The lactate sensor NDRG3 decelerates ER-to-Golgi transport through interaction with the long isoform of syntaxin-5

Significance

Mammalian cells can adapt to low oxygen levels by stabilizing oxygen-dependent transcriptional activators of the hypoxia-inducible factor (HIF) family, which play crucial roles in angiogenesis, cell survival, and proliferation. Clear evidence exists for HIF-independent mechanisms in the cellular response to hypoxia. For example, the accumulation of lactate under hypoxic conditions stabilizes the NDRG3 protein, thereby activating intracellular signaling pathways to promote angiogenesis and cell growth. Here, we show that hypoxia and elevated intracellular lactate levels regulate ER-to-Golgi membrane trafficking through NDRG3 and syntaxin-5 (Stx5), an ER-to-Golgi SNARE protein implicated in a multisystemic disorder. Our data expand the understanding of cellular responses to oxygen deprivation.

Abstract

BET1, GOSR2, and STX5 variants can cause fatal inherited diseases, including epilepsies, muscular dystrophy, and multisystemic disorders. Together with Sec22b, they form a SNARE complex that mediates fusion of ER-derived vesicles with the ER-Golgi-intermediate compartment and the cis-Golgi. The Sec1/Munc18 protein SCFD1/Sly1 accelerates ER-to-Golgi SNARE complex assembly and membrane fusion, but much less is known about downregulation of ER-to-Golgi trafficking under cellular stress conditions. Here, we identify the lactate and hypoxia sensor protein NDRG3 as a binding partner of the ER-to-Golgi SNARE complex. NDRG3 binds via its C-terminal domain to the N-terminal domain of the long isoform of Stx5, thereby impairing ER-to-Golgi trafficking under hypoxic conditions and elevated intracellular lactate levels. In NDRG3-deficient cells, hypoxia- and lactate-induced inhibition of ER-to-Golgi trafficking is abolished. Our work identifies NDRG3 as a negative regulator of ER-to-Golgi SNARE complex function, mechanistically linking hypoxia and lactate to membrane trafficking in the secretory pathway.

SNARE [Soluble-NSF-(N-ethylmaleimide-sensitive-factor)-Attachment-Protein-Receptor] proteins are essential for membrane fusion and are best-characterized in synaptic vesicle exocytosis (1). Several proteins regulating SNARE function such as NSF, soluble NSF-attachment proteins (SNAPs), and Sec1/Munc18 (SM) homologs have been identified (2). Surprisingly, much less is known about nonsynaptic SNAREs and their regulation.

Anterograde transport of vesicles from the endoplasmic reticulum (ER) to the ER-Golgi-intermediate-compartment (ERGIC) and the cis-Golgi is the first step in the secretory pathway. Here, the SNAREs Stx5(Q_a), GOSR2 (Q_b), and BET1 (Q_c) form a complex with the R-SNARE Sec22b, which mediates membrane fusion of ER or ERGIC-derived vesicles in the cis-Golgi (3). Knockdown of GOSR2, Stx5, and Secc22b or the SM protein SCFD1/Sly1 leads to decreased protein secretion due to impairment of ER-to-Golgi transport (4). SCFD1 interacts with Stx5 and facilitates ER-Golgi SNARE complex formation (5). However, a detailed mechanism of complex formation has not been established, and the identity of additional interaction partners remain elusive. The identification of pathological GOSR2, BET1, and STX5 variants emphasizes their physiological role (6–8). Mutations in GOSR2 and BET1 cause neurological and muscular disorders, whereas variants in STX5 cause a fatal multisystemic disease (6–11).

NDRG3 is a member of the N-myc downstream regulated gene family, characterized by a catalytic inactive α/β hydrolase fold. Members share ≈60% sequence identity, similar structures (12–14) and intracellular localization (15). NDRG3 is highly expressed in the brain, testis, and prostate (16). NDRG3 has been implicated in signaling pathways involved in spermatogenesis, prostate cancer, and hypoxia (15–17). Physiologically hypoxic microenvironments are present in a variety of healthy tissues during normal conditions (18–20). Moreover, hypoxia plays a crucial role in tumor promotion as well as immune escape and malignancies (21, 22). A variety of cellular responses to hypoxia have been shown to be mediated by the hypoxia-inducible factors (HIF) family proteins, which are transcription factors stabilized at low oxygen levels (23–25). However, there is clear evidence of HIF-independent mechanisms in the cellular response to hypoxia: Tumors from HIF1α-deficient stem cells show growth advantages through reduced hypoxia-induced apoptosis and increased stress-induced proliferation (25). A HIF-independent pathway under the regulation of NDRG3 has been proposed (26, 27). In normoxia, NDRG3 is hydroxylated by PHD2, leading to ubiquitination and degradation; under prolonged hypoxia, it is stabilized by binding to lactate (26). The physiological effect of NDRG3-upregulation at hypoxic conditions remains unclear, but might involve the Raf-ERK-signaling pathway (26).

It has been shown that hypoxia hampers ER-Golgi membrane trafficking (28, 29), but the molecular mechanisms are unknown. Here, we identify NDRG3 as a lactate-dependent binding partner of the long isoform of Stx5 (Stx5L) and demonstrate that NDRG3 decelerates ER-to-Golgi transport and protein secretion. We found that the hypoxia-induced retardation of protein secretion was lost in NDRG3-deficient cells, indicating that NDRG3 is important for impeding the secretory pathway under hypoxic conditions.

Results

NDRG3 Interacts With the ER-to-Golgi SNAREs Stx5 and GOSR2.

