Contacting domains that segregate lipid from solute transporters in malaria parasites

While membrane contact sites (MCS) between intracellular organelles are abundant1, and cell-cell junctions are classically defined2, very little is known about the contacts between membranes that delimit extracellular junctions within cells, such as those of chloroplasts and intracellular parasites. The malaria parasite replicates within a unique organelle, the parasitophorous vacuole (PV) but the mechanism(s) are obscure by which the limiting membrane of the PV, the parasitophorous vacuolar membrane (PVM), collaborates with the parasite plasma membrane (PPM) to support the transport of proteins, lipids, nutrients, and metabolites between the cytoplasm of the parasite and the cytoplasm of the host erythrocyte (RBC). Here, we demonstrate the existence of multiple micrometer-sized regions of especially close apposition between the PVM and the PPM. To determine if these contact sites are involved in any sort of transport, we localized the PVM nutrient-permeable and protein export channel EXP2, as well as the PPM lipid transporter PfNCR1. We found that EXP2 is excluded from, but PfNCR1 is included within these regions of close apposition. Thus, these two different transport systems handling hydrophilic and hydrophobic substances, respectively, assume complementary and exclusive distributions. This new structural and molecular data assigns a functional significance to a macroscopic membrane domain.


INTRODUCTION
Morbidity and mortality in malaria are due to the apicomplexan Plasmodium spp.
replicating within the host RBC. During its initial invasion of the erythrocyte, the parasite invaginates the RBC plasma membrane to form the PV as a second barrier 3,4 . The parasite must install its own, unique transport systems in order to import and export everything it needs for its survival and proliferation [5][6][7] . Understanding the underlying transport mechanisms between the parasite and its host is useful to identify drug targets. Yet, these crucial transport systems remain incompletely understood for both hydrophilic and hydrophobic substances.
A PVM channel, permeable to water-soluble nutrients like monosaccharides and amino acids 5 , is formed by the 'exported protein 2' or EXP2 8 . EXP2 also facilitates protein export, by serving as the protein-permeant pore for the 'Plasmodium translocon of exported proteins' (PTEX) 9 . A number of PPM channels, including several that are specific for particular nutrients have been identified and studied 10,11 . However, it is not known how lipid substances are transported across the PV, since the two limiting membranes have never been seen to connect or to transport membrane vesicles between each other 6 . The PPM resident protein 'Plasmodium Niemann-Pick C1-related protein' (PfNCR1) is essential for lipid homeostasis 12 but it is unknown how it functions. It is not clear how the PV is organized, in order to support the transport of such a large variety of substrates.
Here we hypothesize that the structure of the PV is built such that it can support direct exchange of lipids across the PV space, directly between the PPM and PVM, in regions that can be defined as membrane contact sites (MCS).

Regions of close PVM-PPM apposition exist
Thin-sections of parasitized red cells were examined in the electron microscope (EM) to determine the separation distance between PVM and PPM. Since transport across the PV is most active in the trophozoite stage, i.e. the stage when parasites have begun to accumulate hemozoin and grow the fastest but have not begun to divide 13 , only this stage was considered. An example image is shown in ( Figure 1A). The distribution of separation distances between PVM and PPM in such organisms was found to be bimodal, indicating two distinct structural regions: regions of "close membrane apposition", separated by ~10 nm, and regions of "PV lumen" with wider membrane separation distances of 20-30 nm. To control for possible artifacts introduced by chemical fixation, infected red cells were also prepared by a quick-freeze, freeze-fracture method that preserved cells in their most lifelike state. Still, these two distinct types of regions could be clearly discerned ( Figure 1B). In conclusion, these two distinct differences in separation are narrow enough that they could be bridged by protein complexes that could interconnect the PVM and PPM membranes, and thus are candidates for membrane contact sites 1 .

EXP2 localizes in domains of the PVM
A functionally characterized marker of the PVM, EXP2 8 , was used to investigate the domain structure of the PVM. To test domain formation in fixed but label-free, unmodified parasites (NF54), an immunofluorescence assay using EXP2 antibodies was performed. EXP2 was found in patches, interrupted by stretches of PVM devoid of anti-EXP2 label ( Figure 2A).
To test if these regions of EXP2 were domains in living parasites and to control for chemical fixation artefacts, a parasite line bearing a C-terminal mNeonGreen (mNG) fusion to the endogenous copy of EXP2 (EXP2-mNG) was examined 14 . In these parasites, the mNG signal also was found in continuous patches, interrupted by stretches of PVM devoid of mNG ( Figure   2B). To quantify and visualize these apparent domains of EXP2, we developed a simple tool: a projection of the maximum fluorescence intensity from the inside of the parasite onto a sphere.
This results in a map of the fluorescence signal of the periphery of the parasite (Figure 2 C-E).
Both samples show a protein domain coverage on the order of 50% around the parasite ( Figure   2F).
Two physically independent techniques to visualize EXP2 spatial organization indicate its domain structure. Thus, EXP2-mNG can serve as a robust and readily detected marker for such PVM domains, labeling regions of protein export and transport of small water-soluble molecules.

