The P. falciparum strain 3D7 was cultured according to modified standard procedures (Trager and Jensen, 1976) at 5% hematocrit using human 0 Rh+ erythrocytes from the University Medical Center Hamburg-Eppendorf (UKE), Germany, at 1% O₂, 5% CO₂ and 94% N₂. RPMI 1640 medium was supplemented with 0.5% (w/v) AlbuMAX II, 20 µg/mL gentamicin and 100 µM hypoxanthine (Thermo Fisher Scientific). Mature schizonts were obtained by treating schizonts at 40 hours post invasion (hpi) with 1 mM compound 2 (4-[7-[(dimethylamino)methyl]-2-(4-fluorphenyl)imidazo[1,2-α]pyridine-3-yl]pyrimidin-2-amine, LifeArc) for 8 h. To count the number of merozoites per mature schizont, Giemsa-stained blood smears were analyzed by light microscopy. Only single-infected cells with one digestive vacuole were taken into account.

To assess parasite proliferation over six days, a previously described assay on the basis of flow cytometry was employed (Malleret et al., 2011). Parasites were synchronized to a 3-h age window by isolating late schizonts from a 60% Percoll (GE Healthcare) gradient and culturing these for 3 h with fresh erythrocytes (Rivadeneira et al., 1983), followed by controlled elimination of advanced parasite stages using 5% (w/v) D-sorbitol (Carl Roth) for 10 min at 37°C (Lambros and Vanderberg, 1979). The growth assay was started at 0.1% parasitemia using the resulting ring-stage parasites at 0 – 3 hpi. The parasitemia was determined at the trophozoite stage every two days by flow cytometry and culture media with the respective supplements were exchanged daily.

To determine parasitemia, 20 µL of resuspended parasite culture was added to 80 μL culture medium and stained with 5 μg/mL SYBR Green I (Thermo Fisher Scientific) and 4.5 μg/mL dihydroethidium (DHE, Sigma-Aldrich) in the dark for 20 min at room temperature. Stained cells were washed with PBS three times and analyzed with an ACEA NovoCyte flow cytometer and NovoExpress Software (version 1.6.1, Agilent). Forward and side scatter gating was used to identify erythrocytes and SYBR Green I fluorescence intensity to determine the number of parasitized cells per 100,000 events recorded for each replicate. For Phen Green SK measurements, uninfected erythrocytes were washed with PBS and incubated with 10 µM Phen Green SK in PBS at 37°C for 60 min. DHE at 4.5 μg/mL was added during the last 20 min of incubation. After three washes with PBS, the cells were analyzed as described above.

Parasites were synchronized to a three-hour window after invasion of erythrocytes from the respective donor as described above. Samples for RNA-sequencing were prepared in triplicate for each condition and time point, i.e., three separate parasite cultures each were grown in parallel for a total of at least two weeks. During the second IDC, two 10-mL dishes each were harvested for parasites at the ring stage (6 – 9 hpi) and one 10-mL dish each for trophozoites (26 – 29 hpi). Samples were collected by centrifuging the culture for 5 min at 800 g and 37°C and dissolving the erythrocyte pellet using 5 mL TRIzol (ThermoFisher Scientific) prewarmed to 37°C, followed by immediate transfer to -80°C for storage. The parasitemia was 0.3% at the start of the experiments with high, control and low-iron donor blood and 2 – 3% at the time of harvest. Parasite cultures treated with 0.7 µM hepcidin (Bachem) had a starting parasitemia of 0.6% and untreated cultures 1% to reach a parasitemia of 4 – 5% during the second cycle. For each experiment, the parasitemia was kept consistent at the point of harvest as high parasite densities can affect transcription (Chou et al., 2018).

For RNA extraction, the samples frozen in TRIzol were thawed, mixed thoroughly with 0.1 volume cold chloroform, and incubated at room temperature for 3 min. Following centrifugation at 20,000 g and 4°C for 30 min, the supernatants were transferred to fresh vials and combined with 70% ethanol of equal volume. RNA was purified using the RNeasy MinElute Kit (Qiagen) by on-column DNase I digest for 30 min and elution with 14 µL water. The GLOBINclear Human Kit (ThermoFisher Scientific) was then employed to deplete human globin mRNA in all samples. The Qubit RNA HS Assay Kit and Qubit 3.0 fluorometer (ThermoFisher Scientific) were used for RNA quantification. Upon arrival at the EMBL Genomics Core Facility (GeneCore Heidelberg, Germany), the RNA quality of each sample was evaluated using the RNA 6000 Nano kit and Bioanalyzer 2100 (Agilent). The median RNA integrity number (RIN) of all samples was 7.30 (IQR: 6.85 – 8.15, Supplementary Figure S1 ). Individually barcoded strand-specific libraries for mRNA sequencing were prepared from total RNA samples of high quality (approximately 150 ng per sample) using the NEBNext^® RNA Ultra II Directional RNA Library Prep Kit (New England Biolabs) for 12 PCR cycles on the liquid handler Biomek i7 (Beckman Coulter) at GeneCore. Libraries that passed quality control were pooled in equimolar amounts, and a 2 pM solution of this pool was sequenced unidirectionally on a NextSeq^® 500 System (Illumina) at GeneCore, resulting in about 500 million reads of 85 bases each.

