Skip to main content
NCBI home page
Molecular Biology of the Cell logoLink to Molecular Biology of the Cell
. 2026 Sep 14;37(10):ar91. doi: 10.1091/mbc.E26-03-0120

Quantitative comparison of PI(3,5)P2 biosensors reveals SnxA is the most sensitive and unbiased

Tiernan Swayhoover a, Claire C Weckerly a,*, Gerald R V Hammond a,*
Editor: Laura L Lacknerb
PMCID: PMC13637729  PMID: 42555823

Abstract

PI(3,5)P2 is an endosomal lipid whose depletion is associated with a variety of pathologies such as neurodegenerative diseases. However, studying this lipid in physiological and disease models has been difficult due to the scarcity of the lipid and the lack of live-cell imaging tools. That is, until recently, when a novel PI(3,5)P2 biosensor, SnxA, was characterized. Despite the exciting promise of this new sensor, it was still unclear whether SnxA reported on PI(3,5)P2 levels without bias and how its sensitivity compared with other PI(3,5)P2 biosensors. In this work, we addressed these gaps by using a recruitable PIKfyve construct to demonstrate that ectopically generated PI(3,5)P2 at mitochondria was sufficient to recruit SnxA. Further, we co-expressed putative PI(3,5)P2 biosensors to definitively show that SnxA is more sensitive to PI(3,5)P2. We also validated previous results by showing that SnxA depends on PI(3,5)P2 for membrane binding, that SnxA responds to PI(3,5)P2 production at endosomes, and that PI(3,5)P2 levels decline quickly when its production is inhibited. Thus, we conclude that SnxA is a robust and sensitive PI(3,5)P2 biosensor that facilitates real-time analysis of this key lipid.

  • Phosphatidylinositol 3,5-bisphosphate (PI(3,5)P₂) is a low-abundance but essential signaling lipid that regulates endosomal physiology and is implicated in multiple neurodegenerative disorders.

  • Despite its importance, mechanistic study of PI(3,5)P₂ has been limited by a lack of reliable live-cell probes.

  • In this work, we perform a systematic and quantitative comparison of PI(3,5)P₂ biosensors and provide independent validation of the recently described SnxA probe, crucially demonstrating the sufficiency of PI(3,5)P₂ for probe localization for the first time, along with a quantitative benchmarking of current sensors.

INTRODUCTION

Phosphatidylinositol 3,5-bisphosphate (PI(3,5)P2) is a low-abundance, endosomal lipid first discovered as a metabolite upregulated in yeast exposed to hyperosmotic stress (Dove et al., 1997). PI(3,5)P2 is produced by a PIP 5-kinase, PIKfyve in mammals or Fab1 in yeast, that phosphorylates phosphatidylinositol 3-phosphate (PI3P) at the 5-OH position (Cooke et al., 1998; Sbrissa et al., 1999). Deletion of this kinase or use of inhibitors such as apilimod results in a phenotype where yeast vacuoles or mammalian endosomal compartments become enlarged; a process called vacuolation (Gary et al., 1998; Ikonomov et al., 2001, 2002; Rutherford et al., 2006; Jefferies et al., 2008; Cai et al., 2013).

PIKfyve exists within a regulatory complex consisting of a homopentamer of the scaffold protein Vac14, one copy of PIKfyve, and one copy of the 5′-phosphatase Fig4 Figure 4, which reverts PI(3,5)P2 to PI3P (Lees et al., 2020). Deletion of PIKfyve is embryonic lethal in mice (Ikonomov et al., 2011) and mutations within the components of the PIKfyve complex have been implicated in neurodegenerative diseases such as Charcot-Marie-Tooth syndrome or amyotrophic lateral sclerosis (ALS; Chow et al., 2007; Chow et al., 2009; Rivero-Ríos and Weisman, 2022). Although PIKfyve is a druggable therapeutic target (Ikonomov et al., 2019; Burke et al., 2022), the specific role of PI(3,5)P2 in these pathologies is still unclear, partly due to a lack of clarity in PI(3,5)P2's subcellular localization.

FIGURE 4:

FIGURE 4:

Open in a new tab

Design of a recruitable PIKfyve construct. (A) The AlphaFold3 predicted structure of the FKBP-PIKfyve-CCR-Kina construct. Both the conserved Cys-rich region (CCR) and the kinase domain, made up of the N-lobe, C-lobe, and linker, were included to facilitate proper domain folding. The predicted aligned error (PAE) is shown to demonstrate high confidence in the predicted structures of the defined domains. The predicted template modeling (pTM) score is given. Scores above 0.5 indicate confidence that the predicted structure aligns with the true structure. (B) Schematic showing how FKBP-PIKfyve-CCR-Kina activity was monitored by EEA1-FYVE and SnxA biosensors after the PIKfyve construct was recruited to Rab5 endosomes with rapamycin. (C) Recruitment of the WT PIKfyve-CCR-Kina to Rab5 endosomes and changes in biosensor localization at these membranes are shown by the fluorescence intensity ratios of the constructs at Rab5 endosomes over the whole cell. Data shown are the grand mean ± SEM of three experiments. A total of 35 cells were analyzed. Confocal images over time are shown to the right. (D) As in C, but now utilizing the catalytically dead K1876E/V1883I mutant of the FKBP-PIKfyve-CCR-Kina as a negative control. Data shown are the grand mean ± SEM for three experiments. A total of 39 cells were analyzed.

Our current understanding of PI(3,5)P2 localization is based on disease phenotypes, the localization of PIKfyve, and the localization of PI(3,5)P2 effector proteins such as WD-repeat proteins interacting with phosphoinositides (WIPI) or membrane channels. Based on this data over the years, PI(3,5)P2 has been associated with membranes across the endosomal pathway, including early endosomes, late endosomes, lysosomes, autophagosomes, and the multivesicular body (Ikonomov et al., 2001; Shisheva et al., 2001; Jeffries et al., 2004; Cabezas et al., 2006; Rutherford et al., 2006; Jefferies et al., 2008; Dove et al., 2009; Dong et al., 2010; Wang et al., 2012; Cai et al., 2013; McCartney et al., 2014; Banerjee et al., 2019; Leray et al., 2022). A more recent study that used immunofluorescence to visualize endogenous PIKfyve tagged by CRISPR/Cas9 showed strong co-localization of PIKfyve and the retromer complex, suggesting PI(3,5)P2 enrichment on membranes undergoing retrograde trafficking (Giridharan et al., 2022). However, the lack of agreement (and in some cases the presence of direct contradictions) in these collective studies highlights the need for more robust tools to directly monitor PI(3,5)P2.

Genetically encoded lipid biosensors are the gold standard for studying lipids by live cell imaging. Valid biosensors must show selectivity for the lipid of interest in vitro, lipid-dependent membrane association, and recruitment to membranes by the lipid alone (Wills et al., 2018). The first PI(3,5)P2 biosensors, ML1N and a higher-affinity counterpart ML1Nx2, were derived from the cation channel TRPML1 (Li et al., 2013; Samie et al., 2013). However, later studies showed that depleting PI(3,5)P2 did not cause the ML1Nx2 sensor to come off of endosomal membranes, indicating that off-target interactions are mediating membrane localization of the sensor (Hammond et al., 2015; Pemberton et al., 2025). Structural studies also showed that the N-terminal domain of TRPML1 used in the sensor was insufficient to bind PI(3,5)P2 (Chen et al., 2017; Hirschi et al., 2017). Thus, ML1Nx2 is not a useful tool to study PI(3,5)P2.

A more recent biosensor, using Dictyostelium sorting nexin-like protein SnxA, which contains PI(3,5)P2-binding phox homology (PX) domains, better fulfills the biosensor requirements (Vines et al., 2023). SnxA showed PI(3,5)P2 to be localized throughout the endosomal pathway, being somewhat enriched on Rab7 endosomes, but especially enriched on specialized endosomal membranes such as macropinosomes and phagosomes (Vines et al., 2023; Maib et al., 2024; Pemberton et al., 2025).

In vitro, both a biosensor made with full-length SnxA and a biosensor using the isolated PX domains in tandem (2xPX-SnxA) show specificity for PI(3,5)P2, with a Kd of 187 nM for SnxA and 217 nM for 2xPX-SnxA. Furthermore, membrane association of either biosensor in cells depends on PI(3,5)P2 production (Vines et al., 2023). This work noted that within cells, 2xPX-SnxA produced more striking images with better signal-to-noise. Interestingly though, overexpression of the 2xPX-SnxA sensor caused some growth defects that were attributed to lipid sequestration (Vines et al., 2023). However, as there is no direct comparison of these sensors’ response to PI(3,5)P2 production in cells, discerning which SnxA sensor to use is not straightforward.

