Palmitoylated CKAP4 regulates mitochondrial functions through an interaction with VDAC2 at ER–mitochondria contact sites

Intracellular organelles coordinate complex signaling, metabolism and gene expression mechanisms in the cell through functional or physical interactions with one another (Wu et al., 2018). Of the various combinations of interactions among organelles, that between the endoplasmic reticulum (ER) and mitochondria plays pivotal roles in cellular functions, including lipid transport, apoptosis control, energy metabolism and Ca²⁺ signaling (Marchi et al., 2018).

The ER is a continuous membrane network that can be divided into sheet-like structures connected to the nuclear envelope and a network of tubules extending throughout the periphery of the cells (Shibata et al., 2006; Westrate et al., 2015). Both the absolute and relative abundance of ER sheets and tubules vary with cell type and their balance is tightly regulated. The ER is dynamic, with sheets rearranging and tubules moving and fusing to form ‘three-way junctions’ (Du et al., 2004; Lee and Chen, 1988). In addition, the ER plays a critical role in many cellular processes, including the regulation of Ca²⁺ homeostasis, as well as protein synthesis, protein modification and lipid synthesis (Baumann and Walz, 2001; Marchi et al., 2018).

The main effectors of the ER Ca²⁺ release machinery are inositol 1,4,5-trisphosphate (IP3) receptors (IP3Rs), which facilitate the release of Ca²⁺ from ER stores in response to IP3 (Patel et al., 1999). Three different gene products (types I–III) assemble as large tetrameric structures. Mitochondria can take up Ca²⁺ into their matrix directly from IP3Rs through voltage-dependent anion channels (VDACs) (Shoshan-Barmatz et al., 2006; Tsujimoto et al., 2006). The three isoforms (VDAC1–VDAC3) show comparable channel properties, despite having different effects on cell death, and form a complex with IP3Rs at ER–mitochondria contact sites (Szabadkai et al., 2006). Under physiological conditions, Ca²⁺ in mitochondria stimulates oxidative metabolism through the modulation of Ca²⁺-sensitive dehydrogenases and metabolite carriers (McCormack et al., 1990). However, mitochondrial Ca²⁺ overload damages mitochondrial morphology through the increase in ER–mitochondria contact sites by various stresses, including fragmentation by the recruitment of the GTPase dynamin-related protein 1 (DRP1; also known as DNM1L) and/or cristae remodeling by optic atrophy 1 (OPA1), and also impairs functions, with one consequence being decreased ATP production (Jahani-Asl et al., 2010; Raffaello et al., 2016; Rizzuto et al., 2012). In addition, loss of cristae structure leads to the release of cytochrome c and cell death (Eisner et al., 2018; Pernas and Scorrano, 2016).

Cytoskeleton-associated protein 4 (CKAP4; also known as CLIMP-63 and ERGIC-63) is a non-glycosylated type II transmembrane protein located in the ER (Schweizer et al., 1993, 1995b). Several possible functions of CKAP4 in the ER have been reported, including segregation of ER sheets close to the nucleus (Klopfenstein et al., 2001), maintenance of luminal width through intermolecular binding of the luminal region of CKAP4 localized on opposing cisternal membranes (Shibata et al., 2010), binding of the cytoplasmic region of CKAP4 to microtubules to create a link between the ER and microtubules (Klopfenstein et al., 1998; Vedrenne and Hauri, 2006), acting as a Dicer-binding protein and regulating the microRNA pathway and mRNA translation by anchoring Dicer to the ER (Pépin et al., 2012), and finally, through binding to gentamicin in the ER lumen, participating in gentamicin-induced apoptosis in proximal tubule cells (Karasawa et al., 2010). Thus, although CKAP4 is likely to be involved in the regulation of the morphology and functions of the ER, the roles of CKAP4 in the functions of other organelles remain obscure.

A small population of CKAP4 is localized to the plasma membrane and functions as a receptor for extracellular ligands (Kikuchi et al., 2017). The extracellular region of plasma membrane-located CKAP4, which is the same as the ER luminal region, functions as a receptor for surfactant protein A (Gupta et al., 2006), tissue plasminogen activator (Razzaq et al., 2003), anti-proliferative factor (Conrads et al., 2006) and dickkopf1 (DKK1) (Kimura et al., 2016), and also binds to integrin to regulate its recycling (Osugi et al., 2019). Thus, CKAP4 has multiple functions, depending on its subcellular localization and interacting proteins. CKAP4 is modified with palmitate at Cys¹⁰⁰ (Schweizer et al., 1995a; Zhang et al., 2008) and this palmitoylation is required for the localization of CKAP4 to the lipid raft of the plasma membrane, for DKK1-dependent AKT activation and for cancer cell proliferation (Sada et al., 2019). However, the role of palmitoylation of CKAP4 in the ER is unclear.

In this study, we observed mitochondrial morphological changes and dysfunction in CKAP4 knockout (KO) cells, in addition to ER morphological changes. We determined that CKAP4 interacts with VDAC2 and modulates the functional coupling between IP3R and VDAC2. In addition, CKAP4 was involved in the formation of ER–mitochondria contact sites, Ca²⁺ influx into mitochondria, mitochondrial respiration and cancer cell proliferation, and these functions required palmitoylation of CKAP4. Our findings provide new roles for CKAP4 at ER–mitochondria contact sites.