Following successful initial quality control of the RNA-sequencing reads with FastQC version 0.11.8 (Andrews, 2010), sequencing adapters were trimmed using Cutadapt version 2.10 (Martin, 2011). A genome index was generated using the FASTA sequence file of the P. falciparum 3D7 genome release 46 (PlasmoDB-46_Pfalciparum3D7_Genome.fasta) and the GFF3 annotation file (PlasmoDB-46_Pfalciparum3D7.gff), both obtained from PlasmoDB (Aurrecoechea et al., 2009), with STAR version 2.7.5c (Dobin et al., 2013). The same R package was used to align reads to the genome with a maximum of three allowed mismatches (–outFilterMismatchNmax 3). To consolidate the results obtained with FastQC and STAR alignments, a single report file was created using MultiQC version 1.9 (Ewels et al., 2016).

The mapped reads were then summarized in Sequence Alignment/Map (SAM) format using featureCounts (Liao et al., 2014) from the R package Rsubread version 2.2.1 (Liao et al., 2013). For counting mapped reads per gene using featureCounts, fragments with a minimum length of 50 bases were considered (minFragLength = 50). Therefore, gene IDs and lengths of transcripts were extracted from PlasmoDB-46_Pfalciparum3D7_AnnotatedTranscripts.fasta with SAMtools faidx version 1.10.2 (Li et al., 2009). The R package edgeR 3.30.3 (Robinson et al., 2010) was used to compute RPKM values (reads per kilobase per million mapped reads) and for differential gene expression analysis. Gene annotations were retrieved from PlasmoDB (Aurrecoechea et al., 2009) and PhenoPlasm (Sanderson and Rayner, 2017). The results of these analyses were visualized with volcano plots using the R package Enhanced Volcano version 1.15.0 (Blighe et al., 2022). The raw and processed data (FASTA files, RPKM values and results of the differential gene expression analysis) can be accessed at https://www.ebi.ac.uk/biostudies/studies/E-MTAB-13411.

The highly polymorphic var, stevor, and rifin gene families were excluded from downstream analyses because of their great sequence diversity between parasites of the same strain during mitotic growth (Bozdech et al., 2003; Kidgell et al., 2006). Genes that were significantly regulated (defined as P < 0.05 according to the exact test for the negative binomial distribution with Benjamini-Hochberg correction (Benjamini and Hochberg, 1995) and an absolute value of log₂ FC ≥ 0.2) were subjected to functional enrichment analysis with g:Profiler (https://biit.cs.ut.ee/gprofiler/gost (Raudvere et al., 2019), accessed on August 17, 2022). The resulting GO, KEGG and REAC terms were summarized using REVIGO (http://revigo.irb.hr/) with the similarity value set to 0.5 (Supek et al., 2011) and visualized as in Thomson-Luque et al (Thomson-Luque et al., 2021) using the scientific color map “roma” (Crameri et al., 2020). To estimate parasite age, an algorithm developed by Avi Feller and Jacob Lemieux (Lemieux et al., 2009) was adapted to use expression data from Broadbent et al (Broadbent et al., 2015) with the time points 6, 14, 20, 24, 28, 32, 36, 40, 44, and 48 hpi as reference. The code and data used were deposited to Zenodo with the record ID 7996302 (https://zenodo.org/record/7996302).

For generating the GFP reporter lines, a homologous region of approximately 800 base pairs (bp) at the 3’ end of the respective gene was amplified without the stop codon from 3D7 gDNA using Phusion high fidelity DNA polymerase (New England Biolabs). A homology region of about 400 bp at the 5’ end of the respective gene was used for targeted gene disruption. The fragments were then inserted into pSLI-GFP (Birnbaum et al., 2017) using NotI and AvrII restriction sites. For glmS constructs, pSLI-GFP-glmS (Burda et al., 2020) was used as a vector instead. All oligonucleotides and plasmids used in this study are listed in Supplementary Table S3 .