Later, Pemberton et al. (2025) further demonstrated using a minimal, recruitable PIKfyve construct that PI(3,5)P₂ was sufficient to recruit SnxA to Rab5 endosomes. However, SnxA's ability to bind to PI(3,5)P₂ outside the endosomal system was not tested. As SnxA showed some basal localization at Rab5 endosomes, the possibility that SnxA binds to endosomes through coincidence detection is still unresolved. To definitively show that SnxA can unbiasedly bind to PI(3,5)P2 in any membrane where it may be produced, PI(3,5)P2 must be shown to recruit SnxA to an orthogonal membrane.

In this study, we independently validated the SnxA biosensor in live HEK293A cells. We showed that SnxA can unbiasedly report on PI(3,5)P2 production at membranes outside of the endosomal pathway and that it does so with more sensitivity than 2xPX-SnxA or ML1Nx2. We further confirmed that SnxA dissociates from membranes upon PI(3,5)P₂ depletion; we used SnxA to track PI3P conversion into PI(3,5)P₂, and we observed PI(3,5)P₂ loss during PIKfyve inhibition. Together, these results provide independent validation of SnxA as a selective and sensitive PI(3,5)P₂ biosensor and demonstrate its utility for tracking PI(3,5)P₂ dynamics in living cells and in disease models.

RESULTS

SnxA depends on PI(3,5)P2 for membrane localization

We first set out to reproduce previous results showing that SnxA depends on PI(3,5)P2 for its membrane localization within HEK293A cells. In other studies, this was done using deletion of PIKfyve or inhibition of PIKfyve with apilimod (Vines et al., 2023; Pemberton et al., 2025). Thus, we set out to validate this result by an independent method. We turned to the phosphatase myotubularin 1 (MTM1), which has been reported to dephosphorylate both PI(3,5)P2 and PI3P at the 3-OH position (Schaletzky et al., 2003). We used MTM1 in a chemically inducible dimerization system using the FK506 binding protein (FKBP) and the FKBP-rapamycin-binding domain from mTOR (FRB). This approach has been used previously to deplete PI3P from Rab5 endosomes (Hammond et al., 2014, 2015). Note that initial reports of MTM1 activity reported selectivity for PI3P, with no activity against PI(3,5)P₂ (Taylor et al., 2000). Therefore, we cannot conclude whether PI(3,5)P₂ depletion occurs due to direct, MTM1-mediated degradation to PI5P (perhaps with slower kinetics than PI3P is converted to PI), or whether instead, the precursor PI3P is depleted, preventing synthesis of a rapidly turning-over pool of PI(3,5)P₂ (which again could therefore occur with slower kinetics than PI3P depletion).

In this assay, rapamycin triggers FKBP-tagged MTM1 dimerization with sorting endosome-localized FRB-Rab5 or lysosome-targeted LAMP1-FRB. This results in acute depletion of PI3P and PI(3,5)P2. We monitored levels of PI3P using the EEA1-FYVE biosensor (Burd and Emr, 1998; Balla et al., 2000), quantified by determining the fluorescence intensity of the biosensor at Rab5 and LAMP1-positive membranes relative to its intensity throughout the whole cell. This same method was used to quantify relative PI(3,5)P2 levels using SnxA (Figure 1A). We saw that rapamycin robustly recruited FKBP-MTM1 to the Rab5-positive endosomes and LAMP1-positive lysosomes (Figure 1B). FYVE-EEA1 was eliminated from all visible membranes by FKBP-MTM1 recruitment to either FRB-RAB5 or LAMP1-FRB (Figure 1B), despite being much more enriched at the Rab5-positive membranes (Figure 1C). We also saw the loss of SnxA by MTM1 recruitment to LAMP1, and a significant reduction by recruitment to Rab5 (Figure 1B), although SnxA was only present at a small subset of either compartment and so showed poor quantitative enrichment at these membranes overall (Figure 1, C and D). The robust depletion of PI3P by MTM1 demonstrated robust activity of the enzyme under these conditions; the loss of predominantly Rab5-localized FYVE by recruitment of MTM1 to LAMP1 was unexpected. Given the depletion of the probe occurred in less than the 30 s image interval (Figure 1B), we speculate that a low level of LAMP1-FRB present in sorting endosomes may have recruited enough FKBP-MTM1 to deplete PI3P. These observations preclude conclusions on the localization of the lipid biosensors based on the extent of depletion following enzyme recruitment to either Rab5 or LAMP1. Indeed, we note measurable overlap of both biosensors with only a subset of Rab5- or LAMP1-positive membranes—except FYVE, which was present on all Rab5-positive membranes. Nonetheless, the data clearly demonstrate that SnxA indeed depends on the presence of PI(3,5)P2 (or PI3P) for localization to Rab5- or LAMP1-positive membranes (Figure 1B).

FIGURE 1:

FIGURE 1:

Open in a new tab

SnxA depends on PI(3,5)P2 for membrane localization. (A) Schematic depicting the recruitment of a chemically inducible dimerization system with MTM1, which leads to the dephosphorylation of both PI(3,5)P2 and PI3P, thereby removing SnxA and EEA1-FYVE from lysosomal membranes. (B) Recruitment of FKBP-MTM1 and loss of EEA1-FYVE and SnxA were quantified as the fluorescence intensity of the constructs at Rab5- or LAMP1-positive membranes relative to that within the whole cell. Data shown are the grand mean of three experiments ± SEM. A total of 34 (Rab5) or 35 (LAMP1) cells were analyzed. Representative confocal images of the first and last time point are shown. (C) Comparison of the relative enrichment of each biosensor on Rab5-positive or LAMP1-positive membranes. Data show means ± 95% confidence interval of the 34 or 35 cells pooled across the three experimental replicates; symbols show individual cell measurements, with triangle, square, and circle symbols matching each cell to its experimental replicate. (D) Representative confocal images of the co-localization of FRB-Rab5 and FRB-LAMP1 with subsets of GFP-SnxA-labeled membranes.

PI(3,5)P2 loss precedes vacuolation during apilimod treatment

As MTM1 depletes two lipid species that SnxA could possibly be interacting with, we next turned to apilimod treatment to inhibit PIKfyve and PI(3,5)P2 production specifically. We stained the endosomal pathway for 1 h with fluorescent dextran, which is passively loaded through fluid phase endocytosis and macropinocytosis. We then took z-stacks of the cells to image a large volume of endosomes during apilimod treatment. After 10 min of apilimod treatment, nearly all of the SnxA was removed from lysosomes/late endosomes (Figure 2A), in agreement with previous data (Vines et al., 2023; Pemberton et al., 2025). In contrast, DIC imaging revealed no sign of the characteristic vacuolation within this time period, which becomes fully apparent after 60 min (Figure 2A). These data support Figure 1 in showing that SnxA depends on PI(3,5)P2 for membrane localization.

FIGURE 2:

FIGURE 2:

Open in a new tab

PI(3,5)P2 loss precedes vacuolation during apilimod treatment. (A) Representative mean intensity projections of 11 0.175 µm equatorial confocal sections showing the loss of SnxA's vesicular localization a few minutes after treatment with 1 µM apilimod. Differential interference contrast images (DIC) show that vacuolation does not occur until 25–30 min after apilimod treatment. (B) Representative confocal projections as in A showing that treatment with 0.02% DMSO does not change SnxA localization. DIC images showing that 0.02% DMSO does not cause vacuolation. (C) Representative confocal projections as in A and B showing that 1 µM VPS34-IN-1 also causes robust and rapid elimination of SnxA localization. (D) Quantification of SnxA localization at dextran-positive lysosomes/late endosomes (LyLE) after treatment with 1 µM apilimod, 1 µM Vps34-IN-1 or 0.02% DMSO. The time series graph shows the mean ± SEM for 16 cells from one experiment (Apilimod), 12 cells from two experiments (DMSO) or 24 cells from three experiments (Vps34-IN-1). Solid lines show a one-phase decay fit for the Apilimod and Vps34 treatment, yielding time constants plotted beneath showing the 95% confidence interval of the fitted value.

In the DMSO control dish, SnxA remained stably associated with the lysosomes/late endosomes, and no vacuolation occurred (Figure 2B). On the other hand, inhibiting PI3P synthesis with VPS34-IN-1 caused a similar rapid decrease in SnxA-labeled structures (Figure 2C). When the loss of SnxA was quantified, fitting a one-phase decay model determined that SnxA came off the membranes with a time constant (τ) of approximately 5 min with either Vps34 or PIKfyve inhibitor, suggesting continuous, rapid turnover of PI(3,5)P₂ and its precursor (Figure 2D).