HeLaS3 cells with CKAP4 KO were generated with the CRISPR/Cas9 system (CKAP4 KO HeLaS3 cells) (Fig. S1A). CKAP4 was primarily localized to ER sheets in control HeLaS3 cells (control cells) and the ER sheets in CKAP4 KO cells were distributed throughout the cytoplasm (Fig. S1B). Electron microscopy revealed that the luminal width was reduced from 60 nm to 40 nm (Fig. S1C). These results are consistent with previous observations (Shibata et al., 2010). To observe the mitochondrial morphology, mitochondrial matrix-targeted GFP (Mito-GFP) was transiently expressed in control and CKAP4 KO HeLaS3 cells (Fig. 1A). The expression level of transiently transfected Mito-GFP was examined in each cell by microscopic imaging, and cells expressing about ±10% of the most frequent expression levels were selected for analyses. These cells were found to have an expression level of Mito-GFP of 51,014±5205 (arbitrary units of fluorescence intensity±standard deviation) in control cells and 51,485±5953 in CKAP4 KO cells (n=100, P=0.48; Mann–Whitney U-test). These results indicate that the variance of transiently expressed Mito-GFP is similar in control and CKAP4 KO cells. Under these conditions, CKAP4 KO cells exhibited a significantly fragmented and attenuated mitochondrial network compared with the tubular mitochondrial reticulum of control cells (Fig. 1A). Mitochondrial fragmentation was observed in 10% of control and 70% of CKAP4 KO cells (data not shown). In addition, the number and the individual size of mitochondria were increased and decreased, respectively, in CKAP4 KO cells compared with control cells under the conditions where the total mitochondrial area was similar (Fig. 1A). The mitochondrial mass, which was analyzed through MitoTracker Green, and the expression levels of proteins, such as VDAC2, glucose-regulated protein 75 (GRP75; also known as HSPA9), adenine nucleotide translocase 2 (ANT2; also known as SLC25A5), COXIV, the type 1 IP3R (IP3R1; also known as ITPR1) and IP3R3 (ITPR3), were not changed in control and CKAP4 KO cells (Fig. S1D,E). These results are similar to the mitochondrial morphological alterations caused by knockdown of YME1L or annexin A6 (Chlystun et al., 2013; Stiburek et al., 2012).

Electron microscopy examination revealed the frequency of alterations in mitochondrial cristae structures in CKAP4 KO cells (Fig. 1B). CKAP4 KO mitochondrion appeared to be round, with loss of the well-defined cristae structures (An et al., 2012; Dalla Rosa et al., 2014; Stiburek et al., 2012). The individual mitochondrion area was calculated to be 0.85 µm² and 0.37 µm² in control and CKAP4 KO cells. The aspect ratio of mitochondria was 1.90 and 1.34 in control and CKAP4 KO cells, respectively (Fig. 1B). The split-GFP system, which can detect ER–mitochondria contact sites (Kakimoto et al., 2018), showed that the numbers of GFP punctate increase in CKAP4 KO HeLaS3 cells (Fig. 1C). The finding was confirmed by electron microscopy, which revealed that the ratio of contact sites to mitochondrial perimeters and the numbers of contact sites per mitochondria were increased in CKAP4 KO cells (Fig. 1D), whereas the length and width of contact sites were not changed in control and CKAP4 KO cells (Fig. 1D). The mitochondrial morphology was also fragmented in CKAP4 KO U2OS cells and the numbers of the contact sites were increased in the cells (Fig. S1F,G). Therefore, CKAP4 may be necessary to maintain healthy mitochondria through ER–mitochondria contact sites, because it is known that the interaction of the two organelles is important for mitochondrial structure and function (Hayashi et al., 2009).

To address the involvement of CKAP4 in mitochondrial functions, CKAP4–HA was stably expressed in S2-CP8 pancreatic cancer cells and CKAP4-binding proteins were precipitated from whole-cell lysates (Fig. 2A). Among candidate proteins, voltage-dependent anion-selective channel protein 2 (VDAC2) was further examined because VDAC2 is involved in various mitochondrial functions, including Ca²⁺ transport, apoptosis and oxidative phosphorylation (Mannella, 1992; Tsujimoto et al., 2006). The mRNA level of VDAC2 was more abundant than that of VDAC1 and VDAC3 in HeLaS3 cells (Fig. S2A). CKAP4 formed a complex with VDAC2 at the endogenous level in HeLaS3 cells and preferentially interacted with VDAC2 compared to VDAC1 and VDAC3 when VDAC family proteins were transiently expressed in X293T cells (Fig. 2B,C). Although the first 11 amino acid residues of VDAC2 are not aligned with the sequence of other VDACs, a VDAC2 mutant with deletion of 11 N-terminal amino acids (Δ1–11) bound to CKAP4 to a similar extent as VDAC2 wild-type (WT) (Fig. S2B).

The N-terminal cytoplasmic region (aa 1–106) of CKAP4 has been reported to regulate ER rearrangement (aa 2–21), link the ER to microtubules (aa 24–101) and be modified with palmitate at Cys¹⁰⁰ (Klopfenstein et al., 1998; Schweizer et al., 1994; Zhang et al., 2008). Thus, N-terminal deletion mutants and a palmitoylation-deficient mutant (C100S) of CKAP4 were transiently expressed in X293T cells (Fig. 2D). CKAP4 (Δ2–21)–HA and CKAP4 (Δ24–99)–HA formed a complex with FLAG–VDAC2 to a similar extent and to a lesser extent, respectively, to CKAP4^WT–HA. On the other hand, CKAP4 (Δ101–106)–HA and CKAP4^C100S–HA did not (Fig. 2D). In X293T cells, transiently expressed CKAP4^WT–HA was palmitoylated but CKAP4^C100S–HA was not (Fig. 2E). Under the same conditions, CKAP4 (Δ2–21)–HA and CKAP4 (Δ24–99)–HA was palmitoylated to a similar extent and to a lesser extent, respectively, to that of CKAP4^WT–HA, but CKAP4 (Δ101–106)–HA was not (Fig. 2E). These results suggest that the amino acid residues 100–106, which include a palmitoylation site, are necessary for the binding of ER-located CKAP4 to mitochondrial VDAC2.

There are 23 members of the ZDHHC protein family containing palmitoyl acyltransferases (PATs) with a DHHC (Asp-His-His-Cys) motif (Fukata et al., 2004; Roth et al., 2002). ZDHHC2 localizes to the ER/Golgi (Ohno et al., 2006) and is required for the palmitoylation of CKAP4 (Zhang et al., 2008). Consistent with the previous report, palmitoylation of CKAP4 was reduced in ZDHHC2-depleted HeLaS3 cells (Fig. 2F; Fig. S2C). Furthermore, it was confirmed that the binding of CKAP4 to VDAC2 was reduced in ZDHHC2-depleted cells (Fig. 2G). Taken together, these results suggest that ER-localized CKAP4 binds to VDAC2 in a palmitoylation-dependent manner.