As described previously (Moon et al., 2013), parasites at the late schizont stage were purified using 60% Percoll (Rivadeneira et al., 1983) and electroporated with 50 μg DNA of the respective plasmid in a 0.2-cm gap cuvette (Bio-Rad Laboratories) using Amaxa Nucleofector 2b (Lonza). Either 4 nM WR99210 (Jacobus Pharmaceuticals) or 2 μg/mL blasticidin S (Life Technologies) was used for selecting transfectants. For the selection of parasites that were genomically modified using the SLI system (Birnbaum et al., 2017), 400 μg/mL G418 (ThermoFisher Scientific) was added to the culture medium once the parasitemia reached 5%. After the selection of modified parasites, genomic DNA was isolated with the QIAamp DNA Mini Kit (Qiagen) and diagnostic tests for correct integration into the genome were performed as specified earlier (Birnbaum et al., 2017).

Erythrocytes infected with parasites at different stages at 3 – 6% parasitemia were incubated in culture medium with 20 nM MitoTracker Red, 200 nM ER Tracker Red or 100 nM LysoTracker Deep Red (Invitrogen, if applicable) at 37°C for 20 min. Then, 200 nM Hoechst-33342 (Invitrogen) was added for 10 min prior to washing the cells with Ringer’s solution (122.5 mM NaCl, 5.4 mM KCl, 1.2 mM CaCl₂, 0.8 mM MgCl₂, 11 mM D-glucose, 25 mM HEPES, 1 mM NaH₂PO₄, pH 7.4) prewarmed to 37°C and seeding on a chambered No. 1.5 polymer cover slip (Ibidi). After 5 min, unbound erythrocytes were removed by washing with Ringer’s solution and the sample was placed into an incubation chamber that maintained the microscope work area including the objective at 37°C. Images and videos were acquired using an SP8 confocal microscope system with a 63x oil-corrected lens (C-Apochromat, numerical aperture = 1.4) and Lightning deconvolution software (Leica), and processed using ImageJ version 2.9.0/1.53t (Schindelin et al., 2012). If fluorescence intensities were to be quantified, no averaging or deconvolution software was applied.

For glmS-based knockdown induction (Prommana et al., 2013), highly synchronous parasites at the early ring stage were cultured with or without supplementation with 2.5 mM glucosamine (GlcN, Sigma-Aldrich). The knockdown was quantified by confocal live-cell microscopy using schizonts 36 h post GlcN treatment initiation. Images of parasites of similar size were acquired with the same settings and background-corrected fluorescence intensities (integrated density) as well as the size of the region of interest were determined using ImageJ version 2.9.0/1.53t (Schindelin et al., 2012), and the data visualized using Graph Pad Prism version 9.4.1.

Structure predictions for monomeric proteins were obtained from AlphaFold Protein Structure Database version 3 (Jumper et al., 2021; Varadi et al., 2022) and homodimeric proteins were predicted using AlphaFold2-multimer version 2.2.2, database version 2.2.0 (Evans et al., 2022) deployed at the EMBL Hamburg computer cluster. Molecular visualization was performed with UCSF ChimeraX version 1.3 (Goddard et al., 2018). UCSF Chimera MatchMaker and Match → Align tools with default settings were used for structural comparison of the predicted structures of P. falciparum proteins with putative orthologs and sequence alignments were generated using the Match → Align tool (Meng et al., 2006). The DeepFRI server (https://beta.deepfri.flatironinstitute.org) was used to identify possible functional residues with the DeepFRI graph convolutional network (Gligorijević et al., 2021).

To investigate whether labile iron levels in the erythrocyte correlate with parasite replication rates, we established different iron conditions in vitro. The first approach was to culture P. falciparum 3D7 parasites in 0 Rh+ erythrocytes from voluntary blood donations by Caucasians aged 18 to 21 at the University Medical Center Hamburg-Eppendorf in Germany. Therefore, samples from a person with an elevated ferritin level (greater than 200 µg/L (Marfil-Rivera, 2015), in this case 231 µg/L ferritin, 18.2 g/dL Hb, 51.5% hematocrit), an iron-deficient individual (serum ferritin < 12 ng/ml (Knovich et al., 2009), here: 3 µg/L ferritin, 11.4 g/dL Hb, 36.3% hematocrit) and a healthy donor (21 µg/L ferritin, 15.0 g/dL Hb, 42.3% hematocrit) were used. Secondly, infection of red blood cells from other healthy individuals with or without the addition of 0.7 µM hepcidin to the culture medium was compared. This concentration was chosen as it had the strongest effect on parasite proliferation in preliminary experiments, and is expected to increase intracellular Fe²⁺ as it is twice as high as the hepcidin level needed to reduce ⁵⁵Fe export from preloaded mature erythrocytes by 30% within one hour of incubation (Zhang et al., 2018).