The quick depletion of PI(3,5)P2 that we measured by SnxA after apilimod treatment is in agreement with data obtained using the PIKfyve inhibitor YM201636. Vacuolation was visualized 35 min after the addition of the inhibitor, but HPLC data showed that PI(3,5)P2 levels were depleted to their minimum level in only 2.5 min (Jefferies et al., 2008; Zolov et al., 2012). This experiment further demonstrates the effectiveness of SnxA, as this biosensor allowed us to confirm biochemical kinetic data in real-time and with live cells.

Note that SnxA also partially labeled the plasma membrane in some cells (for example, Figures 1B and 2, B and C). However, this weak staining is not depleted by inhibitors, and likely does not reflect a PI(3,5)P₂-driven interaction, as noted in the initial description of the sensor (Vines et al., 2023).

PI(3,5)P2 does not significantly overlap with PI3P

When imaging cells, we noticed that often SnxA and the PI3P biosensor EEA1-FYVE do not overlap. Although we saw some degree of overlap of both biosensors with Rab5 and LAMP1 (Figure 1C), we found SnxA only localized to a subset of each, and FYVE only localized to a subset of LAMP1 vesicles (insets in Figure 1D), suggesting they do not reside on the same vesicle. To further define this relationship between the endosomal lipids, we analyzed the images taken throughout this study, before any manipulations were done, to see if there was co-localization between SnxA and EEA1-FYVE. While we see some SnxA and EEA1-FYVE in proximity to each other, we don't often see co-localization of the two. Furthermore, we can find several discrete examples of vesicles only marked with SnxA, as shown by white arrows (Figure 3A).

FIGURE 3:

FIGURE 3:

Open in a new tab

PI(3,5)P2 does not significantly overlap with PI3P. (A) Confocal images of cells used throughout this study expressing both SnxA and EEA1-FYVE. The white arrows show examples of where the biosensors don't co-localize. The figure from which each cell is duplicated is as noted. (B) Time-lapse images of a representative vesicle showing the transition from EEA1-FYVE labelling to SnxA labelling in an untreated cell expressing both biosensors.

Additionally, in the z-stack images, we saw several good examples of vesicles that switched from being EEA1-FYVE-labeled to being SnxA-labeled. First, this indicated the conversion of PI3P to PI(3,5)P2 and thus increased our confidence that SnxA is responding to PI(3,5)P2. Second, this transition was fairly quick, happening in about 4 min. Notably, it only occurred in a small subset of EEA1-FYVE-positive vesicles (Figure 3B). This quick transition could account for the lack of SnxA and EEA1-FYVE overlap seen in static images. Overall, these data demonstrate that the SnxA sensor can provide important insights into the relationship between PI(3,5)P2 and the other phosphoinositides.

Design of a recruitable PIKfyve construct

In the recent study characterizing SnxA, Pemberton and colleagues designed a minimal catalytic fragment of PIKfyve that could be used in the FKBP-FRB system (Pemberton et al., 2025). While that manuscript was in preparation, we had independently designed a similar construct, using AlphaFold3 (Abramson et al., 2024) with some trial and error to determine which domains of the full-length murine PIKfyve were necessary to retain activity. We found that the conserved cysteine-rich (CCR) domain was packed against the kinase domain and therefore likely impacted folding or stability of the kinase domain. However, the CCR itself could only properly fold when the three-helix bundle was present. Then, for use in our imaging experiments, we also included FKBP and a BFP fluorophore with some flexible linkers. Using AlphaFold3, we saw that the structures of the PIKfyve domains in our construct resembled the folded domain structures of full-length PIKfyve. As this construct contains the kinase domain and the CCR, we named this construct FKBP-PIKfyve-CCR-Kina (Figure 4A). We then added mutations (K1876E and V1883I in full-length M. musculus PIKfyve) that inactivate PIKfyve to create FKBP-PIKfyve-CCR-Kina-Dead as a control. Note that for these experiments, the recruitable constructs are transiently expressed in unedited HEK293A cells, which retain their native PIKfyve activity. Therefore, any induced PI(3,5)P₂ synthesis is expected to occur in addition to the baseline levels found in unperturbed cells.

We first used the FKBP-PIKfyve-CCR-Kina to produce PI(3,5)P2 on Rab5 endosomes, where its substrate PI3P endogenously resides (Figure 4B). Rapamycin recruited FKBP-PIKfyve-CCR-Kina to Rab5-positive membranes, and we saw a subsequent depletion of PI3P shown by loss of EEA1-FYVE. The resulting PI(3,5)P2 production then proved to be sufficient for SnxA recruitment, as we saw very robust labeling of Rab5 endosomes with SnxA (Figure 4C). When FKBP-PIKfyve-CCR-Kina-Dead was expressed as a control, we did not see a significant change in either EEA1-FYVE or SnxA localization, indicating that the changes in the biosensors seen previously were dependent on PIKfyve activity (Figure 4D).

Production of PI(3,5)P2 at orthogonal membranes is sufficient to recruit SnxA

Next, to fully confirm that PI(3,5)P2 is sufficient for SnxA recruitment, we targeted the FKBP-PIKfyve-CCR-Kina constructs to the mitochondrial membrane, which lacks PI3P and so presumably also lacks typical endosomal effector proteins. Therefore, producing PI(3,5)P2 at this membrane is a more rigorous way to show that PI(3,5)P2 is sufficient for SnxA membrane association. To make PI(3,5)P2 at mitochondria, we co-expressed FKBP-MavQ along with the FKBP-PIKfyve-CCR-Kina. MavQ is a L. pneumophila effector that produces PI3P on the ER through the phosphorylation of phosphatidylinositol (PI; Hsieh et al., 2021). Therefore, we recruited FKBP-MavQ to mitochondrial membranes so it could supply the FKBP-PIKfyve-CCR-Kina with its PI3P substrate (Figure 5A).

FIGURE 5:

FIGURE 5:

Open in a new tab

Production of PI(3,5)P2 at orthogonal membranes is sufficient to recruit SnxA. (A) Schematic showing ectopic PI(3,5)P2 production on mitochondrial membranes through co-expression and co-recruitment of FKBP-MavQ and FKBP-PIKfyve-CCR-Kina. (B) Quantification of the WT PIKfyve, MavQ, and SnxA at the mitochondria using the ratio of fluorescence at the mitochondria to that within the whole cell. Data shown are the grand mean ± SEM of 5 experiments. A total of 39 cells were analyzed. Representative confocal images of the time course are shown to the right. (C) As in B, but with the kinase-dead mutant of the FKBP-PIKfyve-CCR-Kina. Data shown are the grand mean of 5 experiments ± SEM. A total of 35 cells were analyzed. (D) As in B, except MavQ was replaced with an empty mCherry vector. Data shown are the grand mean of three experiments ± SEM. A total of 20 cells were analyzed.

Interestingly, FKBP-MavQ already showed some co-localization with mitochondrial membranes even before rapamycin addition (Figure 5, B and C). However, we did see that rapamycin induced recruitment of FKBP-PIKfyve-CCR-Kina and increased recruitment of FKBP-MavQ. The resulting production of PI(3,5)P2 then caused robust localization of SnxA at the mitochondria (Figure 5B), which was not seen when the FKBP-PIKfyve-CCR-Kina-Dead was used (Figure 5C) or when FKBP-MavQ was omitted (Figure 5D). Therefore, the recruitment of SnxA to the mitochondrial outer membrane depended on both enzymes that convert PI to PI3P and PI3P to PI(3,5)P₂, eliminating the possibility that SnxA was instead binding PI5P rather than PI(3,5)P₂ in this context (though it should be stated, there is no positive control that PIKfyve was able to generate PI5P from PI at the mitochondrial outer membrane). Overall, these data increased our confidence that SnxA is a robust PI(3,5)P2 biosensor.

SnxA outcompetes 2xPX-SnxA and ML1Nx2 for PI(3,5)P2 binding

Now having seen that SnxA fully meets the criteria for a PI(3,5)P2 biosensor, we wanted to show its sensitivity as compared with the other PI(3,5)P2 biosensors 2xPX-SnxA and ML1Nx2, as direct comparison of the responses of these sensors to PI(3,5)P2 has not been done previously.