EXP2 co-localizes with PV lumen
The EM-findings of two regions with distinct PVM-PPM separation-distances and the light microscopy findings of domains for protein export and nutrient import represented by EXP2 led us to hypothesize that these features could be correlated.
To determine the distribution of the EXP2-domains relative to PV domains with distinct lumenal space, their localization was carefully mapped. Thus, a fluorescent label for the PV lumen was required that could be compared with the distribution of EXP2-mNG. To this end, mRuby3 was targeted to the PV lumen using the signal peptide of HSP101 (PV-mRuby3) 15 . To verify that PV-mRuby3 would serve as a genuine label of the PV lumen, correlative light/electron microscopy after cryo-thin sectioning was performed 16 . PV-mRuby3 was indeed found to localize to the regions of wider separation between PVM and PPM, confirming that it would serve a label for the accessible PV space ( Figure 3A).
Co-expression and two-color imaging of EXP2-mNG and PV-mRuby3 demonstrated a clear-cut colocalization of both labels around the periphery of the parasite ( Figure 3B). The mean Pearson-correlation coefficient for the analyzed sample is 0.81 CI [0.78, 0.83] ( Figure 3C) as a mathematical measure for the degree of signal overlap, with numbers from -1 (perfectly anticorrelated), 0 (not correlated) to 1 (perfectly correlated).
Thus, EXP2 distribution can be correlated with regions of the wide mRuby3 accessible lumen exists. Within the limits of optical microscopy, they colocalize. Therefore, protein export, nutrient import, and presumably aqueous waste export occurs in the regions of the PV lumen.

The lipid transporter PfNCR1 anti-localizes with EXP2 and the PV lumen
Sites of close membrane contact are implicated in direct transfer of lipids via intervening or included proteins that localize to those domains 1 . In the PPM PfNCR1 has been found to be essential for the maintenance of lipid homeostasis 12 . However, its human homolog, Niemann-Pick C1 (hNPC1) relies on a co-factor hNPC2, to transport lipids 17 ; no such co-factor has been identified in Plasmodium spp., suggesting that PfNCR1 may work differently. Thus, localizing PfNCR1 with respect to the separation-distances of the membranes of the PV is indicative in order to determine if PfNCR1 can be expected to transport lipids by interacting with a soluble cofactor from the PV lumen, or whether it interacts directly with the PVM at sites of close membrane apposition.
To determine the distribution of PfNCR1 relative to the domains defined by EXP2 and the PV lumen, an endogenous EXP2-mRuby3 fusion was engineered into a parasite expressing an endogenous PfNCR1-GFP fusion protein, allowing both proteins to be monitored by live fluorescence while preserving their native timing and expression levels. Two-color imaging showed that localization of the two labels is anti-correlated around the periphery of the parasite ( Figure 4A). The mean Pearson-correlation coefficient of this sample is -0.20 CI [-0.10, -0.29], giving a mathematical measure for the anti-correlation of both signals ( Figure 4B). Positive values of the coefficient are caused by small domains, approaching the resolution limit of light microscopy with signals co-localizing at the border of domains ( Figure 4A). Anti-correlation was also observed in parasites where the PV-mRuby3 lumenal reporter was expressed in the PfNCR1-GFP background ( Figure SI 2). From this observed anti-localization of the lumenal vs.
hydrophobic transporter labels, it is most likely that PfNCR1 is localized to regions of close membrane apposition.