Relative labile iron levels in uninfected erythrocytes were estimated by determining the mean fluorescence intensity (MFI) of the iron-sensitive dye Phen Green SK in 100,000 cells per replicate using flow cytometry. As binding of ferrous iron to the metal-binding moiety causes fluorescence quenching of the fluorophore, a reduction in fluorescence intensity indicates higher labile iron levels (Petrat et al., 1999). In erythrocytes from the iron-deficient donor, the Phen Green SK MFI was 43% higher relative to control, confirming reduced labile iron levels ( Figure 1A ). The parasite replication rate after one intraerythrocytic developmental cycle (IDC) decreased by 16% ( Figure 1B ), the DNA content of late schizonts by 19% ( Figure 1C ) and the number of merozoites counted per late schizont by 14% ( Figure 1D ). In contrast, labile iron levels of erythrocytes from the donor with higher iron status were only slightly increased (without statistical support, two-tailed unpaired t tests with Welch’s correction for unequal variances and adjusted with the Holm-Šídák method for multiple comparisons, P = 0.25) relative to blood with normal iron level (healthy control) – as were the parasite proliferation rate, the DNA content and the merozoite number of mature schizonts ( Figure 1 ). To further increase intracellular labile iron levels, we incubated parasites with 0.7 µM hepcidin during one IDC, resulting in 11% reduced Phen Green SK MFI compared to control ( Figure 1A ). Under these conditions, the parasite growth rate increased by 57% ( Figure 1B ), the DNA content per schizont by 16% ( Figure 1C ), and the number of merozoites produced per schizont by 15% ( Figure 1D ).

Taken together, these data show that parasites grown in erythrocytes from an iron-deficient donor displayed significantly reduced growth rates compared to healthy control. Our in vitro results also demonstrate that hepcidin treatment of control erythrocytes elevated intracellular Fe²⁺ concentrations and promoted parasite proliferation.

To identify iron-regulated mechanisms and putative iron transporters in P. falciparum, we carried out whole-transcriptome profiling using bulk RNA-sequencing ( Figure 2 ). P. falciparum 3D7 parasites were cultured either using erythrocytes from a donor with higher, control (healthy) or low iron status (experiment 1); or with red blood cells from another healthy donor in the presence or absence of 0.7 µM hepcidin (experiment 2). Samples from three biological replicates per condition were harvested at the ring and trophozoite stage (6 – 9 hpi and 26 – 29 hpi) during the second IDC under the conditions specified.

To exclude the possibility that differences in mRNA abundance were a consequence of divergent progression through the IDC under different nutritional conditions, we assessed the average developmental age of the parasites in each sample on the basis of a statistical maximum likelihood estimation (MLE) method of transcriptional patterns according to Lemieux et al. (Lemieux et al., 2009). The general transcriptional patterns of parasites were highly similar at individual time points and consistent across different experimental treatments, corresponding to those of a 3D7 reference strain (Broadbent et al., 2015) at approximately 10 hpi and 35 hpi ( Figure 2A ). This indicates that differences in mRNA abundance of parasites were not caused by divergent progression through the IDC but by direct effects of the experimental treatments. As the 3D7 strain we used for the experiments had a reduced total IDC length of 44 h instead of 48 h, possibly because of gene deletions that may have occurred during long-term culturing (Wichers et al., 2019; Stewart et al., 2020), it progresses through the cycle faster than the 3D7 reference strain (Broadbent et al., 2015). This may explain why the calculated MLEs of parasite age were higher than the actual values of 6 – 9 hpi and 26 – 29 hpi ( Figure 2A ).

Using a threshold of 1.5 for the fold change (FC) in gene expression (log₂ FC of 0.585 or -0.585) yielded twelve significantly upregulated and 175 downregulated genes in ring-stage parasites under high vs. low-iron conditions (P < 0.05, exact test for the negative binomial distribution with Benjamini-Hochberg correction (Benjamini and Hochberg, 1995)). As differences in transporter gene transcription are typically small (Küpper and Kochian, 2010; Abrahamian et al., 2016), we examined the 351 upregulated and 770 downregulated genes with a significant expression change and a minimum absolute value of the log₂ FC of 0.2 for this comparison ( Figure 2B ). The full RNA-sequencing datasets are available in the BioStudies repository (Sarkans et al., 2018) under accession number E-MTAB-13411 (https://www.ebi.ac.uk/biostudies/studies/E-MTAB-13411) and differential gene expression test results for individual genes are shown in Supplementary Tables S1 and S2 . The highly polymorphic var, stevor, and rifin gene families were excluded from downstream analyses because of their significant sequence diversity between parasites of the same strain during mitotic growth (Bozdech et al., 2003; Kidgell et al., 2006). Functional Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG) and Reactome (REAC) term enrichment analyses of differentially expressed genes (DEGs) were performed using the g:Profiler web server (Raudvere et al., 2019).