ER–mitochondria contact sites are also known as mitochondria-associated ER membranes (MAMs) (Giacomello and Pellegrini, 2016; Vance, 1990). Consistent with previous observations (Csordás et al., 2006; Poston et al., 2013), subcellular fractionation of HeLaS3 cells confirmed that calnexin is present in MAM, microsome and crude mitochondria fractions (Fig. 3A). On the other hand, IP3R was primarily observed in the MAM fraction, and VDAC2 was detected in the MAM and pure mitochondrial fractions, but not in the microsomal fraction (Fig. 3A). CKAP4 was present in the MAM, microsome and crude mitochondrial fractions, but not in the pure mitochondria fraction, of control HeLaS3 cells, with stably expressed CKAP4^C100S showing a similar subcellular localization in CKAP4 KO cells (CKAP4 KO/CKAP4^C100S cells) (Fig. 3A). Light microscopy analysis showed that a part of CKAP4 is colocalized with ER-mitochondria contact sites, which are indicated by split-GFP (Fig. S3A). Immunoelectron microscopy analysis revealed that both endogenous CKAP4 and CKAP4^C100S were distributed throughout the ER membrane (Fig. 3B). Thus, CKAP4 is present in ER membranes, including MAMs, and unlikely to be enriched in MAMs through palmitoylation.

The split-GFP system revealed that overexpression of CKAP4, but not that of VDAC2, decreases the numbers of ER-mitochondria contact sites (Fig. S3B). The increase in abnormal mitochondria and ER-mitochondria contact sites induced by CKAP4 KO, which were observed by electron and light microscopy, were rescued by the expression of CKAP4^WT but not by that of CKAP4^C100S (Fig. 3C,D). At MAMs, VDAC1 is physically linked to IP3R1 through glucose-regulated protein 75 (GRP75) (Szabadkai et al., 2006). IP3R1 mRNA was more abundantly expressed in HeLaS3 cells compared with the IP3R2 (also known as ITPR2) and IP3R3 mRNAs (Fig. S3C). FLAG–VDAC2 was localized to the mitochondrial in HeLaS3 cells (Fig. S3D). A proximity ligation assay (PLA) showed some signals, indicating that there was interaction between FLAG–VDAC2 and endogenous IP3R3 in control HeLaS3 cells, and the number of signals was increased in CKAP4 KO cells (Fig. 4A). Expression of CKAP4^WT in CKAP4 KO cells decreased the number of PLA signals, unlike CKAP4^C100S (Fig. 4A). In a co-immunoprecipitation assay, FLAG–VDAC2 weakly precipitated endogenous GRP75 in control cells, and CKAP4 KO enhanced the formation of the complex (Fig. 4B). Since the expression levels of proteins constituting ER–mitochondria contact sites were not changed in control and CKAP4 KO cells, as shown in Fig. S1E, it is unlikely that the increase in protein–protein interaction at ER-mitochondria contact sites by CKAP4 KO is due to increased protein levels. The increase in ER–mitochondria contact sites in CKAP4 KO cells expressing split-GFP was also rescued by knockdown (KD) of VDAC2, IP3R1 or IP3R3 (Fig. 4C; Fig. S3E). Taken together, these results suggest that CKAP4 at ER–mitochondria contact sites suppresses the interaction between VDAC2, GRP75 and IP3R, and that CKAP4 is involved in Ca²⁺ transport between the ER and mitochondria.

To examine the role of CKAP4 in the regulation of Ca²⁺ flow from the ER into the mitochondria, intramitochondrial Ca²⁺ concentrations ([Ca²⁺]_m) in HeLaS3 cells were imaged by expressing a mitochondrial Ca²⁺ sensor CEPIA3mt with mCherry, which was used for normalization of the expression levels (Fig. 5A). [Ca²⁺]_m were increased in CKAP4 KO cells under the conditions where mitochondrial mass, which was measured using MitoTracker Deep Red, was not changed (Fig. S4A). The phenotype was rescued by CKAP4^WT expression and also rescued by VDAC2 KO, IP3R1 KD or IP3R3 KD (Fig. 5A,B; Fig. S4B). These results indicate that the basal Ca²⁺ levels are increased in the mitochondria of CKAP4 KO cells. To examine whether CKAP4 KO affects stimulation-dependent Ca²⁺ release from the ER, HeLaS3 cells were stimulated with ATPγS to measure in [Ca²⁺]_m and Ca²⁺ concentrations in the cytosol ([Ca²⁺]_c) chronologically. In control HeLaS3 cells, ATPγS increased [Ca²⁺]_m with a plateau at ∼30 s (Fig. 5C), while it gradually increased [Ca²⁺]_c, reaching a plateau at ∼120 s (Fig. 5D). In CKAP4 KO HeLaS3 cells, basal [Ca²⁺]_m were higher than control cells, but the fold-increase of [Ca²⁺]_m observed upon ATPγS stimulation was similar to that of control HeLaS3 cells [Fig. 5C, fold change compared with basal intensity (mean±s.d.), control, 2.38±0.10; CKAP4 KO, 2.34±0.11]. The stimulation-dependent change in [Ca²⁺]_c did not show significant differences in control and CKAP4 KO HeLaS3 cells (Fig. 5D, fold change compared with basal intensity, control, 2.70±0.20; CKAP4 KO, 2.63±0.14). In addition, the expression levels of mitochondrial Ca²⁺ uniporter (MCU) and its regulators [the regulator of MCU (MICU1; also known as CBARA1) and the essential MCU regulator (EMRE; also known as SMDT1)] (Patron et al., 2014) were almost similar in control and CKAP4 KO cells (Fig. 5E). Thus, the increased [Ca²⁺]_m in CKAP4 KO cells could be caused by the direct influx of ER-released Ca²⁺ due to the increased number of ER–mitochondrial sites.