We first co-expressed SnxA and 2xPX-SnxA in the same cell to simultaneously monitor their response to PI(3,5)P2 production at Rab5 endosomes. 2xPX-SnxA showed strong nuclear localization as described previously. However, this nuclear accumulation can be circumvented through the addition of a nuclear export sequence (Pemberton et al., 2025). After recruitment of FKBP-PIKfyve-CCR-Kina, the 2xPX-SnxA sensor labeled the Rab5 endosomes. In comparison, SnxA showed some slight enrichment at Rab5 endosomes under basal conditions, and then more robustly translocated to Rab5 endosomes after rapamycin addition (Figure 6A). Importantly, neither biosensor was enriched at Rab5 endosomes when FKBP-PIKfyve-CCR-Kina-Dead was recruited there (Figure 6B). Altogether, these data align with the reported high affinity for SnxA and suggest that SnxA is outcompeting 2xPX-SnxA for PI(3,5)P2 binding.

FIGURE 6:

FIGURE 6:

Open in a new tab

Comparison of SnxA, 2xPX-SnxA and ML1Nx2 for ectopic PI(3,5)P2 binding. (A) Recruitment of the FKBP-PIKfyve-CCR-Kina to Rab5 membranes and the subsequent translocation of SnxA and 2xPX-SnxA biosensors was quantified as a fluorescence intensity ratio of fluorescence at Rab5 endosomes to that within the whole cell. Data shown are the grand mean ± SEM of three independent experiments. A total of 34 cells were analyzed. Representative confocal images of the first and last time point are shown to the right. (B) The kinase-dead mutant of the FKBP-PIKfyve-CCR-Kina construct was used in an experiment parallel to that in A as a negative control. Data shown are the grand mean ± SEM of three independent experiments. A total of 38 cells were analyzed. (C) The SnxA biosensor was co-expressed with the ML1Nx2 biosensor to compare their responses to PI(3,5)P2 production at Rab5 endosomes by FKBP-PIKfyve-CCR-Kina. Data shown are the grand mean ± SEM of three independent experiments. A total of 36 cells were analyzed. (D) The control experiment for C, showing the response of SnxA and ML1Nx2 to the kinase-dead FKBP-PIKfyve-CCR-Kina at Rab5 endosomes. Data shown are the grand mean ± SEM of three independent experiments. A total of 36 cells were analyzed.

We next tested ML1Nx2 against SnxA by co-expressing them in the same cells. In this case, we did not see ML1Nx2 respond to PI(3,5)P2 production by FKBP-PIKfyve-CCR-Kina at Rab5 endosomes, while SnxA showed very strong recruitment (Figure 6C). The SnxA translocation was lost when the catalytically dead PIKfyve construct was used (Figure 6D). These data agree with past evidence that suggested that ML1Nx2 targets endosomes through off-target interactions, and therefore PI(3,5)P2 is not sufficient to recruit ML1Nx2 to endosomal membranes (Hammond et al., 2015). However, when the ML1Nx2 sensor was converted into a more sensitive BRET-based probe, a very small response to PI(3,5)P2 production on Rab5 endosomes was seen (Pemberton et al., 2025). Overall, though, our data and previous results all indicate that SnxA and 2xPX-SnxA have far superior sensitivity and selectivity in comparison to ML1Nx2. Therefore, use of the SnxA-based biosensors over ML1Nx2 is highly recommended.

DISCUSSION

In this study, we showed that SnxA membrane localization depends on PI(3,5)P2 (Figures 1 and 2) and that PI(3,5)P2 production by a minimal PIKfyve construct, PIKfyve-CCR-Kina, is sufficient to recruit SnxA to mitochondrial membranes, which aren't expected to contain other endosomal lipids or effectors (Figures 4 and 5). Finally, we demonstrated that SnxA responds to PI(3,5)P2 production more robustly than the 2xPX-SnxA and ML1Nx2 sensors (Figure 6). Overall, our work validated SnxA as a selective and sensitive PI(3,5)P2 biosensor. We showed that SnxA is suitable for live-cell studies of PI(3,5)P2 dynamics, as SnxA allowed us to visualize loss of PI(3,5)P2 during apilimod treatment and its conversion from PI3P (Figures 2 and 3). SnxA will thus be instrumental in future studies investigating PI(3,5)P2's regulation in normal cellular physiology and in disease contexts.

Importantly, our study is the first to show that SnxA responds to PI(3,5)P2 produced in an orthogonal membrane outside of the endosomal pathway (Figure 5). While previous studies showed that SnxA localization at Rab5 endosomes was decreased with PI(3,5)P2 depletion and increased with PI(3,5)P2 synthesis, these experiments did not address whether SnxA was binding to Rab5 endosomes through coincidence detection of PI(3,5)P2 and another lipid or protein (Vines et al., 2023; Pemberton et al., 2025). However, by showing SnxA recruitment to mitochondria upon PI(3,5)P2 production, we have definitively shown that PI(3,5)P2 is sufficient for SnxA to bind to membranes.

The ability of a biosensor to unbiasedly report on lipid levels is a key feature of a reliable biosensor, as it ensures that the biosensor will detect the lipid in any membrane in which it is produced, rather than only “seeing” a specific pool of lipid. For example, PI4P was detected only in the Golgi by PH-FAAP1 biosensors, which localize to the Golgi by interactions with both PI4P and Arf1. However, the P4M biosensor, which was shown to detect ectopic PI4P made in the ER and therefore bind to PI4P but not Arf1, also revealed PI4P pools in endosomes, lysosomes, and at the PM (Hammond et al., 2014).

Furthermore, this work directly compared SnxA, 2xPX-SnxA, and ML1Nx2 by expressing the sensors in the same cell (Figure 6). Pemberton and team compared 2xPX-SnxA to the monomeric probe Px-SnxA, but not the full-length SnxA probe (Pemberton et al., 2025). Vines and team showed SnxA to have higher affinity for PI(3,5)P2 in vitro compared with 2xPX-SnxA, and SnxA showed more basal vesicular localization compared with 2xPX-SnxA, but this study only used static imaging (Vines et al., 2023). We therefore directly compared 2xPX-SnxA to SnxA over time within the same cell. Our data agreed with SnxA having higher affinity for PI(3,5)P2, as we saw SnxA have more basal localization at Rab5 endosomes and a more robust response to PI(3,5)P2. Additionally, we confirmed previous data showing that ML1Nx2 does not accurately respond to PI(3,5)P2 levels.

The difference between full-length SnxA and 2xPX-SnxA is surprising, since the difference in affinity for PI(3,5)P₂ in artificial membranes is very small (Vines et al., 2023). Indeed, the full-length SnxA includes a coiled-coil domain that likely induces dimerization, so both probes are in effect dimeric PI(3,5)P₂-binding PX domains. One trivial explanation could be that the dimer of SnxA proteins includes two copies of the fluorescent protein tag, effectively increasing signal-to-noise. However, this is unlikely to account for the rather large difference in both basal enrichment and PIKfyve-induced endosome recruitment observed in Figure 6. Instead, we speculate that the coiled-coil orients the two PX domains in a more favorable relative conformation for concurrent PI(3,5)P₂ binding compared with 2xPX-SnxA. In this context, it will be interesting to observe the effects of the recently identified higher-affinity G13V mutant in both probe configurations (Nishimura et al., 2025).

While this study confirmed SnxA's high sensitivity, our data do not address whether SnxA sequesters PI(3,5)P2 and disrupts its interactions with effector proteins. This consideration is especially important as PI(3,5)P2 is a very low-abundance lipid. It is estimated to be about 0.038% of the total phosphoinositides in mouse embryonic fibroblasts (Steinfeld et al., 2021). At these low levels, it is a real concern that the overexpressed biosensor will bind to all of the available lipid and sequester it from effector proteins. Aligning with this idea, expression of the 2xPX-SnxA biosensor has been previously associated with growth defects, which are potentially a result of disrupted lipid signaling (Vines et al., 2023). On the other hand, it is somewhat reassuring that extensive vacuolation of the cell is not seen with SnxA or 2xPX-SnxA expression; if the probes completely sequestered PI(3,5)P₂, we would expect the same phenotype as PIKfyve inhibition.

One way to circumvent lipid sequestration is to use a lower affinity biosensor that will dissociate from the lipid more easily. However, a lower affinity biosensor using a single PX domain (Px-SnxA) did not show in vitro binding to PI(3,5)P2 or any basal endosomal localization (Vines et al., 2023; Pemberton et al., 2025). Therefore, we suggest that expressing the full-length SnxA biosensor at lower levels could minimize competition with effector proteins. We have seen this approach work with biosensors for PIP3 (Holmes et al., 2025).