Under high vs. low-iron conditions at the ring stage (6 – 9 hpi), the GO term for biological process GO:0055085 “transmembrane transport” was 2.8-fold enriched (P = 0.007, hypergeometric test) among significantly upregulated parasite genes ( Figure 2C ). Using the recently updated P. falciparum transporter list (Wunderlich, 2022), all genes with differential expression levels at the ring stage were then screened for transport proteins and all of the five genes previously proposed as iron transporters in Plasmodium (VIT, ZIPCO, NRAMP/DMT1, CRT, MRS3/MFRN (Mach and Sutak, 2020; Sloan et al., 2021)) were found differentially expressed ( Table 1 ). The putative transporter PfE140 (Wunderlich, 2022), which is predicted to be essential (Zhang et al., 2018), may also be involved in iron transport, as it was differentially expressed under low-iron vs. control conditions at the ring stage ( Table 1 ) as well as in the presence vs. absence of hepcidin in trophozoites ( Figure 2D ). Other significantly enriched functional terms at the ring stage under high-iron conditions were GO:0009056 “catabolic process”, GO:0020020 “food vacuole”, KEGG:01100 “metabolic pathways”, and GO:0005737 “cytoplasm” ( Figure 2C ). Among downregulated genes under high vs. low-iron conditions at the ring stage, KEGG:03440 “homologous recombination”, KEGG:03410 “base excision repair”, GO:0007049 “cell cycle”, and GO:0015630 “microtubule cytoskeleton” were enriched ( Figure 2C ). At the more metabolically active trophozoite stage (26 – 29 hpi), processes related to mRNA splicing and protein production were overrepresented in upregulated genes, as indicated by the 2.9-fold enrichment (P = 0.00006) of the KEGG:03040 pathway “spliceosome” and the 2.5-fold enrichment (P < 0.05) of the GO:0015934 term “large ribosomal subunit” ( Figure 2C ).

In contrast, hepcidin treatment resulted in reduced metabolism compared to control conditions, as KEGG:00040 “pentose and glucuronate interconversions”, REAC:R-PFA-71291 “metabolism of amino acids and derivatives”, GO:0005737 “cytoplasm”, and GO:0015934 “large ribosomal subunit” were significantly enriched in downregulated genes during the parasite ring stage at 6 – 9 hpi ( Figure 2E ). Among significantly upregulated genes in the presence vs. absence of hepcidin, the terms GO:0070258 “inner membrane pellicle complex” (P < 0.05), KEGG:03430 “mismatch repair” (P = 0.04), and GO:0015630 “microtubule cytoskeleton” (P = 0.04) were enriched at the ring stage. GO:0044409 “entry into host” (P = 0.00008) and GO:0052126 “movement in host environment” (P = 0.0001) were overrepresented at the trophozoite stage ( Figure 2E ), possibly linked to the observed increase in parasite proliferation ( Figure 1B ).

Our RNA-sequencing data also revealed the differential expression of genes involved in epigenetic, transcriptional, translational, and post-translational regulation. Under high vs. low-iron conditions, histone deacetylation and chromatin organization processes as well as GO:1990904 “ribonucleoprotein complex” were significantly enriched in upregulated genes at the trophozoite stage, and GO:000370 “DNA binding transcription factor activity” in downregulated genes at the ring stage ( Figure 2C ). Furthermore, the known iron-regulatory protein PfIRP or aconitate hydratase (Loyevsky et al., 2001; Hodges et al., 2005) was upregulated during the ring stage under high vs. low-iron conditions (log₂ FC = 0.49, P = 0.00003) and downregulated in the presence of hepcidin (log₂ FC = -0.27, P = 0.01) as compared to control. Many protein kinases involved in post-translational modifications and endocytosis were also upregulated at 26 – 29 hpi at high vs. low iron levels, as indicated by the enriched terms GO:0043170 “macromolecule metabolic process” and KEGG:04070 “phosphatidylinositol signaling system” ( Figure 2C ).