Recognizing the pathophysiological importance of ER–Golgi SNAREs and the current gap in defined regulators, we conducted AP-MS profiling of HEK293T cells ectopically expressing wild-type, FLAG-tagged GOSR2 to identify binding partners. We confirmed interactions with SNARE complex members: BET1, Stx5, and Sec22b, validating this approach and notably identified NDRG3 as an interactor (Fig. 1A and SI Appendix, Excel File 1), with the highest mass spectrometry interaction statistics (MiST) score (30).

Fig. 1.

NDRG3 interacts with ER-to-Golgi SNAREs. (A) Summary of AP-MS data for GOSR2. MS interaction statistics (MiST) scores (combining quantity, reproducibility, and uniqueness of prey relative to other bait) are shown. The interaction of GOSR2 and NDRG3 was verified by co-IP of GFP-tagged GOSR2 and HA-tagged NDRG3 using HEK293 cells for expression and GFP-trap (B) or anti-HA magnetic beads (C) for precipitations. (D) Schematic representation of NDRG3. The cap-like motif is shown in orange and the α/β hydrolase fold in green. Full-length NRDG3, including the N- and C-termini are depicted in gray. Structural information of the N- or C-termini of any of the four NDRG members is missing. (E) Ribbon representation of a structural model of NDRG3. The cap-like domain is shown in orange with the α-helix six in magenta and the α/β hydrolase fold in green. (F) Immunoblot showing coimmunoprecipitations using GFP-trap in HEK293T cells coexpressing GOSR2-eGFP fusion proteins, NDRG2 as a negative control, and two NDRG2/NDRG3 chimeras. The numbers indicate the first amino acids of NDRG3 within the NDRG2/NDRG3 chimeras as depicted in (D). (G_I+II) The interaction of NDRG3 and GOSR2 was verified using the BiFC assay. Colocalization of the BiFC signal with GM130 demonstrated interaction of NDRG3 and GOSR2 at the cis-Golgi. (Scale bar, 20 µm, blue: DAPI staining, *transfected cell.) (G_III) Colocalization of GM130 with GOSR2, NDRG3, and the BiFC signals were quantified using Pearson correlation coefficient (PCC; mean PCC ± SD, One-Way Anova, Tukey post hoc comparison, n = 3 (at least 20 cells per replicate). ****P ≤ 0.0001).

To confirm interaction of GOSR2 and NDRG3, we expressed GFP- and HA-tagged proteins, respectively, in HEK293T cells and performed coimmunoprecipitations (co-IPs). As shown in Fig. 1B, we were able to validate the interaction by coprecipitation of NDRG3 with GOSR2 and in reciprocal co-IPs. Specificity was ensured by using NDRG2 as a negative control, which was not precipitated by GOSR2 (Fig. 1F). Next, we generated a three-dimensional homology model of NDRG3 using the X-ray structure of human NDRG2 (Fig. 1 D and E and SI Appendix, Fig. S1 A and B) (31). We refrained from using the NDRG3 crystal structure because of disordered regions in the cap-like domain (SI Appendix, Fig. S1B) (12). However, the NDRG3 model as well as NDRG2 and NDRG3 structures are very similar and consist of a canonical α/β-hydrolase fold domain and a smaller cap-like domain (Fig. 1 D and E and SI Appendix, Fig. S1B). Structural information about the N- and C-terminal domains is missing, likely due to an increased flexibility of these protein regions (12), which is also supported by AlphaFold and low per-residue model confidence scores (pLDDT; https://alphafold.ebi.ac.uk/entry/Q9UGV2).

Subsequently, we generated NDRG2/NDRG3 chimeras to identify structural determinants of GOSR2/NDRG3 interaction. The NDRG2/3(D180) chimera contains the N-terminal segment of NDRG2 fused to the C-terminal segment of NDRG3—including the cap-like motif—whereas the NDRG2/3(T333) chimera is predominantly NDRG2 and includes only the terminal tail from NDRG3 (Fig. 1D). Because both chimeras coimmunoprecipitated with GOSR2 (Fig. 1F), we conclude that the NDRG3 C terminus is sufficient to confer NDRG3-like GOSR2 binding to NDRG2.

Next, we analyzed the subcellular localization of NDRG3 and GOSR2. GOSR2 predominantly localizes to the cis-Golgi (PCC GOSR2/GM130: 0.84 ± 0.02) (SI Appendix, Fig. S2A_I), whereas NDRG3 was found in the cytoplasm with a diffuse staining within the cis-Golgi compartment (SI Appendix, Fig. S2A_II) with a Pearson correlation coefficient of (PCC NDRG3/GM130: 0.55 ± 0.02; Fig. 1G_III). To validate NDRG3/GOSR2 colocalization, we performed bimolecular fluorescence complementation (BiFC) by expressing NDRG3 and GOSR2 tagged with two complementary Venus nonfluorescent fragments (SI Appendix, Fig. S2B). Interaction of NDRG3 and GOSR2 resulted in reconstitution of fluorescence Venus protein that colocalized with the cis-Golgi marker GM130 (PCC BiFC/GM130: 0.83 ± 0.01; Fig. 1G). No BiFC signal was observed when GOSR2-VC and NDRG3-VN were expressed alone (SI Appendix, Fig. S2A_I+II).