Ca²⁺ is required for metabolic regulation, but Ca²⁺ overload also causes the collapse of the mitochondrial membrane potential (Rizzuto et al., 2012). Indeed, when the mitochondrial membrane potential in CKAP4 KO cells was measured using JC-1 (Smiley et al., 1991), the intensity of JC-1 polymer (red) was reduced, whereas that of JC-1 monomer (green) was increased (Fig. 6A), indicating a decrease in the mitochondrial membrane potential. Experiments with MitoTracker Orange, whose intensity is dependent upon the mitochondrial membrane potential, and MitoTracker Green, whose intensity is independent of the mitochondrial membrane potential (Cottet-Rousselle et al., 2011), showed that the MitoTracker Orange intensity was decreased in CKAP4 KO cells under the condition that the MitoTracker Green intensity was the same, further confirming that CKAP4 KO reduces the mitochondrial membrane potential in HeLaS3 cells (Fig. 6B). A similar phenotype was also observed in other cell lines, including U2OS cells, MDCK dog kidney epithelial cells and Eph4 mouse mammary epithelial cells using MitoTracker Orange (Fig. S4C–E). As well as an increase in [Ca²⁺]_m, the phenotype observed in CKAP4 KO cells was also rescued by CKAP4^WT expression, VDAC2 KO, IP3R1 KD or IP3R3 KD (Fig. 6C,D). Thus, CKAP4 may inhibit the binding of IP3R and VDAC2, and thereby appropriately control [Ca²⁺]_m and maintain the mitochondrial membrane potential. CKAP4^C100S did not rescue the phenotypes induced by CKAP4 KO (Fig. 6C), suggesting that palmitoylation is necessary for the functions of CKAP4 at MAMs. It has been reported that the decrease in the mitochondrial membrane potential induces the cleavage of OPA1 and controls mitochondrial fission and fusion (Ishihara et al., 2006). When treated with a m-chlorophenylhydrazone (CCCP), a chemical inhibitor of oxidative phosphorylation, the ratio of long (L)-OPA1 to short (S)-OPA1 (cleavage of L-OPA1 into S-OPA1 impairs mitochondrial fusion; Ishihara et al., 2006) was unchanged in control and CKAP4 KO cells (Fig. 6E), suggesting that mitochondrial fragmentation induced by CKAP4 KO does not depend on OPA1.

VDAC is a master regulator of mitochondrial bioenergetics (Fang and Maldonado, 2018). When the bioenergetics profiles of control and CKAP4 KO HeLaS3 cells were analyzed, the baseline oxygen consumption rate (OCR), which reflects the mitochondrial function of healthy oxidative phosphorylation, and the maximal respiration were decreased in CKAP4 KO cells (Fig. 7A), even though the extracellular acidification rate (ECAR), which reflects the cellular aerobic glycolysis activity, was unchanged by CKAP4 KO (Fig. 7B). The decrease in the OCR in CKAP4 KO cells was rescued by the expression of CKAP4^WT but not by that of CKAP4^C100S (Fig. 7A). The expression levels of oxidative phosphorylation (OXPHOS)-related proteins were not altered in CKAP4 KO cells compared with control cells (Fig. 7C). Therefore, CKAP4 KO may affect the quality but not the quantity of mitochondrial proteins, resulting in decreased respiratory functions in mitochondria.

It is generally believed that cancer cells undergo metabolic reprogramming and mainly generate their ATP from aerobic glycolysis rather than oxidative phosphorylation (Ward and Thompson, 2012). We found that the proliferation of HeLaS3 cells was dependent on glucose concentration (Fig. S5A). When HeLaS3 cells were cultured in the presence of 1 mg/ml glucose (high-glucose conditions), CKAP4 KO did not affect cell proliferation (Fig. 8A). When glucose concentrations were decreased to 0.25 mg/ml (low-glucose conditions), CKAP4 KO suppressed cell proliferation and increased the number of propidium iodide (PI)-positive cells (dead cells), and these phenotypes were rescued by the expression of CKAP4^WT but not by that of CKAP4^C100S (Fig. 8B,C). However, dead cells did not exhibit apoptotic features, such as cleaved caspase-3 expression, in the presence of 1 mg/ml or 0.25 mg/ml glucose (Fig. S5B), although cytochrome c was detected in the cytosol of CKAP4 KO cells (Fig. S5C). In addition, CKAP4 KO did not affect the interaction of VDAC2 and the pro-apoptotic proteins Bak and Bax (Fig. S5D,E). Taken together, cell death induced by CKAP4 KO under low-glucose conditions is unlikely to be associated with apoptosis. Under the low-glucose conditions, the numbers of ER–mitochondria contact sites were increased, but the formation of a complex between GRP75 and VDAC2 and subcellular localization of CKAP4 were not changed in HeLaS3 cells (Fig. S5F–H), suggesting that the increase in ER–mitochondria contact sites by the low-glucose conditions is not caused by suppression of CKAP4 function.

In addition to HeLaS3 cells, U2OS cells also showed similar proliferation patterns, depending on glucose concentrations; in these cells, CKAP4 KO did not affect cell proliferation in the presence of 1 mg/ml glucose, but decreased it in the presence of 0.25 mg/ml glucose (Fig. S6A,B). In contrast, normal (i.e. not cancer) cell lines such as MDCK cells and Eph4 cells showed decreased cell proliferation upon CKAP4 KO, even under 1 mg/ml glucose conditions (Fig. S6C,D).

When HeLaS3 cells were subcutaneously injected into immunodeficient mice, the volumes and weights of the xenograft tumors derived from CKAP4 KO HeLaS3 cells were lower than those of control cells (Fig. 8D–F). These phenotypes were rescued by the expression of CKAP4^WT but not by that of CKAP4^C100S. In conclusion, palmitoylated CKAP4 could be required for in vitro cancer cell proliferation under low-glucose conditions and in vivo xenograft tumor formation.

CKAP4 is primarily localized to the ER of various cells and is also present in the plasma membrane of some types of cells, including cancer cells (Kikuchi et al., 2017; Vedrenne and Hauri, 2006). In this study, we found that CKAP4 localized to the ER binds to VDAC2 in mitochondria and identified new functions of CKAP4 at ER–mitochondria contact sites.

There are three isoforms of VDAC in humans, VDAC1, VDAC2 and VDAC3 (Shoshan-Barmatz and Gincel, 2003). The amino acid sequence of VDACs is highly conserved, with 80–90% overall similarity, indicating the high degree of 3D structural similarity between the three isoforms (Bayrhuber et al., 2008; Schredelseker et al., 2014). Our study revealed that CKAP4 preferentially binds to VDAC2, rather than VDAC1 or VDAC3. VDAC2 has an N-terminal amino acid sequence that is distinct from that of other VDACs. However, the N-terminal region of VDAC2 was not required for its binding to CKAP4. Thus, it is currently unclear which specific region of VDAC2 determines the binding to CKAP4.

