HeLaM, a subline of HeLa, and HeLaM or MCF7 cell lines stably expressing mRFP-histone H2A were cultured as previously described (Wang et al., 2014). To block cells in G1/S, cells were treated with 2.5 mM thymidine (Sigma-Aldrich) for 24 h. To block cells in prometaphase, cells were treated with 2.5 mM thymidine for 16 h, washed and treated with nocodazole at 0.2 µM (60 ng/mL) for 12 h. OTS167 was a gift from Drs. Yusuke Nakamura, Takuya Tsunoda and Yo Matsuo at Onco Therapy Science and was used at 100 nM (Chung et al., 2012). The proteasome inhibitor MG132 and the CDK1 inhibitor RO-3306 were used at 20 μM and 5 μM final concentrations, respectively. Additional information of these inhibitors and other kinase inhibitors are summarized in Supplementary Table S1.

MELK antibody and immunoblotting procedure were previously described (Ji et al., 2016). To prepare CDK1 inhibitors treated cell lysates used in Figure 4A, HeLa cells were treated with nocodazole at 0.2 µM (60 ng/mL) for 12 h, then RO-3306 or Roscovitine were added to final concentrations of 5 µM together with MG132 (20 µM) for 1 h. Mitotic cells were harvested by shake-off. For immunofluorescence, HeLaM cells were seeded on coverslips, treated with 2.5 mM thymidine for 24 h, washed then directly released into drug-free medium. After 9–10 h when cells were observed to enter mitosis, coverslips were fixed with ice-cold methanol for 20 min at −20°C. The fixed cells were gently washed three times in PBS, blocked in PBS containing 5% BSA for 1 h, and left in MELK primary antibody diluted in the blocking buffer overnight. The AlexaFluor 488 conjugated secondary antibody was used at 1:1,000 and incubated for 30 min. The coverslip was mounted using Fluoroshield containing DAPI (Sigma). The images were collected on a Leica TCS SP8 confocal microscope with a ×63 objective (numerical aperture = 1.40).

Human MELK cDNA was cloned into pENTR-TOPO vector (Invitrogen) and point mutations were made following the QuikChange site-directed mutagenesis protocol (Agilent). The primers are listed in Supplementary Table S2. The mutants include D150A, T167A, T167E, ⁵⁵⁵RRLK⁵⁵⁸ to SSSS, ⁶²⁷RRQR⁶³⁰ to SSSS, T446E, and the 5A or 5E mutants (T460, T466, T478, S498, T518 to A or E). All mutants were verified by Sanger sequencing (Genewiz). The wild type and mutants are recombined into an eGFP vector using Gateway LR clonase (Invitrogen). The mCherry-Lifeact-7 was a gift from Michael Davidson (Addgene plasmid # 54491). DNA transfection was carried out using polyethylenimine as described (Ji et al., 2016).

For live cell imaging, cells were plated on 35 mm dish with a coverslip glued bottom (Cellvis) at about 30% confluence and transfected with the desired constructs the next day. Images were captured ∼24 h later on a Leica SP8 confocal microscope usually with 3 min intervals in 2 μm Z stacks spanning the cell dimension. The imaging DMEM medium contains 20 mM HEPES (pH 7.4) but no phenol red, and cells were maintained in an on-stage heating chamber set at 37°C. To quantify cell cortex localized GFP-MELK, the cells were stained with CellBrite Steady 650 Membrane Staining dye (Biotium) to delimit cell membrane. The middle plane of image stacks was selected for quantification. Images were analyzed using the membrane stain to create a mask and apply to the GFP channel. To measure GFP intensity at the cortex in experiments without membrane dye staining as shown in Figures 4, 5, a 1 µm width line was drawn around the cell edge in ImageJ (Schneider et al., 2012). After subtracting background, GFP intensity measured along the line was considered as cortex-localized signals, and the internal circle was measured as cytoplasm. The % of GFP intensity at the cortex was compared to total GFP intensity (cortex + cytoplasm) in the cell.