On the basis of transcriptomic profiles and the P. falciparum transporter list (Wunderlich, 2022), six proteins with a potential role in iron transport were identified ( Tables 1 , 2 ). The subcellular localization of the four proteins that had not yet been localized in P. falciparum (PfMRS3, PfVIT, PfZIPCO, and PfE140) was then examined by endogenous tagging with GFP and confocal imaging of live parasites under physiological control conditions. At least two cell lines were generated per transporter candidate with consistent results and representative example images are shown in Figure 3 . Diagnostic PCRs confirmed the fusion of gfp to the respective gene of interest and the absence of parental DNA at the original locus ( Supplementary Figure S2 ). Only the PfMRS3 reporter cell line still contained wild-type DNA of the parental parasites even after prolonged WR99210/G418 selection and limiting dilution cloning ( Supplementary Figure S2 ), indicating the importance of this mitochondrial transporter for asexual parasite growth during the blood stage.

The GFP-tagged mitochondrial carrier protein PfMRS3 exclusively localized to the mitochondrion, as determined by colocalization with MitoTracker Red ( Figure 3A , Supplementary Video S1 ). PfVIT-GFP displayed a punctate fluorescence pattern within the cytoplasm ( Figure 3B ; Supplementary Video S2 ), which did not colocalize with ER Tracker Red in live cells ( Figure 3B ; Supplementary Video S2 ). Expression of PfZIPCO-GFP resulted in highly similar fluorescent cytoplasmic dots ( Figure 3C ; Supplementary Video S3 ). To test whether these could be acidocalcisomes, we employed LysoTracker Deep Red, commonly used to visualize small acidic organelles in T. brucei (Huang et al., 2014). However, the fluorescent dye only stained the DV in P. falciparum ( Figure 3C ; Supplementary Video S3 ) and no acidocalcisome-specific marker is currently available for this parasite. For both PfVIT-GFP and PfZIPCO-GFP, the number of cytoplasmic foci increased as the parasites matured from the ring to the late schizont stage ( Figures 3B, C ).

GFP-tagged PfE140 (PF3D7_0104100), also known as conserved Plasmodium membrane protein or CPMP (Lerch et al., 2017), localized to the parasite plasma membrane, as evidenced by the ring-like fluorescence pattern around newly formed merozoites ( Figure 3D ; Supplementary Video S4 ). The fluorescence intensity was very low at the ring and early trophozoite stage compared to schizonts. Because of amino acid sequence similarity (E = 9 x 10^-5, 22.5% identity, 66% coverage) to the essential apicoplast transporter PfDER1-2 (Altschul et al., 1997; Spork et al., 2009), we also investigated the potential colocalization with the apicoplast marker PfACP (acyl carrier protein), which could not be detected ( Figure 3D ; Supplementary Video S4 ).

To study the function of the putative transport proteins identified, we used targeted gene disruption (TGD) by selection-linked integration (SLI) to generate the corresponding knockout parasite lines for the putative iron transporters that are non-essential during P. falciparum blood stage: PfVIT and PfZIPCO ( Figure 4A ; Supplementary Figure S2 ). The cloning strategy requires a homology region of at least 400 bp, and GFP was cloned in frame with the truncated version of the respective transporter (the N-terminal 143 amino acids (aa) of 274-aa PfVIT or 117 of the 325 aa of PfZIPCO). The subcellular localization of the resulting GFP fusion protein was also assessed to confirm protein expression and elucidate the position of targeting signals. PfVIT (1–143)-GFP localized to cytoplasmic structures and PfZIPCO(1-117)-GFP to the DV and cytoplasmic vesicles ( Figure 4A ). Hence, both truncated proteins contained (a) sequence(s) for targeting to the observed vesicles. As additional DV staining is often non-specific, another part of the protein located within the C-terminal 208 aa of PfZIPCO may be required for exclusively vesicular localization.

Proliferation assays were then performed to determine the importance of the respective transporter for parasite growth. While the PfVIT knockout had no effect on parasite growth under standard conditions, addition of hepcidin reduced the growth rate of the ΔVIT line by 30% ( Figure 4B ). Of note, hepcidin generally had a smaller effect after two cycles of incubation ( Figure 4B ) than after one cycle compared to the first IDC ( Figure 1B ). Unexpectedly, knocking out PfZIPCO led to a growth rate increase by 42% after two IDCs relative to wild-type 3D7 parasites, and thus a rescue by hepcidin treatment was not tested for this knockout line ( Figure 4B ).