CKAP4 KO enhanced the interaction of IP3R and VDAC2 and that of GRP75 and VDAC2. GRP75 acts as a scaffolding, rather than chaperoning, protein of the IP3R and VDAC complex at ER–mitochondria contact sites to increase the efficiency of mitochondrial Ca²⁺ uptake (Rizzuto et al., 2012; Szabadkai et al., 2006). When mitochondria are exposed to Ca²⁺ overload, opening of the mitochondrial permeability transition pore (mPTP) is triggered, leading to mitochondrial dysfunction (Bock and Tait, 2020; Martinou et al., 2000; NavaneethaKrishnan et al., 2020). Similar to IP3R, calnexin and VDAC2, CKAP4 is present in the MAM fraction. Our present results in CKAP4 KO cells demonstrate an increase in [Ca²⁺]_m, without affecting the expression of MCU and its regulators, increased numbers of ER–mitochondria contact sites, and a reduced mitochondrial membrane potential. These phenotypes are rescued through the depletion of VDAC2, IP3R1 or IP3R3, indicating functional interactions between CKAP4, VDAC2 and IP3Rs. Thus, CKAP4 may appropriately suppress the formation of ER–mitochondria contact sites and the mitochondrial functions of the GRP75–IP3R–VDAC2 complex by binding to VDAC2, resulting in inhibition of mitochondrial Ca²⁺ transport and maintenance of the membrane potential.

GRP75 is localized in extramitochondrial and intramitochondrial sites. Although we speculate that the interaction of extramitochondrial GRP75 and VDAC2 is increased in CKAP4 KO cells, intramitochondrial GRP75 may form a complex with VDAC2. It has recently been shown that ANT3 (SLC25A6), an inner mitochondrial membrane protein, interacts with GRP75 and VDAC in mitochondria (Wu et al., 2020). GRP75 depletion enhances the interaction of ANT3 and cyclophilin D and mPTP opening, suggesting that intramitochondrial GRP75 inhibits the formation of a complex between ANT3 and cyclophilin D and decreases mitochondrial permeability. However, our results revealed that cytochrome c is released into the cytoplasm of CKAP4 KO cells, which reflects mPTP opening. Thus, the increase in the amount of GRP75 co-immunoprecipitated with VDAC2 in CKAP4 KO cells could be derived from extramitochondrial GRP75.

Fragmented mitochondria were frequently observed in CKAP4 KO cells. Although the cleavage of L-OPA1 to S-OPA1, which impairs mitochondrial fusion (Ishihara et al., 2006), was not enhanced in CKAP4 KO cells, the numbers of ER–mitochondria contact sites were increased. It has been reported that ER–mitochondria contact sites coordinate mitochondrial DNA replication with mitochondrial division, and that CKAP4 overexpression increases sheet-like ER structures and reduces mitochondrial DNA synthesis (Lewis et al., 2016). Thus, it is intriguing to speculate that the increase in ER–mitochondria contact sites observed in CKAP4 KO might affect mitochondrial DNA synthesis and mitochondrial fission, which occurs independently of OPA1 cleavage. Taken together with the previous observation that OPA1 cleavage occurs during apoptosis (Arnoult et al., 2005) and our present result that apoptosis does not occur in CKAP4 KO cells, there may be a new mechanism of CKAP4-dependent control of mitochondrial morphology.

CKAP4 is modified with palmitate at Cys¹⁰⁰ in a reaction catalyzed by ZDHHC2 and ZDHHC5 (Planey et al., 2009; Sada et al., 2019). Palmitoylation of membrane proteins has been reported to be involved in the regulation of protein targeting, trafficking, protein–protein interactions and protein conformation (Charollais and Van Der Goot, 2009). CKAP4 is localized to the ER and plasma membrane, and palmitoylation is not required for the trafficking of CKAP4 from the ER to the plasma membrane (Sada et al., 2019). In plasma membrane-located CKAP4, palmitoylation is required for the localization of CKAP4 to lipid rafts, and DKK1 induces depalmitoylation of CKAP4, resulting in its moving to non-lipid rafts (Sada et al., 2019). Lipid rafts are cholesterol- and sphingolipid-rich plasma membrane microdomains that are thicker than other parts of the plasma membrane (Jacobson et al., 2007). Changes in transmembrane domain tilting caused by palmitoylation may control the effective length of hydrophobic CKAP4 segments, causing the protein to partition into lipid rafts.

In addition to local Ca²⁺ transfer from the ER to mitochondria, the lipid transport function of ER–mitochondria contact sites, such as MAMs, has been extensively studied (Hayashi et al., 2009). MAMs are enriched in cholesterol and neutral lipids (Hayashi and Su, 2003). However, in contrast to the plasma membrane lipid rafts, palmitoylation of CKAP4 is not required for localization to ER–mitochondria contact sites or its distribution in the ER membrane. Thus, the difference in the CKAP4 localization in the microdomains of the plasma membrane and ER may be due to differences in their lipid composition.

CKAP4^C100S neither binds to VDAC2 nor rescues the decrease in [Ca²⁺]_m and membrane potential observed in CKAP4 KO. The binding of CKAP4 to VDAC2 was decreased in ZDHHC2-depleted cells. Because palmitate covalently bound to CKAP4 is probably inserted into the ER membrane, it might not directly bind to VDAC2. CKAP4 (Δ101–106) is neither palmitoylated nor binds to VDAC2, even though the palmitoylation site (Cys¹⁰⁰) is intact. Thus, the amino acid sequence (aa 101–106) adjacent to the cell membrane is required for CKAP4 palmitoylation, and its structure, determined by palmitoylation at Cys¹⁰⁰, would be important for its direct or indirect interaction with VDAC2. At least seven amino acids, including the palmitoylation site of CKAP4, are necessary for the formation of the CKAP4 and VDAC2 complex. Taken together, the present study proposes that palmitoylated CKAP4 in the ER regulates mitochondrial functions via its interaction with VDAC2.

Consistent with the results on mitochondrial damage in CKAP4 KO cells, oxidative phosphorylation, but not glycolysis, was reduced in CKAP4 KO cells. Cancer cells, such as HeLaS3 cells and U2OS cells, exhibit increased glycolysis and reduced oxidative phosphorylation (Ward and Thompson, 2012). This is due to the enhanced glycolysis pathway resulting from an interplay between oncogenes and the tumor microenvironment (Zheng, 2012). CKAP4 KO did not affect the proliferation of HeLaS3 cells or U2OS cells when the glucose concentration was 1 mg/ml. However, when the glucose concentration was reduced to 0.25 mg/ml, the proliferation of CKAP4 KO cells was suppressed. Similar results were observed in U2OS cells, but not in MDCK or Eph4 cells. Therefore, cancer cell proliferation is supported by glycolysis when glucose is sufficient, and mitochondrial oxidative phosphorylation, which requires CKAP4, contributes to cancer cell proliferation under low-glucose conditions. These results are consistent with a recent report showing that glycolytic suppression induced mitochondria dependency in cancer cells (Shiratori et al., 2019). In contrast, because normal cells primarily undergo mitochondrial respiration, their proliferation is affected by CKAP4, even in the presence of high concentrations of glucose. In addition to cancer cell proliferation, cancer cell survival is also regulated by CKAP4 under low-glucose conditions. CKAP4 KO increases the number of PI-positive (dead) cells under low-glucose conditions. Cytochrome c, an apoptosis inducer, is released into the cytoplasm of CKAP4 KO cells, but CKAP4 KO neither increases cleaved caspase-3 nor affects the interaction of BAX or Bak with VDAC2. Thus, cell death is induced in a non-apoptotic manner by CKAP4 under low-glucose conditions and may result from mitochondrial dysfunction.