DISCUSSION
In studying how the malaria parasite modifies its protective membrane barriers to import and export materials that it needs to survive and grow, we found that the host cell-parasite interface (HPI) is a unique intercellular junction with clearly definable regions of protein composition and variable separation, consisting of a parasite vacuolar membrane, the lumen of the vacuole (that is the parasite's extracellular space), and the plasma membrane of the parasite.
Different classes of molecules, hydrophobic and hydrophilic, are transported through this one continuous, spheroidal interface with distinct regions. This segregation of function is accomplished by the creation of membrane contact sites (MCS) like those between cellular organelles 18,19 . Electron microscopy resolved the PVM and the PPM, allowing quantification of their separation-distances, and confirming the visual impression that the PVM and the PPM form distinct domains, characterized by a bimodal distribution of space between the two membranes.
Using two complimentary techniques, indirect immunofluorescence and live-cell microscopy of fluorescently tagged EXP2, this solute-transporter was detected in µm-sized domains that correlate spatially with domains where the PVM and the PPM are separated enough from each other for the vacuole to accumulate a visible amount of the lumenal marker PV-mRuby3. In contrast, we found that the parasite lipid transporter PfNCR1 was specifically excluded from these regions and instead accumulated in the intervening regions of close PVM-PPM apposition.
We conclude that the PV has evolved to become laterally segregated into relatively wide regions for hydrophilic transport, and separate close-contact regions for hydrophobic transport.
The distances between the PPM and PVM for both domains could be bridged by proteins, potentially qualifying both regions as MCS 1 . However, in neither domain of the HPI were bridging proteins observed in our "deep-etch" EMs, nor were they observed in our thin-section EMs. However, when the PV is induced to swell experimentally, e.g., when PVM protein export is conditionally impaired, leading to protein accumulation 8,20,21 , it expands inhomogenously into irregular protuberances, suggesting that some sort of adhesion normally exists between the PPM and the PVM that prevents it from swelling uniformly. Still to be determined is whether these adhesions concentrate in the tightly apposed regions or the more open regions. EXP2-mNG is included in the distended regions of the PVM when protein export is impaired ( Figure SI 3), indicating that the EXP2-containing domains form the less strongly connected region.
In contrast, and in keeping with the detrimental effect of water on lipid transport, sites of close PVM-PPM apposition seem to be devoid of PV lumen altogether. This is an extremely close apposition, at the lower end of the membrane distances found at organelle-organelle MSC, i.e. 10-80 nm 1 . PfNCR1, a lipid transporter, localizes to these sites of unusually close membrane apposition. While it can be inferred that lipid transport is taking place at these sites, it is unclear how PfNCR1 goes about exchanging its substrate, presumably cholesterol, with the PVM. It has been demonstrated that the large extra-membranous domain of PfNCR1 is localized in between PPM and PVM. The closely apposed membranes are within the proteins hypothesized radius 12 , so it may well hand over lipids directly 22 , or exchange lipids with the help of a membrane-bound cofactor. PfNCR1 may also function as an anchor between these two membranes, given the relatively constant separation between them at the close contacts. Curiously, the domain structure of the PVM exemplified by the EXP2 distribution shown here, is quite dynamic and variable (cf., SI movie, demonstrating remarkable flexibility in the PV), suggesting there may exist an active mechanism of protein localization driven by processes in the parasite cytoplasm. Additionally, proteins may target to their respective regions by various other mechanisms, such as interacting with structural proteins, sensing of membrane distance, as found with other contact sites 1 , or protein exclusion, as found at gap junctions 2 .
The structure-function relationship described here ( Figure 4C) can potentially guide the study of other functions at the HPI, such as the inhomogeneously distributed PVM proteins noted previously 23,24 . A mechanism for the transport of resident membrane proteins from the PPM to the PVM is still lacking as PTEX is not involved in this process 25,26 . The regions of close apposition are a promising place to look for machinery that would allow transfer of such proteins.
The size of these domains are similar to those created in lipid phase demixing seen in model membranes 27 . However, cellular membranes are complex mixtures that for the most part fail to demix at physiological temperatures and remain almost entirely liquid-disordered 28 .
Moreover, cholesterol-rich domains are not detected by surface imaging using mass spectroscopy 29 . However, molecular dynamic simulations have revealed nanoscopic microdomains of hexagonally packed saturated lipids 30 that may correspond to the nonideal demixing revealed by FRET 31 . Recently a fungal vacuole was reported to exhibit temperaturedependent lipid phase demixing 32 but no function has yet been ascribed to these domains. The specific role of lipid asymmetry and composition, and its role in domain formation and maintenance, in the face of lipid transport, remain to be investigated for any MCS or membrane domain. The HPI offers a larger platform for these studies that may benefit both cell biological and medical investigations.

CODE AVAILABILITY
MATLAB (Mathworks) scripts as described in the Methods are available from the corresponding authors on request.

DATA AVAILABILITY
The data that support the findings of this study are available from the corresponding authors on request.