PfE140 is predicted to be essential (Zhang et al., 2018) and the only putative iron transporter identified that localized to the PPM ( Figure 3 ), thus potentially important for iron uptake in P. falciparum. For an inducible knockdown, a glmS ribozyme sequence (Prommana et al., 2013) was introduced upstream of the 3’ untranslated region in the pSLI plasmid, allowing for conditional mRNA degradation by adding 2.5 mM glucosamine (GlcN) to the culture medium ( Figure 4C ; Supplementary Figure S2 ). The knockdown led to a 61% decrease in total parasite fluorescence intensity after 36 hours of GlcN treatment ( Figure 4D ) without affecting parasite size compared to untreated control ( Figure 4E ). Addition of GlcN also caused a 38% growth rate reduction of the PfE140-GFP-glmS line, which was rescued by hepcidin treatment to a proliferation level that was not significant different from that under standard culture conditions (P = 0.25, Figure 4F ). The generation of a PfMRS3-knockdown line was not successful after four independent attempts of G418 selection for integrants after transient transfection, supporting the essentiality of the gene for asexual parasite growth (Zhang et al., 2018).

We next took advantage of the recent progress in protein structure prediction and generated models of the putative iron transport proteins identified ( Table 2 ) using AlphaFold2 (Jumper et al., 2021; Varadi et al., 2022) and AlphaFold2-multimer (Evans et al., 2022). The transmembrane regions of the proteins typically exhibited the highest confidence score, while some other protein portions appeared unstructured ( Figure 5A ). Regions that are likely located within a membrane were validated by inspecting the molecular lipophilicity potential of the protein surfaces ( Figure 5B ). A clear hydrophobic belt was observed for all proteins and their orientation in the membrane was determined on the basis of orthologous proteins. As transport cavities with negatively charged residues are a hallmark of heavy metal ion transporters, we analyzed the distribution of charge on the surface of the proteins and looked for negatively charged regions to assess the capacity to bind cations like Fe²⁺ ( Figure 5C ). To gain further insights into the functions of the proteins identified, we also compared the predicted structures with those of well-characterized homologs from S. cerevisiae, Eucalyptus grandis, Bordetella bronchiseptica and Staphylococcus capitis ( Table 2 ; Supplementary Figures S3 , S4 ).

The outer surface of PfMRS3 (transport classification (TC): 2.A.29, mitochondrial carrier family) is positively charged ( Figure 5C ) and there is a clear negatively charged patch in the putative binding pocket facing the mitochondrial intermembrane space. We compared the predicted PfMRS3 structure with that of S. cerevisiae MRS3, which is known to import ferrous iron into the mitochondrial matrix across the inner membrane (Mühlenhoff et al., 2003; Froschauer et al., 2009; Brazzolotto et al., 2014). The predicted structures of PfMRS3 and S. cerevisiae MRS3 were superimposed with an average root mean square deviation of Cα atoms (Cα RMSD) of the 205 matched residues of 2.3 Å ( Supplementary Figures S3A , S4A ). The conserved histidine residues His⁴⁸ and His¹⁰⁵ that were required for Fe²⁺ transport by S. cerevisiae MRS3 in reconstituted liposomes (Brazzolotto et al., 2014) are also present in PfMRS3 and are in a similar molecular context in both structures ( Supplementary Figures S3A , S4A ). This suggests that MRS3 may elicit similar molecular functions in S. cerevisiae and P. falciparum.

PfVIT is highly similar to VIT1 from E. grandis (E = 7 x 10^-27, 30.3% identity, 84% coverage), for which an experimental structure is available (Protein Data Bank (PDB) identifier: 6IU4). The plant protein crystallized as a homodimer (Kato et al., 2019), and the same oligomeric state was suggested for the vacuolar iron transporter family (TC: 2.A.89) protein in P. falciparum (Sharma et al., 2021). A PfVIT monomer also has five transmembrane domains and comprises a negatively charged region facing the cytosol that may enable cation transport ( Figure 5C ). In agreement with this, one Fe²⁺ ion and two Zn²⁺ ions were bound by a strongly charged region on the cytosolic side of the E. grandis VIT1 monomer (Kato et al., 2019) and a highly similar putative binding pocket is present in the parasite protein ( Supplementary Figure S4B ). In the structural alignment, 219 of the 227 residues of the experimental EgVIT1^23-249 structure are within 5 Å of the predicted structure of PfVIT with an average Cα RMSD of 1.9 Å and the key residues in the metal-binding domain (Glu¹⁰², Glu¹⁰⁵, Glu¹¹³, Glu¹¹⁶, using EgVIT1^23-249 numbering) are placed in a similar molecular context in the predicted structure of PfVIT ( Supplementary Figures S3B , S4B ). The residues in the transmembrane domain that are in the vicinity of the Co²⁺ ion in the EgVIT1^23-249 structure (Met⁸⁰ and Asp⁴³) are also conserved ( Supplementary Figure S3B ), which is in line with a similar function of PfVIT and EgVIT1.