To address the question of whether NDRG3 binds preferentially one specific member or a subset of ER-Golgi SNARE complex members, we performed AP-MS by using HA-tagged NDRG3. We included NDRG1, -2, and -4 to control specificity and to determine whether SNARE binding is a common feature of the NDRG-protein family. We found GOSR2 and Stx5 among the most abundant NDRG3 interaction partners in addition to other SNAREs (e.g., GOSR1, Stx8, and YKT6) and their binding partners SCFD1, NSF, and NAPA (Fig. 2A). None of these proteins coprecipitated with NDRG1, -2, and 4 (Fig. 2A and SI Appendix, Fig. S2C and Excel File 2). Indicating that although NDRG proteins are highly homologous this SNARE interaction is specific for NDRG3. Protein network analysis of AP-MS data revealed that the majority of NDRG3 interactors have a known role in membrane trafficking, strongly suggesting a role of NDRG3 in vesicular transport (SI Appendix, Fig. S2D). In contrast, protein network analysis of data for NDRG1, -2, and -4 interactors identified other cellular pathways, respectively (SI Appendix, Fig. S2E), further highlighting the specific role of NDRG3 in membrane trafficking.

Fig. 2.

NDRG3 interacts with Stx5. (A) Volcano Plot of AP-MS data using HA-tagged NDRG2 and NDRG3 as baits. (B_I) Co-IPs using HA-tagged NDRG2 and NDRG3 constructs expressed in HEK293T cells. HA-tagged proteins were immunoprecipitated with α-HA magnetic beads. IP samples and cell lysates were subjected to immunoblotting with α-GOSR2, α-NDRG3, α-BET1, α-SEC22b, α-Stx5, α-HA, and α-GAPDH antibodies, respectively. GAPDH served as negative and loading control. (B_II) Quantification of co-IP efficiencies. Protein levels of the (co-)IP samples were normalized to the respective input levels (One-Way Anova, Tukey post hoc comparison n = 3; *P ≤ 0.05; **P ≤ 0.01). (C_I) Co-IPs using FLAG-tagged NDRG3 and HA-tagged GOSR2 and Stx5L constructs. HA-tagged constructs were precipitated with α-HA magnetic beads. IP samples and cell lysates were subjected to immunoblotting with NDRG3 and HA antibodies, respectively. GAPDH served as negative and loading control. (C_II) Quantification of co-IP efficiencies (n = 3; ****P ≤ 0.0001). (D_I) Representative confocal images showing PLA signal (red) from NDRG3 and Stx5 in wild-type cells (Upper panel), in NDRG3-deficient cells as negative control (Lower Left panel), and GOSR2 and Stx5 in wild-type cells as positive control with DAPI staining (blue). (Scale bar, 20 μm.) (D_II) Immunoblot of NDRG3-deficient cell lines. (D_III) Relative quantification of PLA puncta per cell and PLA area per cell (±SD), One-Way Anova, Dunnett post hoc comparison N = 3, from n = 20 images (GOSR2/Stx5 in WT), n = 22 images [NDRG3/Stx5 in −/− (I)] n = 13 images [NDRG3/Stx5 in −/− (II)] n = 19 images (NDRG3/Stx5 in WT), confirming the endogenous interaction between NDRG3 and Stx5 under physiological conditions.

To corroborate our AP-MS results, we confirmed ER-Golgi SNARE complex interaction with NDRG3 by co-IPs (Fig. 2B_I). NDRG3 but not NDRG2 was able to efficiently precipitate GOSR2 and the long isoform of Stx5 (Stx5L), whereas the short isoform Stx5S, Sec22b, and BET1 were significantly less efficient, when compared to input levels, respectively (Fig. 2B_II). To further characterize the interactions of NDRG3 with Stx5L and GOSR2, we transiently expressed NDRG3 with HA-tagged Stx5L and GOSR2 and performed co-IPs. As shown in Fig. 2C_I+II, Stx5L efficiently coprecipitated NDRG3, in contrast to GOSR2, which pulled-down NDRG3 less efficiently. These results indicate that Stx5L is a main ER-to-Golgi SNARE interactor of NDRG3. To demonstrate NDRG3 and Stx5 interact under physiological conditions, we utilized Proximity Ligation Assay, which confirm interactions of endogenous proteins in close proximity (≤40 nm) of each other (31, 32). We showed that NDRG3/Stx5 PLA was specific to NDRG3 and Stx5 and significantly reduced to background levels in two independent NDRG3-deficient cell lines (Fig. 2D_I−III). NDRG3 deficiency was confirmed by immunoblot analysis as well as unaltered Stx5 and GOSR2 expression levels (Fig. 2D_I). As a positive control, the GOSR2–Stx5 PLA signal was ~2 to 3-fold higher and produced ~2 to 3-fold more puncta per cell than the NDRG3–Stx5 PLA (Fig. 2D_I+III).

In summary, these findings suggest that NDRG3 is an interaction partner of Stx5.

The N- and C-Terminal Domains of Stx5L and NDRG3, Respectively, Are Necessary for Interaction.

Stx5S and GOSR2 share the same general domain architecture, consisting of an N-terminal H_abc domain, a SNARE motif and a C-terminal transmembrane domain. However, Stx5L has an additional N-terminal domain of 54 amino acids and an adjacent short helical region necessary for SCFDI binding (Fig. 3A). Given the preferential interaction of Stx5L with NDRG3, we next asked whether NDRG3 directly interacts with Stx5L and whether this interaction is mediated by the extra 54 N-terminal amino acids. Therefore, we expressed 6xHis-tagged NDRG3 and the N-terminal aa1-54 domain of Stx5L as recombinant thioredoxin (TRX) and glutathione-S-transferase (GST) fusion proteins in Escherichia coli, respectively, to determine whether the N-terminal domain is sufficient to convey the interaction properties of Stx5L with NDRG3. The GST-Stx5L_aa1-54 fusion protein efficiently pulled down NDRG3 when compared to the negative control GST (Fig. 3B_I+II), indicating that the extra 54 N-terminal residues of Stx5L are sufficient for the interaction of Stx5L with NDRG3. Overall, these data show that NDRG3 directly interacts with Stx5L by binding to the N-terminal domain.

Fig. 3.