Compared with what was seen in the in vitro experiments, which depended on the glucose concentration, the effect of CKAP4 KO on cancer cell proliferation was clearly observed in vivo. This difference between the in vitro and in vivo experiments may be related to the tumor microenvironment. Inhibition of ER-located CKAP4 in cancer cells using, for example, an antisense oligonucleotide (ASO) against CKAP4 may suppress cancer cell proliferation in vivo. Although normal cells express ER-located CKAP4 in various organs, ASO tends to accumulate in tumor lesions rather than normal cells (Harada et al., 2019; Kimura et al., 2020). Thus, CKAP4 ASO may be a novel nucleic acid treatment for patients with tumors expressing high levels of CKAP4.

Anti-CKAP4 monoclonal antibodies (3F11-2B10 and 5A6-17A11) and anti-CKAP4 polyclonal antibody were generated as described (Kimura et al., 2016, 2019). Other primary antibodies used in this study are listed in Table S1. pLV-Mito-GFP (#44385), pCMV-CEPIA3mt (#58219) and pCAG cyto-RCaMP1h (#105014) were from AddGene (Cambridge, MA, USA). The primers for real-time PCR and the siRNA target sequences used in this study are described in Tables S2 and S3, respectively. Other materials were obtained from commercial sources.

HeLaS3 cervical cancer cells were provided by Kunihiro Matsumoto (Nagoya University, Aichi, Japan) in May 2002. MDCK type I (MDCK) dog renal tubule cells and Eph4 mouse mammary epithelial cells were provided by Sachiko Tsukita (Osaka University, Suita, Japan) in January 2013 and October 2011. S2-CP8 pancreatic cancer cells were purchased from Cell Resource Center for Biomedical Research, Institute of Development, Aging and Cancer, Tohoku University, in April 2014. Lenti-X 293T (X293T) cells were purchased from Takara Bio Inc. in October 2011. U2OS osteosarcoma cells were purchased from American Type Culture Collection (ATCC, Manassas, VA, USA) in February 2010. Initial cell lines were frozen in liquid nitrogen and early passages of cells (<1 month in culture) were used in all experiments (no authentication was done by the authors). Cells were checked for mycoplasma using the e-Myco plus Mycoplasma PCR Detection Kit. HeLaS3, MDCK, S2-CP8, X293T and U2OS cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS). For cell proliferation assays, glucose-depleted DMEM and dialyzed FBS by filtration (10,000 molecular weight cut-off) were used.

CKAP4 KO and VDAC2 KO cells were generated by the CRISPR/Cas9 system as previously described (Harada et al., 2017). The target sequences for human CKAP4 (5′-GGGTGGGCACCCTTCTCCGA-3′) and human VDAC2 (5′-TATGATGGAGGAATACACAT-3′) were designed with the help of CRISPR Genome Engineering Resources (https://zlab.bio/guide-design-resources). To generate cells stably expressing proteins for rescue experiments, parental cells were infected with lentivirus and selected with G418 and blasticidin.

Immunocytochemistry was performed as described previously (Matsumoto et al., 2019; Osugi et al., 2019). Cells were fixed with phosphate-buffered saline (PBS) containing 4% paraformaldehyde (PFA) and then permeabilized in PBS containing 0.2% (w/v) Triton X-100 and 2 mg/ml bovine serum albumin (BSA) for 10 min. The cells were blocked in blocking buffer (PBS containing 2 mg/ml BSA). Samples were incubated with the primary antibodies diluted in PBS for 1 h or overnight, washed three times with PBS, and then stained with a secondary antibody (conjugated to Alexa Fluor 488 or 546; Invitrogen) diluted in PBS for 1 h. After washing, the samples were covered with PBS containing 50% glycerol. Samples were viewed and analyzed using an LSM810 laser scanning microscope.

The number per cell, mean size and total area of mitochondria were measured in ImageJ by using the Mito Morphology Macro plugin (Dagda et al., 2009). Z-stack confocal microscopy images of Mito-GFP were acquired using an LSM810 laser scanning microscope and the images were imported into ImageJ, where the program was used to set a common threshold and calculate the mitochondrial parameters.

Conventional electron microscopy was performed as previously described (Hayashi-Nishino et al., 2009). Briefly, cells were fixed in 2.5% glutaraldehyde in 0.1 M PBS at pH 7.4 for 1 h. The specimens were postfixed in buffer containing 1% OsO₄ and 1.5% potassium ferrocyanide, dehydrated in a series of graded ethanol solutions, and embedded in epoxy resin. Ultra-thin sections were collected and stained with uranyl acetate and lead citrate and observed under a JEM-2100 transmission electron microscope (JEOL, Tokyo, Japan) at an accelerating voltage of 80 kV.

Normal or abnormal cristae morphology of mitochondria was defined as previously described (Deng et al., 2015). Briefly, normal mitochondria with numerous well-organized cristae or abnormal mitochondria showing round mitochondria containing unstructured cristae were scored and quantified as the percentage of total mitochondria examined (n=160–170).

To quantify the amount of ER–mitochondria contact sites, the percentage of the mitochondrial surface closely apposed to the ER was calculated (<25 nm distance between membranes, n=25–30) (Demetriadou et al., 2017; Hirabayashi et al., 2017).

The split-GFP system, which can detect ER–mitochondria contact sites, was described previously (Kakimoto et al., 2018). HeLa cells stably expressing TOM70 (1–70)–FLAG–GFP11 in mitochondria and doxycycline-dependently expressing ERj1 (1–200)–V5–GFP1-10 in the ER were treated with 30 ng/ml doxycycline for 24 h to induce the expression of ERj1 (1–200)–V5–GFP1-10. Split-GFP signals were viewed and analyzed using an LSM810 laser scanning microscope.