PfZIPCO contains seven transmembrane domains and was modeled as a homodimer ( Figure 5A ), as it is part of the zinc (Zn²⁺)-iron (Fe²⁺) permease (ZIP) family (TC: 2.A.5), whose members usually function as homo- or heterodimers (Saier et al., 2016). The negatively charged patch in each binding pocket facing the vesicle lumen ( Figure 5C ) may be involved in cation transport to the cytosolic side. In an overlay of the PfZIPCO model with the cryo-EM structure (PDB: 8GHT) of a ZIP transporter from B. bronchiseptica in the presence of either Zn²⁺ or Cd²⁺ ions (Pang et al., 2023), the average Cα RMSD of 140 sequence-aligned residues was 2.0 Å ( Supplementary Figures S3C , S4C ). Several key residues of the metal binding site M1 of BbZIP (Met⁹⁹, His¹⁷⁷, Glu¹⁸¹, Glu²¹¹) were also found in PfZIPCO, whereas others (Asn¹⁷⁸, Gln²⁰⁷, Asp²⁰⁸, Glu²⁴⁰) were different ( Supplementary Figure S3C ), possibly resulting in divergent substrate specificity.

PfNRAMP (TC: 2.A.55, metal ion (Mn²⁺-iron) transporter family) is a homolog of the human endosomal Fe²⁺ transporter 2/DMT1 (Martin et al., 2005) and contains twelve transmembrane domains ( Figure 5A ). Like PfCRT (TC 2.A.7, drug/metabolite exporter family), for which a recent cryo-EM structure (PDB: 6UKJ) is available (Kim et al., 2019), the predicted structure possesses a negatively charged region within its binding pocket facing the digestive vacuolar lumen ( Figure 5C ). This is consistent with binding of cations such as Fe²⁺. The PfNRAMP model was superimposed on the solved crystal structure of S. capitis NRAMP/DMT (PDB: 5M95, E = 1 x 10^-32, 26.3% identity, 60% coverage), which was shown to bind Mn²⁺, Fe²⁺, Co²⁺, Ni²⁺, Cd²⁺ and Pb²⁺ (Ehrnstorfer et al., 2014). In the overlay, the average Cα RMSD of the 349 matched residues was 1.6 Å and the negatively charged cavity of PfNRAMP was in close proximity to the Mn²⁺ ion bound to S. capitis NRAMP ( Supplementary Figure S3D , Supplementary Figure S4D ). Two of the four key residues required for ion coordination in the binding pocket of the bacterial protein (Asn⁵² and Asp⁴⁹) are present in PfNRAMP, whereas the two other residues (Met²²⁶ and Ala²²³) are changed to serine. The functional implications of the latter are unclear; Ehrnstorfer et al (Ehrnstorfer et al., 2014) showed that Met²²⁶Ala mutant of S.capitis NRAMP exhibited reduced transport activity and binding, however, transport could not be completely abolished by these mutations. While potential effects of these differences on substrate specificity and/or transport activity remain to be elucidated, PfNRAMP is likely to perform cation transport from the DV into the cytosol.

PfE140 is predicted to be anchored in the parasite plasma membrane by a bundle of five transmembrane domains (Meerstein-Kessel et al., 2021). In the AlphaFold prediction, it forms a coiled coil with a hydrophilic region that displays negatively charged patches exposed to the extracellular side ( Figures 5A, B ). No human orthologs could be identified for this highly conserved Plasmodium protein (Altschul et al., 1997). As there is no obvious channel or cavity in the transmembrane region of the PfE140 monomer ( Figures 5B, C ), the helical bundles may form a dimer to enable ion transport. However, we were not able to obtain a PfE140 dimer model with AlphaFold2-multimer because of its sequence length. To predict functional residues on the basis of the amino acid sequence and the AlphaFold2 structure of PfE140, we used DeepFRI graph convolutional network (Gligorijević et al., 2021), which has significant denoising capability and can reliably assign GO terms to residues in the protein. In particular, the terms GO:0022857 “transmembrane transporter activity” (DeepFRI gradCAM score 0.94), GO:0015075 “monoatomic ion transmembrane transporter activity” (score 0.78), and GO:0046873 “metal ion transmembrane transporter activity” (score 0.67) were assigned to a putative transmembrane region of PfE140 with high confidence ( Supplementary Figure S5 ). We thus speculate that the protein is a transporter of metal ions.