Thermo Fisher Scientific, #A11004, RRID:AB_2534072). Confocal images were acquired on a Nikon A1 laser scanning confocal microscope using a ×100×oil immersion objective. In SIM, images were captured using a 100 x oil immersion objective on a Nikon SIM microscope as previously described (Wan et al., 2024).

In IP experiments, cells were lysed in IP buffer (25 mM HEPES [pH 7.4], 138 mM NaCl, 10 mM Na3PO4, 2.7 mM KCl, 0.5% CHAPS, 1 mM DTT, and a protease inhibitor cocktail). After centrifugation, proteins were immunoprecipitated from cell extracts using primary antibodies and protein A/G agarose beads (Thermo Scientific, #WF324079). For immunoblotting, immunoprecipitates or whole cell lysates were resolved on 8% Bis-Tris SDS–polyacrylamide gel electrophoresis (SDS-PAGE) and transferred to PVDF membranes. Proteins were detected using unlabeled primary antibodies and horseradish peroxidase (HRP)-conjugated secondary antibodies, or HRP-conjugated primary antibodies. Primary antibodies used in immunoblotting include anti-COMMD3 (Bethyl, #A304-092A, RRID:AB_2621341), anti-VPS35L (Invitrogen, #PA5-28553, RRID:AB_2546029), anti-CCDC93 (Santa Cruz Biotechnology, #sc-514600), anti-alpha-tubulin (DSHB, #12G10, RRID:AB_1210456), HRP-conjugated anti-FLAG M2 (Sigma-Aldrich, #A8592, RRID:AB_439702), and HRP-conjugated anti-HA (Roche, #12013819001, RRID:AB_390917). Secondary antibodies used in this work include HRP-conjugated anti-rabbit IgG (Sigma-Aldrich, #A6154, RRID:AB_258284) and HRP-conjugated anti-mouse IgG (Sigma-Aldrich, #A6782, RRID:AB_258315). All experiments were run in biological triplicates. Intensities of protein bands on immunoblots were quantified using ImageJ.

Mass spectrometry was carried out as previously described (Wang et al., 2023). Immunoprecipitates on protein A/G beads were snap-frozen and stored at –70 °C. Peptides were pre-fractionated using high-pH fractionation and analyzed on the Thermo Ultimate 3000 RSLCnano System via direct injection. Data were processed using MaxQuant/Andromeda (version 1.6.2.10) and compared to Uniprot-annotated protein sequences. False discovery rates were set to 0.01 for protein and peptide assignment with a minimum peptide length of four residues and a minimum peptide number of one.

The structural model of the ARF1-GTP:COMMD3-NTD (a.a. 1–120) heterodimer was predicted using AlphaFold3 with default settings (Abramson et al., 2024). Five independent structural models were generated for each protein complex, and the quality of the predicted models was assessed through their interface predicted template modeling (ipTM) scores, predicted template modeling (pTM) scores, predicted alignment error (PAE) plots, and predicted local distance difference test (pLDDT) scores (Abramson et al., 2024; Mirdita et al., 2022). Structural analysis was conducted using UCSF ChimeraX (v1.8) (Goddard et al., 2018) and PDBePISA (Krissinel and Henrick, 2007).

All data shown in the figures are from at least three independent biological replicates. Biological replicates were plated, treated, and analyzed in parallel. Statistical analyses were performed using GraphPad Prism 10. Student’s t-tests were used for comparisons between two groups. One-way ANOVA with Dunnett’s multiple comparisons test (comparing all groups to control) or Sidak’s multiple comparisons test (for internal group comparisons) was used when analyzing experiments with more than two groups. Significance is indicated as follows: n.s., not significant; *p<0.05; **p<0.01; ***p<0.001. Additional details on sample size, error bars, and statistical tests are provided in the figure legends.