The N-terminus of STX5L is sufficient, and the C-terminus of NDRG3 is necessary for interaction. (A) Ribbon representation of an AlphaFold-generated structural model of Stx5L. The three helices of the H_abc (blue) domains, the SNARE (red), the transmembrane domains (magenta), the SCFD1 binding site (yellow), and the N-terminal domain (green) are shown. Known interaction partners and binding sites for the ER-to-Golgi SNAREs, SCFD1, and NDRG3 are depicted. The two substitutions for generating a SCFD1-binding-deficient Stx5L construct are shown as well. (B_I) Direct interaction of the N-terminal region of Stx5L with NDRG3 was verified through pulldown experiments by using recombinant proteins encompassing the first 54 amino acids of Stx5L expressed as GST and full-length NDRG3 His- and TRXA-tag fusion proteins, respectively. GST served as a negative control. (B_II) Quantification of efficiency from pulldown experiments shown in B_I. (C) Co-IPs using HA-tagged wild-type Stx5L and Stx5L SCFD1-binding-deficient (Stx5L-T61A+F64A) constructs expressed in HEK293 cells. HA-tagged proteins were immunoprecipitated with α-HA magnetic beads. IP samples and cell lysates were subjected to immunoblotting with α-NDRG3, α-SCFD1, and α-HA. α-GAPDH served as negative and loading control. (D) Direct interaction of the N-terminal region including the H_abc domain of Stx5 only with full-length NDRG3 and not with NDRG3_aa1-315 was verified through pulldown experiments by using recombinant proteins expressed as GST and His-tag fusion proteins, respectively. GST served as negative control. * bands from incomplete stripped His antibody. (E) Experiments in (D) were repeated using recombinant proteins encompassing the first 54 amino acids of Stx5L expressed as GST fusion protein. The first 54 aa of STX5 are sufficient for direct interaction with full-length NDRG3 but not NDRG3_aa1-315. (F) Pulldown experiments as described in (B) with or without the presence of L-lactate. (G) Endogenous co‐IPs of NDRG3 (Upper four panels) and Stx5 (Lower four panels). Immunoblots of wild-type and NDRG3-deficient HEK293 cell lysates incubated with NDRG3 or Stx5 antibodies followed by precipitation with protein A/G beads. NDRG3-deficient cellular lysates served as negative controls. + Lactate cells were preincubated with 50 mM lactate in the tissue culture medium for 24 h. NDRG3 or Stx5 signals were only present in lactate-treated samples. SCFD1 antibodies were used as a positive control for coprecipitation. (H_I) (Co)-localization of NDRG3 (magenta) and Stx5 (green) in HEK293 cells pretreated with or without 50 mM lactate. Blue: DAPI staining. (Scale bar, 10 µm.) (H_II) PCC values for NDRG3 versus Stx5 [− lactate: n = 3 (282 cells in total), + lactate: n = 3 (248 cells in total)]. (**P ≤ 0.01, two tailed unpaired t test).

According to the Stx5L AlphaFold model (Fig. 3A) (33), the domains for NDRG3 and SCFD1 binding are in close proximity, suggesting that the two proteins may compete for binding to Stx5L. To test this hypothesis, we generated a Stx5L construct, carrying two mutations (Stx5L-T61A + F64A) in the SCFDI binding site (34) and tested it in co-IP experiments. In cells expressing HA-tagged Stx5L, co-IP of endogenous SCFDI and NDRG3 with Stx5L was observed, but the amount of NDRG3 was reduced. In contrast, HA-tagged Stx5L T61A + F64A mutant was unable to precipitate SCFD1 but coprecipitated NDRG3 more efficiently when compared to Stx5L wt (Fig. 3C). Together, these data suggest that NDRG3 binds to the N-terminal domain of Stx5L and may compete with SCFD1 for Stx5L binding. Given our results from the NDRG2/3 chimeras and GOSR2 binding (Fig. 1C_III) and the low sequence homology of the C termini of NDRG2 and NDRG3 (SI Appendix, Fig. S1A), we next asked whether the C terminus of NDRG3 is important for binding to Stx5L. Thus, we tested whether recombinant NDRG3 lacking the C terminus (NDRG3_aa1-315) interacts with GST-Stx5-H_abc and GST-Stx5L_aa1-54 in pull-down experiments compared to NDRG3. Whereas full-length NDRG3 was coprecipitated with GST-Stx5-H_abc (Fig. 3D) and GST-Stx5L_aa1-54 (Fig. 3E) the C-terminal truncated NDRG3_aa1-315 protein was unable to interact with Stx5-H_abc and Stx5L_aa1-54, respectively (Fig. 3 D and E).

These data strongly suggest that the C terminus of NDRG3 mediates interaction with Stx5L.

Lactate triggers NDRG3 and Stx5 binding.

Since it has been shown that NDRG3 binds lactate, we hypothesized that lactate might increase the NDRG3 and Stx5 interaction and tested whether lactate modifies NDRG3 and Stx5L binding in GST-pulldown experiments. Indeed, NDRG3 more efficiently coprecipitated with Stx5L_aa1-54 in the presence of lactate (Fig. 3F). Next, we aimed to confirm the NDRG3-Stx5 native interaction by endogenous co‐immunoprecipitations. As shown in Fig. 3G, we were unable to coimmunoprecipitate Stx5 when NDRG3 antibodies were used for pull-downs and vice versa. However, we found that NDRG3 was able to coimmunoprecipitate Stx5 in the presence of lactate and vice versa. NDRG3-deficient HEK293T cells were used to control for unspecific binding of antibodies (Fig. 3G). Since isoform-specific Stx5 antibodies and Stx5-deficient cell lines were unavailable, we could not test whether NDRG3 interacts specifically with endogenous Stx5L. However, we consistently detected SCFDI in the NDRG3/Stx5 complex, independent of whether endogenous coimmunoprecipitation was performed with NDRG3 or Stx5 antibodies. Furthermore, lactate did not affect the ability of Stx5 to coprecipitate SCFD1 at the endogenous level, which argues against NDRG and SCFD1 competing for Stx5 binding. However, Stx5 isoform-specific antibodies or isoform-specific Stx5-deficient cells will be necessary to answer this question in future studies.