Statistical analysis was performed using GraphPad Prism software (version 10.1.0). Data is presented as mean ± SD. The student’s t-test was used to assess the significance of differences between two samples. For multiple samples, 2-way ANOVA with Tukey’s multi comparison test was used (for Figures 4, 5 experiments).

Previously it was found that MELK re-localizes to the cell cortex in anaphase and telophase Xenopus or HeLa cells (Chartrain et al., 2006; Le Page et al., 2011; Tipton et al., 2012). To gain more insights into the translocation, live cell imaging was used to track GFP-MELK localization in a HeLa cell line stably expressing mRFP-tagged histone H2A (Figures 1A–C). GFP-MELK primarily resides in the cytoplasm until metaphase but translocates to the cell cortex within ∼3 min after the metaphase-to-anaphase transition and remains associated until late telophase. The cortex localization for MELK lasts 33 ± 8 min (mean ± SD, n = 11) during anaphase and telophase (Figure 1C). Endogenous MELK was also found to localize at cell cortex in anaphase cells by immunofluorescence (Figure 1D), similarly as reported before (Chartrain et al., 2006). Translocation of MELK to cell cortex during anaphase cells has also been observed in other cell lines including MCF-7 and hTERT-RPE1, suggesting it is a common feature for MELK regulation (Supplementary Figure S1). We checked the GFP-MELK localization in interphase especially in early G1 and late G2 phases and found GFP signals were enriched in the cytoplasm with no clear cell cortex localization (Supplementary Figure S2A).

Using a lipid binding fluorescent dye to delimit the cell membrane, the GFP intensity of MELK at the cortex was quantified (Figure 1E). The GFP intensity localized at the cell cortex increased from 10% ± 2% (n = 12 cells) of the total GFP intensity in metaphase to 50% ± 8% (n = 12 cells) in anaphase. Line scans drawn across the cortex in metaphase and anaphase cells stained with the lipid binding dye supported MELK localization to the membrane during anaphase (Supplementary Figure S2B). Actin cytoskeleton is a major component of cell cortex (Kunda and Baum, 2009). Live cell fluorescence microscopy was performed in cells transfected with Lifeact that binds to filamentous actin (F-actin) (Riedl et al., 2008; Belyy et al., 2020). The temporal control of MELK was indicated by GFP-MELK co-localization with LifeAct signals in anaphase but not metaphase cells (Supplementary Figure S3). This is consistent with the earlier result that cortical MELK co-localized with filamentous actin in fixed samples (Chartrain et al., 2006). The above results supported that MELK translocates from the cytoplasm to cell cortex within ∼3 min of anaphase onset.

The temporally restricted cortex localization of MELK during late mitosis suggested that MELK localization is under control of mitotic kinases. Since the translocation occurs after the anaphase onset, it is hypothesized that the drop in CDK1 kinase activity regulates MELK translocation. To test the hypothesis, HeLa cells stably expressing mRFP-histone H2A were transfected with GFP-MELK, and treated with nocodazole and MG132 to arrest cells in prometaphase. When exposed to CDK1 inhibitor RO-3306 (Vassilev et al., 2006), GFP-MELK localized to the cortex within 2–3 min and became stable at the cortex onwards, even though the cells stayed in prometaphase-like state based on chromosome configuration (Figures 2A,B). Quantitation found that 48% ± 12% of GFP signals (n = 12 cells) got enriched at the cell membrane after 3 min of RO-3306 treatment (Figure 2B). Similar results were observed using another CDK1 inhibitor roscovitine (Meijer et al., 1997) or using MCF-7- mRFP H2A cell line (Figures 2C,D). This showed that reduced CDK1 activity triggers MELK localization onto the cortex.

We have shown before that inhibiting mitotic kinases Plk1, Aurora B or MPS1 did not prematurely target MELK to the cortex (Ji et al., 2016). We expanded the test and found inhibiting Src/Abl kinases (PD166326), Aurora A kinase (MLN8237), p38 MAPK (SB202190), MEK1 (PD98059), JNK (JNK-IN-8), and MEK1/2 (U0126) also did not affect the timing of cortex association of GFP-MELK (Supplementary Figure S4).