Liquid chromatography-tandem mass spectrometry analysis for identification of CKAP4-binding proteins was performed as described previously (Osugi et al., 2019) using an UltiMate 3000 nano LC system (Thermo Fisher Scientific, Waltham, MA) coupled to a Q-Exactive hybrid quadrupole-Orbitrap mass spectrometer (Thermo Fisher Scientific) with a nanoelectrospray ionization source.

The immunoprecipitation assay was performed as described previously (Matsumoto et al., 2019; Osugi et al., 2019). Cells (60-mm diameter dish) were lysed in 500 µl of Nonidet P-40 (NP-40) buffer (20 mM Tris-HCl pH 8.0, 10% glycerol, 137 mM NaCl, and 1% NP-40) with protease inhibitors (10 µg/ml leupeptin, 20 µg/ml aprotinin and 1 mM phenylmethanesulfonyl fluoride). After centrifugation, the lysates were incubated with primary antibodies and 40 µl of a 50% slurry of protein G–Sepharose beads (GE Healthcare Bio-Sciences, Buckinghamshire, UK) for 1 h at 4°C. After washing three times with NP-40 buffer, the precipitates were probed with the indicated antibodies.

The acyl-PEGyl exchange gel shift (APEGS) assay was performed as previously described (Sada et al., 2019). X293T CKAP4 KO cells (60-mm diameter dish) expressing CKAP4^WT–HA, CKAP4^C100S–HA, CKAP4 (Δ24-99)–HA or CKAP4 (Δ2-21)–HA were then lysed in 1 ml of PBS containing 5 mM EDTA, 4% SDS and protease inhibitors. After sonication and centrifugation at 20,000 g for 15 min at room temperature (RT), the soluble proteins (0.4 to 0.5 mg/ml, 1 ml) were reduced with 25 mM Tris (2-carboxyethyl) phosphine (TCEP) for 1 h at 55°C, and free cysteine residues were alkylated with 50 mM N-ethylmaleimide for 3 h at RT. After chloroform and methanol precipitation (CM ppt), the precipitates were suspended in 250 µl of PBS containing 5 mM EDTA, 4% SDS, and 10 µg/ml pepstatin A and then centrifuged at 20,000 g for 10 min at RT to completely remove the undissolved protein pellet. The supernatant (125 µl) was mixed with 375 µl of either 1.33 M hydroxylamine (pH 7.0), 0.2% Triton X-100, and 5 mM EDTA or 1.33 M Tris-HCl (pH 7.0), 0.2% Triton X-100, and 5 mM EDTA, and the mixtures were incubated for 1 h at 37°C. After CM ppt, the precipitates were resuspended in 100 µl of PBS containing 5 mM EDTA, 4% SDS and 10 µg/ml pepstatin A. Soluble proteins (0.5 to 0.75 mg/ml, 100 µl) were PEGylated with 20 mM mPEG-5k in the presence of 10 mM TCEP for 1 h so that newly exposed cysteinyl thiol groups could be labeled with mPEG-5k, which causes a mobility shift of palmitoylated proteins in SDS-PAGE gels. After CM ppt, the precipitates were resuspended in Laemmli's SDS-sample buffer and boiled at 100°C for 5 min. Western blotting was then used to detect palmitoylated bands and non-palmitoylated bands. Protein concentrations were measured with the bicinchoninic acid protein assay at individual steps.

Microsomes, mitochondria and MAMs from HeLaS3 cells were isolated following previously described protocols (Wieckowski et al., 2009). HeLaS3 cells were harvested from 10 dishes (100-mm-diameter dishes) at 90–100% confluence and homogenized in isolation buffer (30 mM Tris-HCl pH 7.4, 225 mM mannitol, 75 mM sucrose and 0.1 mM EGTA) using a Teflon pestle until 80–90% of the cells were disrupted. Then, the homogenate was centrifuged twice at 600 g for 10 min each to remove nuclei and debris. The supernatant was collected and centrifuged for 15 min at 7000 g to obtain crude mitochondria. For ER isolation, this step was repeated with the supernatant until a pellet was no longer visible, and the supernatant was centrifuged at 20,000 g for 30 min. Next, the mitochondria-free supernatant was centrifuged again at 100,000 g for 1 h. The pellet was then resuspended for the ER fraction and the supernatant was kept for the cytosolic fraction. For the pure mitochondria and MAM fraction, the crude mitochondria pellet was resuspended in a 5× volume of resuspension buffer (5 mM HEPES-NaOH pH 7.4, 250 mM mannitol and 0.5 mM EGTA), and the fraction was added to the top of 30% Percoll medium [25 mM HEPES-NaOH pH 7.4, 225 mM mannitol, 1 mM EGTA and 30% Percoll (v/v)] in a centrifuge tube. Centrifugation was performed at 95,000 g for 30 min, and then the upper (MAMs) and lower (pure mitochondria) layers were collected with a Pasteur pipette. Both fractions were diluted with a 10× volume resuspension buffer and centrifuged at 6300 g for 15 min. Then, the pure mitochondria fraction was obtained from the pellet after one more wash under the same conditions. The supernatant containing the MAMs was centrifuged at 100,000 g for 1 h, and the resulting pellet was resuspended in resuspension buffer. Each fraction was evaluated by western blotting using mouse anti-CKAP4 (3F11-2B10), rabbit anti-calnexin (Sigma), mouse anti-IP3R (BD Bioscience), rabbit anti-VDAC2 (Invitrogen) and mouse anti-Hsp90 (BD Bioscience) antibodies.

Immunoelectron microscopy was performed following the previously described protocols (Sobajima et al., 2018). Control and CKAP4 KO HeLaS3 cells were fixed with 3% PFA and 0.1% glutaraldehyde in PBS, pH 7.4, for 30 min at RT, washed with PBS, and permeabilized in 5% normal goat serum and 0.25% saponin in PBS. After quenching with 1 mg/ml NaBH₄ in the same solution for 30 min at room temperature, the cells were washed with PBS and treated with anti-CKAP4 monoclonal antibody (5A6-17A11) for 2 h at 37°C and then with Alexa Fluor 488-labeled FluoroNanogold anti-mouse IgG (1:30; Nanoprobes, Yaphank, NY, USA) in 5% normal donkey serum and 0.25% saponin in PBS for 1 h at RT. The cells were washed with PBS and then with distilled water. The fixed cells were incubated with HQ silver enhancement solution (Nanoprobes) for 7 min at RT, extensively washed with distilled water, and incubated with selenium toner (1:20; Kodak) for 7 min at RT to prevent erosion of the silver during the OsO₄ fixation. Samples were postfixed in 2.5% glutaraldehyde in PBS for 15 min and in 1% OsO₄ for 1 h on ice, followed by staining in 4% uranyl acetate for 2 h at RT. The samples were dehydrated and embedded as previously described. They were examined and photographed with a JEOL-1010 electron microscope.