To follow up on the finding that the NDRG3 and Stx5 interaction is lactate dependent, we analyzed the subcellular localization of NDRG3 and Stx5 by immunofluorescence in cells preincubated with or without lactate (Fig. 3H_I). In the presence of lactate, NDRG3 shifted from a cytoplasmic/nuclear distribution to a punctate cytoplasmic pattern, where it partially colocalized with Stx5 (PCC values: without lactate = 0.47 ± 0.02; with lactate = 0.57± 0.02; Fig. 3H_II). Notably, nuclear NDRG3 staining was reduced after lactate treatment. In summary, these data strongly suggest that lactate promotes binding of endogenous NDRG3 to Stx5.

NDRG3 Mediates Deceleration of ER-to-Golgi Transport in Response to Hypoxia.

Since Stx5L has been implicated in ER-Golgi vesicular transport (8) and NDRG3 in hypoxia, we next investigated the impact of the Stx5L–NDRG3 interaction on the secretory pathway under norm- and hypoxia, by using a Regulated Secretion/Aggregation assay.

Cells expressing a constitutively secreted reporter protein, pC4—a GFP-tagged variant of human growth hormone (hGH) fused to multiple aggregation domains was generated. These domains cause newly synthesized GFP-hGH to aggregate and be retained within the ER (4). Upon incubation with a specific ligand/solubilizer, the aggregates dissolve, leading to a synchronous release of GFP-hGH from the ER to the Golgi complex. In the Golgi, the reporter undergoes Furin-mediated cleavage before being secreted (SI Appendix, Fig. S3A).

To analyze the impact of NDRG3 and hypoxia on regulated secretion, we incubated wild-type and NDRG3-deficient HEK293T cells expressing the reporter construct with or without the hypoxia mimetic deferoxamine (DFX) and determined the intracellular retention of the reporter by flow cytometry. While cells at control conditions showed a decrease of intracellular fluorescence over time, treatment with DFX led to a strong retention of intracellular GFP fluorescence over 80 min (Fig. 4A). Next, we asked whether this effect is dependent on NDRG3 and repeated the pC4-reporter assay in NDRG3-deficient cells (Fig. 2D_II), which showed a significantly reduced intracellular GFP half-life under normoxia, when compared to wild-type cells, suggesting that NDRG3 regulates the secretory pathway also at normoxic conditions (Fig. 4A). In addition, reporter protein secretion was not DFX-dependent in NDRG3-deficient cells (Fig. 4A). NDRG3 deficiency led to a significant decrease of intracellular GFP half-life (NDRG3^−/−: t_1/2 = 72.2 ±5.3 min) compared to wild-type cells (wt: t_1/2 = 114.0 ±13.2 min) and the half-life of the intracellular GFP reporter was not significantly changed by DFX treatment (NDRG3^−/− (+DFX): t_1/2 = 79.4 ± 2.6 min) (Fig. 4A). The intracellular GFP half-life of wild-type cells treated with DFX could not be determined, as no decrease of GFP fluorescence over our time course was observed. In summary, these results strongly suggest that NDRG3 negatively regulates the secretory pathway under norm- and hypoxia.

Fig. 4.

Hypoxia reduces ER-to-Golgi trafficking, a process that is dependent on NDRG3 and lactate. (A) HEK293 cells were transfected with the pC4 reporter construct and cultured with or without 5 mM DFX for 24 h. After addition of solubilizer for the indicated time periods, mean cellular GFP fluorescence signal was measured by flow cytometry. Under control conditions, the intracellular GFP signal decreased over time and plateaued after 80 min. Signals at time point 0 were set to 1, and a nonlinear fit for one phase decay was performed using Graphpad Prism. Treatment with DFX led to stable GFP signal over 80 min, and no reduction in the GFP fluorescence could be detected. For HEK293 NDRG3-deficient cells intracellular GFP signal declined more rapidly in NDRG3-deficient cells, when compared to wild-type cells. Treatment with DFX as described above had no measurable effect on the transport rates of the pC4 reporter. (B_I-II) HeLa cells stably expressing the pC4 reporter construct were cultivated with or without 5 mM DFX. After 24 h, solubilizer was added for 0 or 10 min, respectively, followed by fixation of cells. ±DFX-treated cells were labeled with GM130 and PDI as markers for cis-Golgi and ER, respectively. PCCs were determined and plotted (DFX/0 min: n = 30, DFX/10 min: n = 21. Ctr/0 min: n = 25, ctrl/10 min: n = 31 cells. ***P ≤ 0.001 two-sited Student’s t test, Scale bar, 20 µm, blue: DAPI staining). (C_I-III) HEK293 WT or NDRG3^−/− cells were transfected with the Str-li_SBP-mcherry-Golgin84 RUSH Plasmid and cultivated at normoxia (21% O₂) (Fig. S3B), hypoxia (1% O₂), or in the presence of 50 mM L-lactate (Fig. S3B) for 24 h. At Time point zero most of the RUSH signal was located for all conditions in the ER, respectively. 20 min after addition of biotin, the RUSH signal shifted toward the cis-Golgi apparatus for wild-type and NDRG3^−/− cells under normoxia. Remarkably, less colocalization was detected for the ER marker PDI in NDRG3^−/− cells compared to wild-type after 20 min. While under hypoxic and lactate conditions 20 min after biotin treatment the RUSH signal in wild-type cells was significant less localized in the Golgi apparatus compared to normoxic conditions, no effect was detected under hypoxia and lactate for NDRG3^−/− cells (n = 3 at least 30 cells per replica and condition) (PCC; mean PCC ± SD, 2Way Anova, Tukey post hoc comparison, n = 3 (at least 20 cells per replicate). ****P ≤ 0.0001). (Scale bar, 10 µm, * transfected cells, blue DAPI staining.)