We confirmed that OTS167 exposure also prematurely targeted MELK to the cortex (Ji et al., 2016) (Supplementary Figure S4, third row, part I). Although OTS167 was suggested to be a MELK specific inhibitor, later results found it promiscuously inhibits many other kinases (Ji et al., 2016; Huang et al., 2017; Klaeger et al., 2017; Giuliano et al., 2018). Applying two more specific MELK inhibitors, HTH-01-091 and NVS-MELK8a, revealed no premature localization of GFP-MELK in the same assay (Supplementary Figure S4, bottom row) (Toure et al., 2016; Huang et al., 2017). In addition, two kinase dead mutants of MELK (D150A or N137A), when fused with GFP, displayed similar translocation kinetics as wild type MELK (Supplementary Figure S5). Other MELK kinase mutants such as T167A and T167E mutants at the key T167 residue in the activation loop (Cao et al., 2013) did not alter the localization pattern either (Supplementary Figure S5). These results suggested that MELK kinase activity is likely not required for its own localization pattern.

To further understand the cortex localization of MELK, we next shifted attention to its kinase associated 1 (KA1) domain. KA1 domains exist primarily in kinases in the Kin1/PAR-1/MARK family, but also in other kinases such as Chk1 and RNA processing enzymes (Moravcevic et al., 2010; Gong et al., 2018; Paung and Seeliger, 2018; Aoyama et al., 2020; Ju et al., 2023). The KA1 domains in the MARK family show plasma membrane localization in S. cerevisiae, S. pombe and mammalian cells, and MARK1-KA1 binds to acidic phospholipids in cells and in vitro (Moravcevic et al., 2010; Rincon et al., 2014; Emptage et al., 2017a).

Alphafold predicted that human MELK KA1 has similar fold as other solved KA1 structures (Meng et al., 2023) (Figure 3A). The MELK-KA1 domain (550–651 residues) has a theoretical isoelectric point: pI = 9.330 and at physiological pH 7.4: z = +5.886. GFP-MELK-KA1 is indeed localized to the cortex throughout the cell cycle (Supplementary Figure S6), in contrast to full length MELK. Live cell imaging of GFP-KA1 domain confirmed persistent localization at cortex form metaphase to the end of cytokinesis (Figure 3B).

To further investigate the requirement of KA1 domain for MELK association with the cell cortex, a KA1-truncated MELK was fused with GFP and transfected into HeLa cells. As shown in Figure 3C, no translocation was observed even during anaphase. Two conserved basic patches in the KA1 domains of the MARK family kinases were found to be essential for membrane association through directly binding to phospholipids (Moravcevic et al., 2010; Emptage et al., 2017a; Emptage et al., 2017b). When corresponding patches in human MELK (⁵⁵⁵RRLK⁵⁵⁸ and ⁶²⁷RRQR⁶³⁰) were mutated, the resulted GFP-MELK-KA1 double mutant (“DM”) also failed to re-locate to the cortex in anaphase cells (Figure 3C). The same DNA constructs showed the same localization patterns despite differential expression levels in individual cells (Supplementary Figure S7). These results supported the idea that the KA1 domain particularly its two conserved basic patches provide the physical foundation of MELK association with the cell cortex.

We then hypothesized that CDK1 activity temporally controls MELK localization through phosphorylating MELK and preventing its KA1 binding to cell membrane during prometaphase. Human MELK contains a so-called Threonine-Proline (TP) rich unstructured region (335–550 amino acids) between its kinase-UBA domain and KA1 domain, whose TP rich composition is not shared with other MARK family kinases (Vulsteke et al., 2004) (Figure 3A). The MELK-TP region contains 10 TP and 1 SP sites with five of them conforming the CDK1 substrate consensus motif S/T-P-x-K/R (Songyang et al., 1994). Phosphoproteomics studies have confirmed in vivo phosphorylation of several sites in the TP region and several sites were proposed to be directly phosphorylated by CDK1 (Badouel et al., 2006; Hornbeck et al., 2015). Western blot of cell lysates prepared from nocodazole and MG132 arrested HeLa cells showed reduced MELK mobility shift after treatment with RO-3306 or Roscovitine, two CDK1 inhibitors, supporting CDK1 phosphorylation of MELK (Figure 4A). Although some earlier reports indicated that RO-3306 could reduce CDK1 after long time treatment, under our experimental conditions, RO-3306 did not affect CDK1 level but inhibit its kinase activity (Supplementary Figure S8). The serine/threonine residues in the five conserved (S/T)P sites within the TP region (⁴⁶⁰TPNR⁴⁶³, ⁴⁶⁶TPSK⁴⁶⁹, ⁴⁷⁸TPIK⁴⁸¹, ⁴⁹⁸SPER⁵⁰¹ and ⁵¹⁸TPKR⁵²¹) were mutated to alanines to create a phosphoresistant 5A mutant in otherwise full length MELK. Six out of 7 cells transfected with GFP-MELK-5A prematurely localized GFP to cell cortex in prometaphase cells, in contrast to GFP-MELK wild type transfected cells (Figure 4B), indicating the importance of phosphorylation status of the five (S/T)P sites.