One of the most important uses of SnxA will be in the study of PIKfyve-related diseases. PIKfyve dysfunction has been linked to several diseases, yet the role of PI(3,5)P2 in disease pathology is still unclear. PIKfyve-generated PI(3,5)P2 is the precursor for phosphatidylinositol 5-phosphate (PI5P) by way of MTM-mediated dephosphorylation of PI(3,5)P2 (Zolov et al., 2012). Therefore, the effects of PI(3,5)P2 loss are intertwined with the effects of PI5P loss in PIKfyve-related diseases. Notably, there is no characterized biosensor for PI5P, so before SnxA's discovery, distinguishing between the role of PI(3,5)P2 and PI5P in cellular events has been difficult. Furthermore, the relationship between PIKfyve and levels of these lipids is not straightforward in disease states. Mouse embryonic fibroblasts with only one PIKfyve allele show a 60% reduction in PI(3,5)P2 and PI5P levels despite only a 50% decrease in PIKfyve protein levels (Ikonomov et al., 2011). Therefore, the fully validated SnxA biosensor will be invaluable for detailed studies of PI(3,5)P2 in disease models.

In this work, SnxA has already corroborated important information for our understanding of PIKfyve-related diseases. SnxA confirmed that PI(3,5)P₂ depletion precedes cellular changes. Apilimod treatment eliminated SnxA's association with late endosomes and lysosomes within 9.5 min, whereas endosomal vacuolation began only after 25–30 min (Figure 2). The time course of PIKfyve inhibition gradually leading to vacuolation is well-documented (Choy et al., 2018; Sbrissa et al., 2018). Additionally, a rapid loss of PI(3,5)P₂ with PIKfyve inhibition ( < 5 min) has been determined biochemically (Zolov et al., 2012). However, when the ML1Nx2 biosensor was used to monitor PI(3,5)P2 loss with the inhibitor YM201636, a gradual decrease in PI(3,5)P2 was seen. Significant loss of PI(3,5)P2 occurred after 3 h of inhibitor treatment, and further loss was seen with overnight treatment (Li et al., 2013). Here we show that the SnxA sensor, which better fulfills the biosensor criteria as compared with ML1Nx2, is more in agreement with the biochemical data than the ML1Nx2 studies were. This shows the power of SnxA; it allowed us to confirm PI(3,5)P2 kinetics in real-time that had previously only been seen biochemically.

Additionally, the lagging appearance of large vacuoles after PI(3,5)P2 reduction suggests that the lipid regulates endosome size through a delayed mechanism, but the exact mechanism underlying this phenotype is still unclear. One model hypothesizes that PIKfyve inhibition leads to vesicle swelling through changes in ion homeostasis and osmosis. For example, PI(3,5)P2 activation of two-pore channels (TPCs) causes Na+ efflux from the lysosome (Wang et al., 2012). In macropinosomes, this TPC-dependent ion movement results in loss of water and vesicle shrinkage (Freeman et al., 2020). Therefore, the increase in vacuolar size with PIKfyve inhibition could be due to the inability of the TPCs to drive water efflux from the vesicles. In support of this hypothesis, applying a hyperosmotic solution to PIKfyve-inhibited cells reversed the swelling of the vacuole (Freeman et al., 2020).

Similarly, PI(3,5)P2 was shown to regulate Cl-/H+ exchangers in endosomes. The voltage-gated chloride channel 3 (ClC-3) regulates vacuolation of endosomes by increasing luminal Cl- levels, which causes osmotic swelling (Sonawane et al., 2002). However, ClC-3 can be inhibited by a regulatory T9A subunit blocking the ion pore. PI(3,5)P2 was shown to stabilize the association of ClC-3 and T9A. Apilimod-induced depletion of PI(3,5)P2 overrode T9A overexpression to cause constitutive ion influx and ClC-3-mediated vacuolation (Schrecker et al., 2025).

In another model, the loss of PI(3,5)P2 affects endosomal pH. This occurs through dysregulation of v-ATPases, causing improper endosomal/vacuole acidification (Yamamoto et al., 1995; Gary et al., 1998; Banerjee and Kane, 2020). While it is still unclear how pH regulates vacuolation under PIKfyve inhibition, it is thought that proper pH levels are needed to maintain endosomal identity and regulate fission or fusion events, or that a tightly controlled pH is needed for proper ion homeostasis and water flux. Therefore, upon PIKfyve inhibition, enlarged endosomes are speculated to occur due to dysregulated pH levels leading to increased fusion or increased water influx. Changing the pH of endosomal compartments has been seen to reverse vacuolation (Nicoli et al., 2019), but other studies show that pH does not affect vacuole size (Leray et al., 2022). Therefore, more work needs to be done to determine the exact relationship between pH and endosomal size and to uncover how PIKfyve inhibition alters it.

Finally, PIKfyve inhibition is thought to increase the size of endosomes through membrane remodeling mechanisms. In these cases, inhibiting PIKfyve activity disrupts the balance of lysosomal fusion and fission events (Choy et al., 2018), or reduces the capacity of endolysosomes to form lysosomes (Bissig et al., 2017). Both of these mechanisms result in the stalling of endosomal trafficking where endosomes gain volume from upstream trafficking events but then don't dissipate that volume through further trafficking or fission events.

While any of these mechanisms could explain the delay in the appearance of the endosomal phenotype after rapid PI(3,5)P2 loss, all of them will need further testing. Our ability to test these models is now more achievable using the SnxA biosensor and the FKBP-FRB systems that we describe here. Using these tools in combination with live-cell imaging will allow us to determine the effects of acute changes in PI(3,5)P2 on things like endosomal ion levels or fusion events.

The SnxA biosensor will also help us uncover more detail about PI(3,5)P2 in the context of the other phospholipids. There was evidence that PI(3,5)P2 was synthesized from PI3P since the first discovery of PI(3,5)P2 (Dove et al., 1997). However, there has been some conflicting data as to the localization of PI(3,5)P2 relative to that of PI3P. Some early studies suggested that PIKfyve and PI3P localize in separate membrane domains within early endosomes (Cabezas et al., 2006), which brings up the question: how much overlap is there between PI3P and PI(3,5)P2 in the endosomal pathway? Consistent with prior work (Vines et al., 2023; Pemberton et al., 2025), we observed little overlap between the two lipids in static images of a population of labeled membranes (Figure 3). But we could observe a short temporal overlap between the two lipids within a single membrane compartment, lasting only a few minutes. Furthermore, we observed that this transition happens spontaneously in cells (Figure 3), and we did not see any evidence of lipid-specific membrane domains. This short window of co-localization will be important to consider as we continue to elucidate the functions of PI(3,5)P2 effector proteins, some of which, like WIPI49, bind to both PI3P and PI(3,5)P2 (Jeffries et al., 2004).

Overall, we have found SnxA to be a selective and sensitive biosensor that fills a critical gap in the tools available to study the sparse but essential lipid PI(3,5)P2.

MATERIALS AND METHODS

Cell culture and transfection

HEK293A cells (Invitrogen R70507) were maintained at 37°C and 5% CO2. They were passaged 1:5 twice weekly using the following protocol: flasks were washed with 10 ml 1x PBS (Fisher Scientific BP3994), 1 ml TrypLE (Life Technologies 12604039) was used to dissociate the cells, and 4 ml complete Dulbecco's modified Eagle's media (DMEM; Life Technologies 10567022, supplemented with 10% HI-FBS (Life Technologies A5669801), 1% 100 u/ml penicillin and 100 µg/ml streptomycin (Life Technologies 15140122), 0.1% chemically defined lipid supplement (Life Technologies 11905031) was added to the flask and the cells were resuspended by pipetting. The suspended cells were then seeded into 35 mm glass bottom dishes with a 20 mm aperture (Thermo Fisher Scientific D35-20-1.5-N). Dishes were pre-coated with 10 µl 1 mg/ml ECL cell attachment matrix (Sigma 08-110) diluted in 0.5 ml incomplete DMEM. The coating was incubated on the dishes for 30+ min at 37°C. After the incubation period, the coating solution was replaced with 2 ml pre-warmed complete DMEM. The cells were added into this medium, seeding enough cells that the dishes would be 100% confluent on the day of imaging. The cells were then left to settle onto the dish for 2+ h at 37°C before transfecting.

To transfect each dish of cells, 1 µg of DNA (the needed plasmids were mixed to obtain 1 µg total) was diluted in 100 µl of Opti-MEM (Life Technologies 51985091). Then, 3 µl Lipofectamine2000 (Life Technologies 11668019) was mixed in 100 µl Opti-MEM. These two solutions were mixed and then incubated for 5–20 min at room temperature. 200 µl of the resulting solution was added to each dish of cells. The dishes were swirled to mix and then incubated at 37°C for 4 h. At this time, the transfection solution was removed and replaced with 2 ml Complete HEPES-buffered Imaging Media (CHIM; 10% HI-FBS, 1% Glutamax (Life Technologies 35050061), 0.1% chemically defined lipid supplement, 2.5% 1 M Na-HEPES buffer at pH 7.4 (VWR EM-5320) diluted in 500 ml Fluorobrite media (Life Technologies A1896702). The cells were then kept in the incubator overnight and imaged the next day.