To explore ER-to-Golgi trafficking in more detail, we analyzed the subcellular localization of the pC4 reporter protein in stably transfected HeLa cells (4) that were incubated with or without DFX at 0 and 10 min after adding solubilizer by costaining with the ER and cis-Golgi marker proteins PDI and GM130, respectively (Fig. 4B). At time point 0 in both conditions, the pC4 reporter was predominantly present in the ER (PCCs: pC4/PDI 0.58 ± 0.04 (−DFX and pC4/PDI 0.56 ± 0.04 (+DFX) and not in the Golgi (PCCs: pC4/GM130 0.27 ± 0.02 (−DFX) and pC4/GM130 0.29 ± 0.02 (+DFX) (Fig. 4B_I and B_II). 10 min after addition of the solubilizer, the reporter was enriched within the Golgi of cells incubated without DFX (PCC: pC4/PDI 0.28 ± 0.02, pC4/GM130 0.61 ± 0.02), whereas in DFX-treated cells the reporter protein was found mainly in the ER (Fig. 4B_II), PCC: pC4/PDI 0.67 ± 0.04 and pC4/GM130 0.28 ± 0.03), indicating that the induction of hypoxia decelerates ER-to-Golgi transport.

Next, we corroborated these results by inducing hypoxia by oxygen deprivation (1% O₂) or preincubation of cells with lactate and by utilizing the retention-using-selective-hooks (RUSH) assay to monitor the synchronous release of the Golgi reporter Golgin-84 from the ER to the Golgi compartment (35). As shown in Fig. 4C and SI Appendix, Fig. S3B), before biotin treatment Golgin-84 was trapped in the ER in all conditions (normoxia, hypoxia, and lactate, Fig. 4C_I and SI Appendix, Fig. S3B). After biotin treatment for 20 min at normoxia, Golgin-84 was present in the cis-Golgi in wild-type and NDRG3-deficient HEK293 cells (SI Appendix, Fig. S3B_II). In contrast, biotin treatment for 20 min at hypoxia or in the presence of lactate resulted in a significantly reduced amount of Golgin-84 in the cis-Golgi in wild-type but not NDRG3-deficient cells. This strongly suggests that the transport rate of the Golgin-84 reporter from the ER to the cis-Golgi is dependent on NDRG3 during hypoxia and increased intracellular lactate levels. In addition, under normoxia, we found less Golgin-84 reporter colocalizing with the ER-marker PDI after 20 min of biotin treatment and a stronger colocalization with GM130 in NDRG3-deficient HEK293 cells. This indicates that NDRG3 also decelerates ER-to-Golgi transport under normoxic conditions (Fig. 4C_III).

In summary, we have identified NDRG3 as an interacting protein of the long isoform of Stx5, linking mechanistically hypoxia to ER-to-Golgi transport and the secretory pathway.

Discussion

We have identified NDRG3 as a regulator of the ER-to-Golgi SNARE complex and show that it interacts with the N-terminal domain of Stx5L. Stx5 exists in animals as a long (Stx5L) and a short isoform (Stx5S) with a large degree of overlapping localization within the ER, ERGIC, and Golgi compartments. However, Stx5L predominantly resides in the ER/ERGIC, due to an arginine/lysine–based ER retrieval signal, whereas Stx5S is primarily localized to the Golgi apparatus (36, 37). Accordingly, Stx5L is likely involved in ER-to-ERGIC/Golgi and Stx5S in intra-Golgi trafficking (8). Recently, patients with a fatal multisystemic disease, including liver failure, were identified as carrying a homozygous mutation in an alternative start codon of Stx5S. Patient-derived fibroblasts show normal levels of Stx5L but no expression of Stx5S and display compromised ER-to-ERGIC/ Golgi trafficking and mislocalized glycosyltransferases, causing glycosylation defects such as mucin-type O-glycosylation (8), suggesting that Stx5L is unable to compensate for Stx5S.

Our results indicate that NDRG3 specifically binds to the N-terminal region of Stx5L, thereby impacting Stx5L function in ER-to-ERGIC/Golgi trafficking, but most likely not intra-Golgi trafficking mediated by Stx5S (Fig. 5) (8, 37). Since the structure and function of the N-terminal domain of Stx5L are unknown, we can only speculate about the mechanism of NDRG3-mediated regulation of Stx5L. Given the close proximity of the NDRG3 and SCFD1 binding sites, one explanation could be that both proteins compete for Stx5L binding. Our finding that a SCFD1-binding-deficient Stx5L variant precipitated NDRG3 more efficiently when compared to wild-type Stx5L agrees with a competition model (Fig. 3C). However, the fact that SCFD1 could be endogenously coprecipitated with Stx5 even in the presence of NDRG3 and lactate argues against a competitive binding model of NDRG3 and SCFD1 for Stx5L (Fig. 3G). Thus, further experiments are required, for example, endogenous co-IPs of Stx5 and SCFD1 in isoform-specific Stx5 deficient cell lines under conditions dependent on NDRG3 and lactate to confirm or refute a competition model. Alternatively, Stx5 can adopt an active “open” and an inactive “closed” conformation (the latter is depicted in the AlphaFold prediction model in Fig. 3A) characterized by the H_abc domain folded back onto the SNARE motif, preventing SNARE complex assembly (5, 38). It is also conceivable that binding of NDRG3 to the N-terminal domain stabilizes the “closed” conformation thereby interfering with ER-to-Golgi SNARE complex function. However, the molecular mechanism by which NDRG3 negatively regulates ER-to-Golgi complex-mediated membrane fusion remains to be elucidated.