To further probe the effect of TP phosphorylation, we co-transfected HeLa cells with GFP-KA1 domain and mCherry fused wild type TP fragment or TP-5A (phosphoresistant) or TP-5E (phosphomimetic) mutants and arrested cells in prometaphase by treatment with nocodazole and MG132. As shown in representative images in Figure 4C, ∼20% of GFP-KA1 is localized to cell cortex when co-transfected with mCherry vector in prometaphase cells. The cortex fraction of GFP-KA1 drops to 10% when mCherry-TP was co-expressed. Interestingly when mCherry-TP-5A was expressed, accumulation of GFP-KA1 at the cell cortex was observed again. Conversely, when mCherry-TP-5E was expressed, GFP-KA1 localization to the cortex was comparable as in mCherry-TP co-expressed cells (Figures 4D, E). Despite variations of mCherry and GFP expression levels, only TP-5A co-expressed cells but not TP or TP-5E expressed cells showed detectable cell cortex accumulation of GFP-KA1 signals. The series of experiments are consistent with the idea that phosphorylation in the MELK TP region by CDK1 could affect the interactions between TP and the KA1 domain, hence affecting KA1 availability for phospholipid binding which further controls the timing of MELK cortex localization.

Many mitotic phosphoproteins are dephosphorylated by protein phosphatases as cells exit from mitosis (Holder et al., 2019; Nilsson, 2019). Recently Ueki et al. identified two overlapping potential binding motifs (FXXP) on MELK for PP4 protein phosphatase: FMFP and FPEP within ⁴³⁹FMFPEP⁴⁴⁴ in the MELK TP region (Ueki et al., 2019) (Supplementary Figure S9A). We wonder whether PP4 dephosphorylates MELK TP region and triggers MELK cortex localization. To test the idea, we mutated FMFP and FPEP to AMFA and APEA respectively in GFP-MELK. When transfected into HeLa-mRFP-H2A cells, the APEA mutant showed 32% ± 8% (n = 9) cortex localized GFP, similarly to 41% ± 7% (n = 7) in wild type MELK transfected anaphase cells (Figure 5; Supplementary Figure S10). However, the AMFA mutant only had 10% ± 3% (n = 5) GFP translocated to the anaphase cortex (Figure 5; Supplementary Figure S10). This indicated that the FMFP motif is required for MELK anaphase cortex translocation, probably through recruiting PP4. PP4 binding to the FXXP motif can be negatively impacted by adjacent phosphorylation (Ueki et al., 2019). T446 is an in vivo phosphorylation site adjacent to the FMFP motif (Hornbeck et al., 2015). We therefore tested the T446E mutant, and found the phosphomimetic mutant also reduced anaphase cortex GFP signals, although to a lesser degree, to 26 ± 10 (n = 11) (Figure 5; Supplementary Figure S10). We tested but did not find significant PP4 catalytic subunit in the MELK immunoprecipitates using either prometaphase or anaphase cell lysates (Supplementary Figure S9B). These results indicate that PP4 is a strong candidate phosphatase to counter CDK1 phosphorylation of MELK TP region, hence contributing to timing the MELK anaphase cortex translocation. However, the PP4-MELK interaction might be transient.