PLA was performed according to the manufacturer's protocol (Sigma) as described previously (Ibuka et al., 2015; Osugi et al., 2019). Cells transiently expressing FLAG–VDAC2 grown on glass coverslips were fixed for 15 min at RT in PBS containing 4% PFA. The cells were permeabilized in PBS containing 0.2% (w/v) Triton X-100 and 2 mg/ml BSA for 10 min. The glass coverslips were blocked in blocking buffer for 30 min and incubated with anti-FLAG and anti-IP3R antibodies diluted in blocking buffer for 1 h at RT. After washing, the coverslips were incubated with Duolink PLA anti-rabbit minus and PLA anti-mouse plus proximity probes (Sigma). PLA dots were counted using an LSM810 laser scanning microscope.

To measure intramitochondrial Ca²⁺ concentrations, cells were plated on glass-bottom dishes (Iwaki) and transiently transfected with CEPIA3mt (Suzuki et al., 2014) and mCherry using FuGENE HD (Promega) and incubated for 1 day. The medium was replaced with DMEM supplemented with 2.5 mM HEPES (pH 7.4). The fluorescence intensities of CEPIA3 mt and mCherry were measured using a laser scanning microscope (LSM810 or FV1000) and the intensity of the CEPIA3mt signal was normalized to that of mCherry signal. To measure cytoplasmic Ca²⁺ concentrations, cells were plated on glass-bottom dishes (Iwaki) and transiently transfected with RCaMP1h (Hirabayashi et al., 2017) and EGFP using FuGENE HD (Promega) and incubated for 1 day. The medium was replaced with DMEM supplemented with 2.5 mM HEPES (pH 7.4). The fluorescence intensities of RCaMP1 h and EGFP were measured using a laser scanning microscope (LSM810 or FV1000). To monitor the mitochondrial membrane potential, the cells were incubated with 2 µM JC-1 (Dojindo), 200 nM MitoTracker Orange CMXRos (Invitrogen), 200 nM MitoTracker Green FM (Invitrogen) or 200 nM MitoTracker Deep Red FM (Invitrogen) for 30 min and then imaged using an LSM810 laser scanning microscope.

Respirometry of intact HeLaS3 cells was performed using an XF96 Extracellular Flux Analyzer (Seahorse Biosciences) (VanLinden et al., 2015). Cells were seeded at a density of 4.0×10⁴ cells/well in 96-well XF microplates and cultured with DMEM containing 10% FBS for 24 h. At 1 h before measurements, the medium was replaced with XF base medium supplemented with 2 mM L-glutamate and 2 mM pyruvate. After 1-h incubation in a CO₂-free incubator at 37°C for temperature and pH equilibration, the baseline oxygen consumption rate (OCR) was measured. This was followed by sequential injections with 2 μg/ml oligomycin to measure the ATP-linked OCR, 1 μM FCCP (an oxidative phosphorylation uncoupler) to determine maximal respiration, and 0.1 μM rotenone and 0.1 μM antimycin A to determine non-mitochondrial respiration. For cell glycolysis measurement, HeLaS3 cells seeded in XF96 microplates as described above were cultured in XFbase medium supplemented with 2 mM L-glutamine in a CO₂-free incubator for 1 h. For extracellular acidification rate (ECAR) measurement, the cells were treated sequentially with 10 mM (1.8 mg/ml) glucose to measure glucose metabolism, oligomycin (2 μg/ml) to measure glycolytic capacity, and 2-deoxy-D-glucose (2-DG) (100 mM) to measure non-glycolytic acidification.

Cytochrome c release assays were performed as described previously (Waterhouse et al., 2004). Cells were homogenized in plasma membrane permeabilization buffer (PBS containing 200 µg/ml digitonin and 80 mM KCl) and incubated on ice for 5 min. Then, the homogenate was centrifuged at 8000 g for 5 min, and the supernatant (the cytosol-enriched digitonin-soluble fraction) was probed with anti-cytochrome c antibody (Santa Cruz Biotechnology).

HeLaS3 cell (5×10⁶ cells) pellet was suspended in 100 μl of PBS and sububcutaneous injected into the back of anesthetized 8-week-old male BALB/cAJcl-nu/nu mice (nude mice; CLEA Japan, Tokyo, Japan). The mice were euthanized 28 days after transplantation. The areas containing transplanted cells were measured and weighed. Tumor volumes were calculated using the following formula: (major axis)×(minor axis)×(minor axis)×0.5. All animal experiments were performed according to guidelines approved by the Animal Research Committee of Osaka University, Japan (No. 21-048-1).

Each experiment was performed at least three times, and the results are presented as the mean±s.d. A Mann–Whitney U-test was used to determine the statistical significance between the means of two groups. Analysis of variance (ANOVA) with Tukey, Bonferroni or Dunnett's post hoc tests was used to compare three or more group means. Statistical analysis was performed using Excel 2010 (Microsoft, Redmond, WA, USA). The Wilcoxon rank sum test was used for analysis of xenograft tumor weight (Fig. 8F). P values <0.05 were considered statistically significant.

We are grateful to Drs N. Ishihara, Y. Shintani and H. Kato for helpful discussion about mitochondrial functions, and Drs Y. Tamura and S. Tashiro for donating cells. We also thank Grants-in-Aid for Scientific Research on Innovative Area ‘Organelle zone’ for conventional electron microscopy and immunoelectron microscopy. We are grateful to the technical staff, especially I. Nishimura, of the Comprehensive Analysis Center of the Institute of Scientific and Industrial Research at Osaka University. We would like to thank the Center of Medical Research and Education, Graduate School of Medicine, Osaka University for in-gel digestion and LC-MS/MS analysis.

The peer review history is available online at https://jcs.biologists.org/lookup/doi/10.1242/jcs.249045.reviewer-comments.pdf