Dextran labeling

For cells labeled with dextran, the transfection mixture was replaced with 2 ml complete DMEM rather than CHIM the night before. Then, on the day of imaging, a solution of 100 µg/ml Alexa Fluor 647 Dextran (ThermoFisher D22914) was made up in CHIM. Note that the dextran solution must be protected from light as much as possible. The media on the cells was aspirated, and 500 µl of the dextran solution was added to the glass inset of each dish. The cells were incubated in the dextran for 1 h at 37°C. After the incubation, the cells were washed with 1 ml CHIM, the wash was aspirated, and then 2 ml CHIM was added to the dish for imaging.

Plasmids and cloning

The plasmids used in this study are listed in Table 1. Plasmids were cloned using PCR and the NEBuilder HiFi DNA Assembly Cloning Kit (New England Biolabs E5520S) or restriction digest (FastDigest Thermo Scientific) followed by ligation (T4 DNA Ligase, Thermo Scientific EL0011). Vector and insert sequences were either ordered as a custom gene block (IDT) or isolated using primers for an existing plasmid. Proper plasmid synthesis was confirmed by Sanger sequencing over the insertion sites and/or by long-read nanopore sequencing over the whole plasmid (GENEWIZ from Azenta Life Sciences).

TABLE 1:

Plasmids used in this study. Numbered residues refer to their position in the full-length protein. All genes are H. sapiens unless otherwise specified. Residues written out in 1-letter code indicate linker sequences.

Name Sequence Reference
EGFP-SnxA EGFP-SGLRS-D.discoideum SnxA1-186 (DDB_G0289833)- RILQSTVPRARDPPDLDN (Vines et al., 2023)
mCherry-SnxA mCherry-SGLRS-D.discoideum SnxA1-186(DDB_G0289833)- RILQSTVPRARDPPDLDN (Vines et al., 2023)
mCherry-FYVE-EEA1 mCherry- SGLRS-EEA11253-1411 This study, adapted from (Balla et al., 2000)
iRFP-LAMP1-FRB LAMP11-417- SEFGSTGSTGSTGADPTRSANSGAGAGAGA-MTOR2021-2113- TSYPYDVPDYA-iRFP (Goulden et al., 2019)
BFP-FKBP-MTM1 TagBFP- SGLRSRSAAAGAGGAARAALG-FKBP1A3-108- SAGGSAGGSAGGSAGGSAGGPRAQASNSAVD-MTM11-603 This study, adapted from (Hammond et al., 2014)
iRFP-Rab5-FRB iRFP713-SGLRSRAGGAGAILSR-MTOR2021-2113-GGSAGGSAQASNSAVDGT-C. familiaris Rab5a1-215 (Hammond et al., 2014)
BFP-FKBPflex-PIKfyve-CCR-kina TagBFP- SGLRSRSAAAGAGGAARAALG-FKBP1A3-108- SAGGSAGGSAGGSAGGSAGGPRAQ-M. musculus PIKfyve isoform 11384-2097 This study
BFP-FKBPflex-PIKfyve-CCR-kina-dead TagBFP- SGLRSRSAAAGAGGAARAALG-FKBP1A3-108- SAGGSAGGSAGGSAGGSAGGPRAQ-M. musculus PIKfyve isoform 11384-2097,K1876E,V1883I This study
mCherry-FKBP-MavQ L. pneumophila MavQ11-590- GGSGGSGGSGGVPMG- FKBP1A3-108-RDPPVAT-mCherry This study
iRFP-Mito-FRB iRFP-SGLRSRAGGAGAILSR-MTOR2021-2113-GGSAGGSAQASNSAVDGTA-Fis1122-152 (Doyle et al., 2024)
EGFP-2xPX-SnxA EGFP-SGLRSM-D.discoideum SnxA55-186 (DDB_G0289833)-QNSAVDGTAGPGSM- D.discoideum SnxA55-186 (DDB_G0289833)- QNSDPPDLDN (Vines et al., 2023)
GFP-ML1Nx2 EGFP-YSDLT-M. musculus Mcoln11-68-RILT- M. musculus Mcoln11-68-DPPDLDN (Li et al., 2013)
puC19 Empty plasmid used to bring total DNA mass up to 1 µg as needed (Yanisch-Perron et al., 1985)
Open in a new tab

Confocal imaging

Imaging was done using a Nikon A1R confocal microscope with a Nikon TiE inverted stage and a heated stage incubator to keep the dishes at 37°C (Tokai Hit). A dual fiber-coupled laser LU-NV launch was used to excite the cells: 405 nm for blue fluorophores, 488 nm for green fluorophores, 561 nm for red fluorophores, and 640 nm for far-red fluorophores. A channel series was used so that cells were excited with 405/561 nm lasers in one pass and 488/640 nm lasers in a separate pass in order to avoid crosstalk. Emission was collected by 500–550 nm (green), 663–737 nm (far-red), 570–620 nm (red), and 425–475 nm (blue) fibers. A 100 × 1.45 NA plan apochromatic oil immersion objective was used, and the pinhole opening was set to 1.2x the Airy disk size of the longest wavelength being imaged.

To take the images, 8x line averaging, a 512-pixel scan area size, and 2x zoom were used. In some experiments, the Nikon Elements denoising software was also used. To image, first 10+ cells appropriately expressing the plasmids were found and their positions marked. A time-lapse was then set to take images every 30 s for the total duration as noted in the figure legends. After 2 min of baseline imaging, the cells were stimulated with 500 µl of the solutions described in Table 2. Solutions were made up at a concentration 5x of that listed in Table 2 since adding 500 µl solution to the 2 ml of CHIM on the cells resulted in a 1:5 dilution.

TABLE 2:

Cell stimulations used throughout this study.

Reagent Manufacturer Catalogue number Stock solution Concentration used on cells (diluted in CHIM) Figures used
Rapamycin EMD 80054-244 1 mM 1 µM Figure 1, 4, 5, 6
Apilimod Millipore-Sigma SML2974-5MG 5 mM 1 µM Figure 2
DMSO Millipore-Sigma D2653-5 × 5ML 100% 0.02% Figure 2, 3
VPS34-IN-1 Millipore-Sigma 532628 5 mM 1 µM Figure 2
Open in a new tab

Z-stack imaging

To image z-stacks of the cells, the confocal microscope was used as described above, with the additional use of the Nikon A1 Piezo Z Drive. The z-stack was set relative to a manual slice set in the middle of the cell. Then, the stack imaging moved from bottom to top in 0.175 µm steps across a range of 10 µm (−5.00 and +5.00), with focus being maintained during time-lapse by the “perfect focus” feedback system. To present and analyze these images, only 11 slices in the middle of the cell were used as an average z-projection, as these slices contained the majority of the SnxA-labeled vesicles.

Confocal image analysis

Fiji was used to complete all image analysis (Schindelin et al., 2012). To quantify the intensity ratios of constructs, a membrane marker was used to create a binary mask: either LAMP1-FRB, dextran, Rab5-FRB, or Mito-FRB.

The membrane marker's fluorescence underwent Gaussian blur filters at 4x, 3x, 2x, or 1x the size of the Airy disk of the fluorophore. The resulting images were then subtracted from the image created by the next smaller filter in order to create wavelets. The wavelets were multiplied together and had a threshold of 3x the SD of the original image applied. A 1 or 2-pixel dilation cycle was used on the wavelet image, and the background was subtracted using a threshold of 1.5x the mode of the fluorescence intensity. Cells that moved too much during the time course or were dead/dying as determined by their morphology in the DIC channel were excluded from analysis.

The intensity of the construct within the mask was measured and then divided by its intensity throughout the whole cell (as defined by a manually drawn ROI) to create the given ratio. For further details on analysis, please refer to (Wills et al., 2021). All quantification was plotted using GraphPad Prism 9 or later.

ACKNOWLEDGMENTS

This work was supported by NIH grants R35GM119412 (G.R.V.H) and 1F31HL170755-01 (C.C.W.).

Abbreviations used:

CCR

conserved cysteine-rich

DIC

differential interference contrast

FKBP

FK506-binding protein

FRB

FKBPrapamycin-binding domain

LAMP1

lysosome-associated membrane protein 1

MTM1

myotubularin 1

PX

phox homology.