Fig. 5.

Role of NDRG3 in ER-to-ERGIC/Golgi transport. A simplified cartoon or ER-to-ERGIC/Golgi transport illustrating the role of NDRG3 in this process under norm- and hypoxia. At normoxic conditions, SCFD1 binds to both isoforms of Stx5, regulating the activity of SNARE complexes Stx5L or Stx5S+GOSR2+Bet1-Sec22b promoting ER-to-Golgi and intra-Golgi transport, respectively. At hypoxia, NDRG3 bound to lactate negatively regulates ER-to-Golgi transport by binding to Stx5L (created with bioRender.com).

Our data suggest that during normoxia, SCFD1 promotes ER-to-ERGIC/Golgi and intra-Golgi vesicular transport by binding to both isoforms of Stx5, whereas NDRG3 hinders ER-to-ERGIC/Golgi transport by preferentially binding the Stx5L isoform. SCFD1 belongs to the family of SM proteins, which act as molecular clasps cooperating with SNARE complexes in membrane fusion (37). The inhibitory effect of NDRG3 on ER-to-ERGIC/Golgi transport becomes more apparent under hypoxia in the presence of accumulating lactate by binding of NDRG3 to Stx5L (Figs. 3G and 5). Since NDRG3 has been described as a lactate binding protein that contributes to the adaption of cells to hypoxic conditions (26), it is tempting to speculate that decelerated ER-to-Golgi transport supports viability of eukaryotic cells under oxygen restriction. Similar to HIF-1, NDRG3 is oxygen-dependently regulated by PHD enzymes, but little is known about the role of NDRG3 in the adaptation of cells to hypoxia. We found that hypoxia efficiently impairs reporter protein secretion in wild-type but not NDRG3-deficient cells (Fig. 4A). In addition, hypoxia and lactate effectively decelerates ER-to-Golgi transport (Fig. 4 B and C). These data strongly suggest that NDRG3 is part of the molecular machinery connecting hypoxia to membrane trafficking likely by lactate-bound NDRG3 acting on Stx5L (Fig. 5).

Previous studies have analyzed the effect of hypoxia on ER-to-Golgi trafficking in primary beta cells. It was shown that hypoxic conditions lead to severely reduced ER-to-Golgi transport and induction of the unfolded protein response (29). Reduced expression of coatomer protein complex proteins indicates that reduction of the ER-to-Golgi trafficking protein machinery could be part of cellular adaptation to hypoxic conditions (29). Notably, these effects were independent of HIF-1α, but dependent on c-Jun N-terminal kinase (JNK) since inhibition of JNK led to partial or complete rescue of the ER-to-Golgi trafficking (29). Regulation of the SNARE complex by phosphorylation has been reported previously, for example, phosphorylation of non-neuronal VAMPs by PKC, or phosphorylation of the SM protein Munc-18 that impairs its interaction with Syntaxin-1A and SNAP25 (39, 40). Given that the C terminus of NDRG3 is both necessary and sufficient for Stx5L binding and that this region contains numerous potential phosphorylation sites, the NDRG3–Stx5L interaction is likely regulated by phosphorylation (41).

In summary, our results uncover an unexpected mechanism by which hypoxic conditions decelerate ER–Golgi transport, requiring both NDRG3 and lactate. We found that NDRG3, a primarily cytosolic protein, interacts with the N-terminal domain of Stx5L to inhibit ER-to-ERGIC/Golgi trafficking. Given that elevated NDRG3 expression is linked to tumor development and poor prognosis, we propose that suppression of the secretory pathway may represent an important adaptive response of tumor cells to hypoxic stress.

Materials and Experimental Procedures

For a detailed description of plasmids and antibodies used, cultivation and transfection of cells, CRISPR-Cas9 nickase mediated knockout of NDRG3, Volcano Plot, Molecular modeling, and PLA assay, please refer to SI Experimental Procedures.

Affinity Purification–Mass Spectrometry (AP–MS).

HEK293T cells were transfected with wild-type GOSR2-3xFLAG or NDRG3-HA, NDRG2-HA, NDRG1-HA, and NDRG4-HA, respectively, and AP-MS were performed as described previously (7, 42, 43). Briefly, after transient transfection of bait constructs, cells were pooled and lysed, and cellular debris was removed by centrifugation. The supernatant was used for anti-flag precipitation or anti-HA precipitation, and the eluates were subjected to mass spectrometry. Data were analyzed with the MiST (mass spectrometry interaction statistics) AP-MS scoring algorithms or represented as a volcano plot as described previously (7).

Coimmunoprecipitation.