Molecular Biology
For generation of an endogenous EXP2-mRuby3 fusion, the blasticidin-S deaminase (BSD) cassette between SalI and BglII in the plasmid pGDB 34  To generate an endogenous EXP2-mNeonGreen fusion in the HSP101 DDD background, the plasmid pyPM2GT-EXP2-mNeonGreen was co-transfected with pAIO-EXP2-CT-gRNA into NF54attB: HSP101 DDD 8 . Selection was applied with 2µM DSM1 24 hours post transfection and parasites were cloned by limiting dilution when they returned from selection.

Immunofluorescence assay (IFA)
IFAs were performed as described previously 8

Light microscopy
Images were obtained on a Zeiss 880 with Airyscan module using a 63x 1.4NA Zeiss Plan-Apochromat, 37 °C immersion oil (Zeiss, Oberkochen, DE). Images were collected using Zen black (Zeiss) in the Airyscan mode, following the programs recommendation for the optimal pixel size and slice thickness, pixel dwell times were kept at 1-2 µs. Live parasites were transferred into a hybridization chamber (HybriWell HBW20, Grace Bio-Labs, Bend, OR) for observation on the microscope 15  Analysis was done using custom scripts in MATLAB 8.5 (MathWorks, Natick MA). Briefly, the center of the cell was determined from the center of mass of a mask created from pixels in between the 1 st and the 2 nd level of a 2-level threshold using the "multithresh" function. Each voxel was then assigned an altitude and azimuth with respect to this center using the "acos" and "atan2" functions respectively. The altitude and azimuth were then used to create a mask with a 1° resolution (equivalent to 17.5 nm at a distance of 1 µm from the center). The maximum of masked intensity was then recorded as angular intensity value for the altitude and azimuth.
The 95% confidence interval of the correlation coefficient was calculated in Prism (Graphpad, San Diego, CA) after a Fisher transform of the data, the result was then back transformed.
The source code of the scripts used will be made available through the authors and online repository.

Electron Microscopy on thin sections
Erythrocyte cultures infected with NF54attb parasites were enriched for late stages using a LD where the membranes were cut obliquely were not considered. The minimal distance of the two membrane traces was determined using a MATLAB script.

Correlative Light Electron Microscopy
EXP2-mNeonGreen-PV-mRuby3 parasites were isolated on a 65% percoll (MilliporeSigma) interface. Cells were fixed using 4 % formaldehyde + 0.4 % glutaraldehyde in 1x PHEM (pH 6.9) (Electron microscopy sciences) 38  direct magnification (7.4 nm pixel size) were used for the alignment to the light microscopy image (pixel size following the "optimal" pixel size settings of the Airyscan module is 42.6 nm).

Freeze Fracture Replica
Late stage NF54attb infected red blood cells were isolated using a magnetic separator as for the thin sections. Cell were gently pelleted at 1000 RPM in a clinical centrifuge, then layered as a thick slurry on a tiny 3x3mm class coverslip mounted on a lung cushion, in preparation for quick-freezing with the liquid helium cooled copper-block "slammer" 42 . Thereafter, they were transferred to a Balzers 400 freeze-fracture apparatus, where they were fractured through their well-frozen surfaces, "deep-etched" for 2 min at -104 °C, and then rotary-replicated with 4 nm of platinum deposited from a 20° angle. Thereafter they were "backed" with 10 nm of carbon deposited from 90°, removed from the Balzers, thawed, and the platinum replica was floated on 25% SDS to partially remove organic material from underneath it 43 . After washing, the replica was picked up on an EM grid. Thereafter, these replicas were examined in the electron microscope in the same manner as the thin sections, above.

Figure SI 3. EXP2 is targeted to distensions when protein export is impaired
Compared to the broken but smooth outline of EXP2-mNeonGreen in unmodified parasites (see figure 2) the outline after impairing protein export by inactivating HSP101 using a (DHFR)based destabilization domain (DDD) leads to the appearance of a distended signal. This suggests that EXP2 is present in the distended membrane reported in 8 . Green: Exp2-mNeonGreen, gray: brightfield, scale bar: 5 µm. Trimethoprim was withdrawn 24 hours before imaging. Image was acquired on an Axio Imager M1 equipped with a Plan-Apochomat 100x/1.4NA objective and ORCA-ER CCD camera, using the GFP filterset for the mNeonGreen imaging.

Figure SI movie. Dynamics of relatively small EXP2 domains.
Time course of Z-projection of a stack of the PVM face closest to the cover slide. Green: EXP-mNG. Scale bar: 1 µm.