Footnotes

This article was published online ahead of print in MBoC in Press (http://www.molbiolcell.org/cgi/doi/10.1091/mbc.E26-03-0120) on August 5, 2026.

REFERENCES

  1. Abramson J, Adler J, Dunger J, Evans R, Green T, Pritzel A, Ronneberger O, Willmore L, Ballard AJ, Bambrick J, et al. (2024). Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630, 493–500. [DOI] [PMC free article] [PubMed] [Google Scholar]
  2. Balla T, Bondeva T, Várnai P (2000). How accurately can we image inositol lipids in living cells? Trends Pharmacol Sci 21, 238–241. [DOI] [PubMed] [Google Scholar]
  3. Banerjee S, Clapp K, Tarsio M, Kane PM (2019). Interaction of the late endo-lysosomal lipid PI(3,5)P2 with the Vph1 isoform of yeast V-ATPase increases its activity and cellular stress tolerance. J Biol Chem 294, 9161–9171. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Banerjee S, Kane PM (2020). Regulation of V-ATPase activity and organelle pH by phosphatidylinositol phosphate lipids. Front Cell Dev Biol 8, 510. [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Bissig C, Hurbain I, Raposo G, van Niel G (2017). PIKfyve activity regulates reformation of terminal storage lysosomes from endolysosomes. Traffic 18, 747–757. [DOI] [PubMed] [Google Scholar]
  6. Burd CG, Emr SD (1998). Phosphatidylinositol(3)-phosphate signaling mediated by specific binding to RING FYVE domains. Mol Cell 2, 157–162. [DOI] [PubMed] [Google Scholar]
  7. Burke JE, Triscott J, Emerling BM, Hammond GRV (2022). Beyond PI3Ks: targeting phosphoinositide kinases in disease. Nat Rev Drug Discov 1–30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Cabezas A, Pattni K, Stenmark H (2006). Cloning and subcellular localization of a human phosphatidylinositol 3-phosphate 5-kinase, PIKfyve/Fab1. Gene 371, 34–41. [DOI] [PubMed] [Google Scholar]
  9. Cai X, Xu Y, Cheung AK, Tomlinson RC, Alcázar-Román A, Murphy L, Billich A, Zhang B, Feng Y, Klumpp M, et al. (2013). PIKfyve, a class III PI kinase, is the target of the small molecular IL-12/IL-23 inhibitor apilimod and a player in toll-like receptor signaling. Chem Biol 20, 912–921. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Chen Q, She J, Zeng W, Guo J, Xu H, Bai X, Jiang Y (2017). Structure of mammalian endolysosomal TRPML1 channel in nanodiscs. Nature 550, 415–418. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Chow CY, Landers JE, Bergren SK, Sapp PC, Grant AE, Jones JM, Everett L, Lenk GM, McKenna-Yasek DM, Weisman LS, et al. (2009). Deleterious variants of FIG4, a phosphoinositide phosphatase, in patients with ALS. Am J Hum Genet 84, 85–88. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Chow CY, Zhang Y, Dowling JJ, Jin N, Adamska M, Shiga K, Szigeti K, Shy ME, Li J, Zhang X, et al. (2007). Mutation of FIG4 causes neurodegeneration in the pale tremor mouse and patients with CMT4J. Nature 448, 68–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Choy CH, Saffi G, Gray MA, Wallace C, Dayam RM, Ou Z-YA, Lenk G, Puertollano R, Watkins SC, Botelho RJ (2018). Lysosome enlargement during inhibition of the lipid kinase PIKfyve proceeds through lysosome coalescence. J Cell Sci 131, jcs213587. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Cooke FT, Dove SK, McEwen RK, Painter G, Holmes AB, Hall MN, Michell RH, Parker PJ (1998). The stress-activated phosphatidylinositol 3-phosphate 5-kinase Fab1p is essential for vacuole function in S. cerevisiae. Curr Biol 8, 1219–S2. [DOI] [PubMed] [Google Scholar]
  15. Dong X, Shen D, Wang X, Dawson T, Li X, Zhang Q, Cheng X, Zhang Y, Weisman LS, Delling M, Xu H (2010). PI(3,5)P2 controls membrane trafficking by direct activation of mucolipin Ca2+ release channels in the endolysosome. Nat Commun 1, 38. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Dove SK, Cooke FT, Douglas MR, Sayers LG, Parker PJ, Michell RH (1997). Osmotic stress activates phosphatidylinositol-3,5-bisphosphate synthesis. Nature 390, 187–192. [DOI] [PubMed] [Google Scholar]
  17. Dove SK, Dong K, Kobayashi T, Williams FK, Michell RH (2009). Phosphatidylinositol 3,5-bisphosphate and Fab1p/PIKfyve underPPIn endo-lysosome function. Biochem J 419, 1–13. [DOI] [PubMed] [Google Scholar]
  18. Doyle CP, Rectenwald A, Timple L, Hammond GRV (2024). Orthogonal targeting of SAC1 to mitochondria implicates ORP2 as a major player in PM PI4P turnover. Contact 7, 25152564241229272. [DOI] [PMC free article] [PubMed] [Google Scholar]
  19. Freeman SA, Uderhardt S, Saric A, Collins RF, Buckley CM, Mylvaganam S, Boroumand P, Plumb J, Germain RN, Ren D, Grinstein S (2020). Lipid-gated monovalent ion fluxes regulate endocytic traffic and support immune surveillance. Science 367, 301–305. [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Gary JD, Wurmser AE, Bonangelino CJ, Weisman LS, Emr SD (1998). Fab1p is essential for PtdIns(3)P 5-kinase activity and the maintenance of vacuolar size and membrane homeostasis. J Cell Biol 143, 65–79. [DOI] [PMC free article] [PubMed] [Google Scholar]
  21. Giridharan SSP, Luo G, Rivero-Rios P, Steinfeld N, Tronchere H, Singla A, Burstein E, Billadeau DD, Sutton MA, Weisman LS (2022). Lipid kinases VPS34 and PIKfyve coordinate a phosphoinositide cascade to regulate retriever-mediated recycling on endosomes. eLife 11, e69709. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Goulden BD, Pacheco J, Dull A, Zewe JP, Deiters A, Hammond GRV (2019). A high-avidity biosensor reveals plasma membrane PI(3,4)P2 is predominantly a class I PI3K signaling product. J Cell Biol 218, 1066–1079. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Hammond GRV, Machner MP, Balla T (2014). A novel probe for phosphatidylinositol 4-phosphate reveals multiple pools beyond the Golgi. J Cell Biol 205, 113–126. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Hammond GRV, Takasuga S, Sasaki T, Balla T (2015). The ML1Nx2 phosphatidylinositol 3,5-bisphosphate probe shows poor selectivity in cells. PLoS One 10, e0139957. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Hirschi M, Jr, MAH, Wie J, Suo Y, Borschel WF, Ren D, Lander GC, Lee S-Y (2017). Cryo-electron microscopy structure of the lysosomal calcium-permeable channel TRPML3. Nature 550, 411–414. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Holmes VL, Ricci MMC, Weckerly CC, Worcester M, Hammond GRV (2025). Single-molecule lipid biosensors mitigate inhibition of endogenous effector proteins. J Cell Biol 224, e202412026. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Hsieh T-S, Lopez VA, Black MH, Osinski A, Pawłowski K, Tomchick DR, Liou J, Tagliabracci VS (2021). Dynamic remodeling of host membranes by self-organizing bacterial effectors. Science 372, 935–941. [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Ikonomov OC, Sbrissa D, Delvecchio K, Xie Y, Jin J-P, Rappolee D, Shisheva A (2011). The phosphoinositide kinase PIKfyve is vital in early embryonic development: preimplantation lethality of PIKfyve−/− embryos but normality of PIKfyve+/− mice*. J Biol Chem 286, 13404–13413. [DOI] [PMC free article] [PubMed] [Google Scholar]
  29. Ikonomov OC, Sbrissa D, Mlak K, Kanzaki M, Pessin J, Shisheva A (2002). Functional dissection of lipid and protein kinase signals of PIKfyve reveals the role of PtdIns 3,5-P2 production for endomembrane integrity*. J Biol Chem 277, 9206–9211. [DOI] [PubMed] [Google Scholar]
  30. Ikonomov OC, Sbrissa D, Shisheva A (2001). Mammalian cell morphology and endocytic membrane homeostasis require enzymatically active phosphoinositide 5-kinase PIKfyve*. J Biol Chem 276, 26141–26147. [DOI] [PubMed] [Google Scholar]
  31. Ikonomov OC, Sbrissa D, Shisheva A (2019). Small molecule PIKfyve inhibitors as cancer therapeutics: translational promises and limitations. Toxicol Appl Pharmacol 383, 114771. [DOI] [PubMed] [Google Scholar]