Cells were harvested by scraping in PBS with cOmplete™ Protease Inhibitor Cocktail (Merck, Darmstadt, Germany) and centrifugation. Cell pellets were resuspended in EBC lysis buffer (120 mM NaCl, 50 mM Tris-Cl, 1% (v/v) NP-40) and sonified using a Branson Sonifier. Cell debris was removed by centrifugation. 50 µL of the supernatant, mixed with Laemmli buffer was used as the lysate sample and the rest was incubated with primary antibodies for 12 h at 4 °C The lysate was then incubated with 50 µL Dynabeads Protein G (Thermo Fisher Scientific, Dreieich, Germany) that have been blocked with 5% (w/v) BSA (Sigma-Aldrich, Taufkirchen, Germany) in EBC buffer. The beads were then washed 5 times with EBC buffer. After the final wash, the beads were resuspended in Laemmli buffer and incubated at 55 °C for 20 min. Supernatants were collected and analyzed by immunoblots. Using GFP-trap or HA-magnetic beads: Cellular lysates were prepared as described above. Lysates expressing GFP-tagged proteins were incubated for 2 h with 50 µL of GFP-trap beads (Chromotek, Martinsried, Germany) or HA-magnetic beads (Thermo Fisher Scientific, Dreieich, Germany).

PC4 and RUSH Reporter Assays.

For the quantitative pC4 assay analyzed by flow cytometry, cells were incubated with solubilizer (rapamycin; Cayman Chemical Company, Ann Arbor, USA) for indicated time periods. Cells were resuspended in PBS and analyzed in a FACS Calibur (BD Bioscience, Germany, Heidelberg). A cell population was selected, and the mean GFP fluorescence signal was measured for each sample. Three technical replicates were measured four times each. For subcellular analysis, the cells were seeded on poly-L-lysine hydrobromid (PLL, Sigma-Aldrich, Taufkirchen, Germany) coated cover slips and incubated for indicated time points with rapamycin. Subsequently, cells were analyzed by immunocytochemistry. DFX was applied for 24 h with a final concentration of 5 mM.

On coverslip seeded HEK293T wild-type or HEK NDRG3^−/− cells were transfected with Str-Ii_SBP-mCherry-Golgin84 48 h prior Biotin treatment. After 24 h, hypoxia incubated cells were transferred to 1% O₂ and cultivated for 24 h. For lactate-treated cells, medium was replaced with complete medium containing a final concentration of 50 mM lactate for 24 h. Cells were treated for either 0 min or 20 min with 40 µM Biotin final concentration. After treatment, cells were fixed and stained as described in section “Immunocytochemistry”. As primary antibodies, anti-GM130 mouse (BD Bioscience, USA) and anti-PDI rabbit (Sigma-Aldrich, Steinheim, Germany) antibodies were used.

Immunocytochemistry.

On coverslips seeded cells were washed with PBS and fixed for 10 min with 4% PFA/PBS. Cells were washed again with PBS and permeabilized with 0.1% Triton-X-100/PBS. After three more washing steps with PBS, cells were blocked with 2% goat serum in PBS for 1 h. Incubation with the primary antibody, diluted in 2% goat serum/PBS occurred for 4 h at room temperature or overnight at 4 °C. Cells were washed again and incubated with the secondary antibody in PBS for 1 h. After final washing steps, cover slips were mounted on microscope slides with Mowiol 4-88 (Sigma-Aldrich, Taufkirchen, Germany) including Hoechst Solution (Invitrogen, Carlsbad, USA) following the manufacturer’s recommendations.

Recombinant Protein Expression, Purification, and GST Pulldown.

For recombinant protein expression, E. coli BL21 or E. coli BL21 tuner (expressing GroEL chaperone after arabinose induction) were used. Bacteria were transformed with pGEXT-I-GST-hSTX5 w/o TM, pGEXT-I-GST-hSTX5-Habc, pGEXT-I-GST-hGOSR2 w/o TM, pGEXT-I-GST-hGOSR2-Habc, or pET32b-His-hNDRG3, respectively. Expression cultures were inoculated by overnight culture with an optical density of 0.1 at 600 nm (OD600) and incubated at 37 °C. At an OD600 of 0.6, cultures were induced with 0.2 mM isopropyl-β-D-thiogalactopyranoside (IPTG, Roth, Karlsruhe, Germany) and incubated for 5 h at 24 C. Cells were then pelleted at 6,000 x g and 4 C for 30 min, washed with PBS, pelleted again, and stored at −20 C until further use. Pellets were resuspended in PBS with cOmplete (Sigma-Aldrich, Taufkirchen, Germany) or His buffer (300 mM NaCl, 50 mM NaH₂PO₄, 10 mM imidazol) with cOmplete (Sigma-Aldrich, Taufkirchen, Germany) and lysed either by French press or by sonication. GST-tagged proteins were purified using Glutathion Sepharose Superflow beads (Thermo Fisher Scientific, Dreieich, Germany), and His-tagged proteins were purified using Ni²⁺-NTA Agarose Protino beads (Macherey-Nagel, Düren, Germany) by incubating the lysates for 1 h at 4 °C under rotation. The supernatant was then discarded, the beads washed, and eluted with GSH buffer (1.5 mg/mL reduced glutathione in PBS) or His elution buffer with increasing imidazole concentration (50 mM, 100 mM, 250 mM, and 500 mM) at 2 mL each. The elution fractions were stored at −20 °C until further use. 0.4 µM GST or GST fusion protein was incubated with 0.4 µM His-NDRG3 in 500 µL PBS + 0.1% Triton X-100 for 1 h at 4 °C under rotation. After centrifugation at 20,000×g, input samples were taken, and the remainder was incubated with 50 µL of equilibrated GST beads for 1 h at 4 °C under rotation. Beads were then washed with PBS + 0.1% Triton X-100, and proteins were eluted by addition of Laemmli sample buffer (Bio-Rad, Feldkirchen, Germany) and incubation at 55 °C for 25 min.

Supplementary Material

Appendix 01 (PDF)

Dataset S01 (XLSX)

Dataset S02 (XLSX)

Data, Materials, and Software Availability

Raw MS data has been deposited at MassIVE under accession number MSV000099757 (44) and Proteome Exchange under accession number PXD070322 (45). All other data are included in the article and/or supporting information.