  32. Jefferies HBJ, Cooke FT, Jat P, Boucheron C, Koizumi T, Hayakawa M, Kaizawa H, Ohishi T, Workman P, Waterfield MD, Parker PJ (2008). A selective PIKfyve inhibitor blocks PtdIns(3,5)P2 production and disrupts endomembrane transport and retroviral budding. EMBO Rep 9, 164–170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Jeffries TR, Dove SK, Michell RH, Parker PJ (2004). PtdIns-specific MPR pathway association of a novel WD40 repeat protein, WIPI49. Mol Biol Cell. 15, 2652–2663. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Lees JA, Li P, Kumar N, Weisman LS, Reinisch KM (2020). Insights into lysosomal PI(3,5)P2 homeostasis from a structural-biochemical analysis of the PIKfyve lipid kinase complex. Mol Cell 80, 736–743.e4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  35. Leray X, Hilton JK, Nwangwu K, Becerril A, Mikusevic V, Fitzgerald G, Amin A, Weston MR, Mindell JA (2022). Tonic inhibition of the chloride/proton antiporter ClC-7 by PI(3,5)P2 is crucial for lysosomal pH maintenance. eLife 11, e74136. [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Li X, Wang X, Zhang X, Zhao M, Tsang WL, Zhang Y, Yau RGW, Weisman LS, Xu H (2013). Genetically encoded fluorescent probe to visualize intracellular phosphatidylinositol 3,5-bisphosphate localization and dynamics. Proc Natl Acad Sci USA 110, 21165–21170. [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Maib H, Adarska P, Hunton R, Vines JH, Strutt D, Bottanelli F, Murray DH (2024). Recombinant biosensors for multiplex and super-resolution imaging of phosphoinositides. J Cell Biol 223, e202310095. [DOI] [PMC free article] [PubMed] [Google Scholar]
  38. McCartney AJ, Zolov SN, Kauffman EJ, Zhang Y, Strunk BS, Weisman LS, Sutton MA (2014). Activity-dependent PI(3,5)P2 synthesis controls AMPA receptor trafficking during synaptic depression. Proc Natl Acad Sci USA 111, E4896–E4905. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Nicoli E-R, Weston MR, Hackbarth M, Becerril A, Larson A, Zein WM, Baker PR, Burke JD, Dorward H, Davids M, et al. (2019). Lysosomal storage and albinism due to effects of a de novo CLCN7 variant on lysosomal acidification. Am J Hum Genet 104, 1127–1138. [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Nishimura T, Tsuboyama K, Nakagaki Y, Yamamoto E, Mizushima N (2025). Rational engineering of lipid-binding probes via high-throughput protein-lipid interaction screening. Biorxiv 10.1101/2025.01.05.627504. [DOI] [Google Scholar]
  41. Pemberton JG, Barlow-Busch I, Jenkins ML, Parson MAH, Sarnyai F, Bektas SN, Kim YJ, Heuser JE, Burke JE, Balla T (2025). An advanced toolset to manipulate and monitor subcellular phosphatidylinositol 3,5-bisphosphate. J Cell Biol 224, e202408158. [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Rivero-Ríos P, Weisman LS (2022). Roles of PIKfyve in multiple cellular pathways. Curr Opin Cell Biol 76, 102086. [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Rutherford AC, Traer C, Wassmer T, Pattni K, Bujny MV, Carlton JG, Stenmark H, Cullen PJ (2006). The mammalian phosphatidylinositol 3-phosphate 5-kinase (PIKfyve) regulates endosome-to-TGN retrograde transport. J Cell Sci 119, 3944–3957. [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Samie M, Wang X, Zhang X, Goschka A, Li X, Cheng X, Gregg E, Azar M, Zhuo Y, Garrity AG, et al. (2013). A TRP channel in the lysosome regulates large particle phagocytosis via focal exocytosis. Dev Cell 26, 511–524. [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Sbrissa D, Ikonomov OC, Shisheva A (1999). PIKfyve, a mammalian ortholog of yeast Fab1p lipid kinase, synthesizes 5-phosphoinositides. Effect of insulin*. J Biol Chem 274, 21589–21597. [DOI] [PubMed] [Google Scholar]
  46. Sbrissa D, Naisan G, Ikonomov OC, Shisheva A (2018). Apilimod, a candidate anticancer therapeutic, arrests not only PtdIns(3,5)P2 but also PtdIns5P synthesis by PIKfyve and induces bafilomycin A1-reversible aberrant endomembrane dilation. PLoS One 13, e0204532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  47. Schaletzky J, Dove SK, Short B, Lorenzo O, Clague MJ, Barr FA (2003). Phosphatidylinositol-5-phosphate activation and conserved substrate specificity of the myotubularin phosphatidylinositol 3-phosphatases. Curr Biol 13, 504–509. [DOI] [PubMed] [Google Scholar]
  48. Schindelin J, Arganda-Carreras I, Frise E, Kaynig V, Longair M, Pietzsch T, Preibisch S, Rueden C, Saalfeld S, Schmid B, et al. (2012). Fiji: an open-source platform for biological-image analysis. Nat Methods 9, 676–682. [DOI] [PMC free article] [PubMed] [Google Scholar]
  49. Schrecker M, Son Y, Planells-Cases R, Kar S, Vorobeva V, Schulte U, Fakler B, Jentsch TJ, Hite RK (2025). Structural basis of ClC-3 transporter inhibition by TMEM9 and PtdIns(3,5)P2. Nat Struct Mol Biol 1–8. [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Shisheva A, Rusin B, Ikonomov OC, DeMarco C, Sbrissa D (2001). Localization and insulin-regulated relocation of phosphoinositide 5-kinase PIKfyve in 3T3-L1 adipocytes*. J Biol Chem 276, 11859–11869. [DOI] [PubMed] [Google Scholar]
  51. Sonawane ND, Thiagarajah JR, Verkman AS (2002). Chloride concentration in endosomes measured using a ratioable fluorescent Cl− indicator evidence for chloride accumulation during acidification*. J Biol Chem 277, 5506–5513. [DOI] [PubMed] [Google Scholar]
  52. Steinfeld N, Giridharan SSP, Kauffman EJ, Weisman LS (2021). Phosphoinositides, methods and protocols. Methods Mol Biol 2251, 1–17. [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Taylor GS, Maehama T, Dixon JE (2000). Myotubularin, a protein tyrosine phosphatase mutated in myotubular myopathy, dephosphorylates the lipid second messenger, phosphatidylinositol 3-phosphate. Proc Natl Acad Sci USA 97, 8910–8915. [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Vines JH, Maib H, Buckley CM, Gueho A, Zhu Z, Soldati T, Murray DH, King JS (2023). A PI(3,5)P2 reporter reveals PIKfyve activity and dynamics on macropinosomes and phagosomes. J Cell Biol 222, e202209077. [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Wang X, Zhang X, Dong X, Samie M, Li X, Cheng X, Goschka A, Shen D, Zhou Y, Harlow J, et al. (2012). TPC proteins are phosphoinositide- activated sodium-selective ion channels in endosomes and lysosomes. Cell 151, 372–383. [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Wills RC, Goulden BD, Hammond GRV (2018). Genetically encoded lipid biosensors. Mol Biol Cell 29, 1526–1532. [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Wills RC, Pacheco J, Hammond GRV (2021). Quantification of genetically encoded lipid biosensors. Methods Mol Biol 2251, 55–72. [DOI] [PubMed] [Google Scholar]
  58. Yamamoto A, DeWald DB, Boronenkov IV, Anderson RA, Emr SD, Koshland D (1995). Novel PI(4)P 5-kinase homologue, Fab1p, essential for normal vacuole function and morphology in yeast. Mol Biol Cell. 6, 525–539. [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Yanisch-Perron C, Vieira J, Messing J (1985). Improved M13 phage cloning vectors and host strains: nucleotide sequences of the M13mpl8 and pUC19 vectors. Gene 33, 103–119. [DOI] [PubMed] [Google Scholar]
  60. Zolov SN, Bridges D, Zhang Y, Lee W-W, Riehle E, Verma R, Lenk GM, Converso-Baran K, Weide T, Albin RL, et al. (2012). In vivo, Pikfyve generates PI(3,5)P2, which serves as both a signaling lipid and the major precursor for PI5P. Proc Natl Acad Sci USA 109, 17472–17477. [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Molecular Biology of the Cell are provided here courtesy of American Society for Cell Biology

RESOURCES