Improving reproducibility of Plasmodium falciparum culture: a large-language-model-driven literature review and a parasite growth variability assessment of donor red blood cells Authors: Sara M. Mahomed, Thomas Hanscheid, Yara Almeida, Martin P. Grobusch, Maria Rebelo Published: 2025 ABSTRACT BackgroundIn vitro culture of Plasmodium falciparum forms the basis for many key malaria research areas globally, yet the human red blood cells (RBC) essential for culture remain poorly characterized in most studies, despite growing evidence of substantial donor-dependent variability that can influence parasite growth and experimental outcomes.MethodsA dual investigation was performed comprising: (1) a large language model-assisted systematic review of 963 PubMed-indexed articles (2000–2025) to assess reporting of RBC sourcing in malaria culture studies; and (2) an experimental analysis of parasite growth variability using standardized P. falciparum strain 3D7 culture with RBC from 69 malaria-naïve volunteers, including detailed haematological profiling of 24 participants.ResultsAmong the 963 studies, 314 provided explicit information on the RBC source. Of these, 61.1% failed to specify donor origin, 72.6% omitted blood group information, and only 6.0% reported the anticoagulant used. Type O RBC were used in 54.7% of studies reporting the blood group, despite limited evidence for biological advantage. In the experimental cohort, parasite growth after 96 h varied 6.4-fold across volunteers (range: 3.0–19.3), with growth phenotypes reproducible over three years. Low-growth volunteers had lower haemoglobin concentrations (12.3 vs. 14.8 g/dL, p = 0.03), more frequent personal or family history of anaemia (p = 0.01), and greater ancestry from malaria-endemic regions (p = 0.03). However, extended laboratory testing failed to identify any consistent clinical or biochemical predictors of growth capacity.ConclusionsDonor RBC are a major and reproducible source of biological variability in P. falciparum culture, yet are inconsistently reported and poorly understood. To address this, the STRuBeC (Standards for Reporting for use of (Red) Blood Cells in Culture) is proposed as a minimum reporting standard for studies using P. falciparum in vitro culture, to improve transparency, reproducibility, and cross-study comparability.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12936-025-05722-x. Background The in vitro culture of Plasmodium falciparum, first established in 1976, has played an essential role in malaria research for nearly five decades [1, 2]. However, P. falciparum culture remains entirely dependent on primary red blood cells (RBC) from human donors; cells that are biochemically and genetically variable. While methods sections meticulously describe parasite lines, media, and gas conditions, RBC are often treated as neutral reagents and typically reduced information is provided, such as ‘O-positive blood’. Studies have shown that parasite invasion and replication vary substantially between donors, even under otherwise identical conditions [3, 4]. These differences reflect stable, intrinsic properties of the RBC, meaning that key parasite phenotypes, including drug sensitivity and growth rate, may be influenced by donor variability rather than parasite biology. RBC characteristics such as haemoglobin S (HbAS), haemoglobin C (HbC), alpha-thalassaemia, and G6PD deficiency can impair parasite development [5–7]. Although these well-known variants are unlikely to be overlooked in controlled laboratory studies, more recent studies, have identified the Dantu blood group variant as an additional protective factor, associated with 50% reduction in merozoite invasion [8]. Other individual differences, such as RBC age distribution, deformability, redox status, and surface receptor density, also modulate parasite invasion and replication [4, 9, 10]. Even transient physiological states, including iron deficiency or mild anaemia, may alter the RBC environment, and thus possibly impact the parasite growth [11–13]. In addition, technical factors, such as the anticoagulant (e.g. heparin, citrate, and EDTA) used for blood collection may interfere differently with P. falciparum. Heparin impacts merozoite surface protein binding and reduces cytoadherence under certain conditions [14], while EDTA has been reported to inhibit parasite invasion and gametocyte development [15, 16]. Citrate, widely used in blood banking, is generally considered acceptable and is included in storage solutions that also contain adenine, phosphate and glucose. However, RBC stored in these citrate-based additives undergo progressive metabolic and structural changes that may influence parasite viability [17, 18]. One of the most persistent practices in P. falciparum culture is the preferential use of group O RBC. Historically, concerns about ABO incompatibility in serum-supplemented media likely led to the perception of group O as a safe universal choice. Additionally, group O is among the most common blood groups in the world, further contributing to its widespread use [2]. However, with the widespread shift to serum-free bovine additives [19], this rationale no longer applies. Regarding parasite growth, some studies have reported slightly higher invasion rates into group O cells [20, 21], possibly related to differential H antigen expression [22], but the observed effects are modest. In practice, P. falciparum is capable of growing efficiently in RBC from any ABO group [3, 23, 24]. Interestingly, population studies show that individuals with blood group O are protected against severe malaria; however, this is not due to limitations in parasite invasion, but to reduced rosetting and cytoadherence [20, 25, 26]. Despite growing recognition of host-cell variability, P. falciparum culture protocols lack standardised reporting for RBC sourcing. In contrast to fields with established guidelines, such as MIQE for qPCR [27] or ARRIVE for animal studies [28], malaria studies in general seem to ignore critical RBC parameters. Possibly, to improve comparability and reproducibility, the malaria research field would benefit from a consensus framework for reporting P. falciparum culture conditions, such as reporting donor type (volunteer vs. blood bank), ABO/Rh group, anticoagulant, and storage duration. Little is known about how well P. falciparum culture studies document RBC sources, a problem compounded by the lack of PubMed indexing for culture conditions and the fact that key details are often buried in full-text PDFs. While inter-donor variation in parasite growth has been noted [3, 29–31], previous studies were small and did not assess whether common clinical parameters might predict growth outcomes. This study aimed to assess how well RBC sourcing is reported in the literature, and to investigate the extent of inter-individual variability in parasite growth across a larger cohort. To this end, large language models (LLM) were applied to extract and classify unstructured data from published studies and investigated in a large cohort whether simple donor characteristics, such as clinical history or routine laboratory values, could predict P. falciparum growth. Based on these findings, the STRuBeC (Standards for Reporting use of (Red) Blood Cells in Culture) framework is proposed as a minimum reporting standard for P. falciparum blood-stage culture, aiming to improve reproducibility and transparency across studies. Methods This study comprised two distinct components: (1) a comprehensive AI-assisted systematic review of the literature to analyse blood source reporting practices in published P. falciparum culture studies; and (2) an experimental laboratory investigation examining interindividual variations in P. falciparum growth using volunteer red blood cells. LLM- assisted literature review This study employed a multi-phase, LLM-assisted systematic review to identify and analyse blood source reporting in P. falciparum in vitro culture studies. The complete computational workflow (Fig. 1) is detailed in the Supplementary Methods A. All Python scripts used for data processing and visualization, including PubMed retrieval, abstract triage, PDF-to-text conversion, verbatim extraction, and classification prompts, are publicly available at 10.5281/zenodo.17534229 (Repository: PfCultureReproducibility-LLM-Scripts2025).Fig. 1PRISMA 2020 flow diagram adapted for a large-language-model–assisted systematic review on Plasmodium falciparum red blood cell sources. Schematic representation of the large-language-model (LLM)–based pipeline used to identify, extract, and classify methodological information on red blood cell (RBC) sourcing from published P. falciparum in vitro culture studies PRISMA 2020 flow diagram adapted for a large-language-model–assisted systematic review on Plasmodium falciparum red blood cell sources. Schematic representation of the large-language-model (LLM)–based pipeline used to identify, extract, and classify methodological information on red blood cell (RBC) sourcing from published P. falciparum in vitro culture studies Initial literature retrieval was conducted using a custom Python script to query PubMed via the Entrez API with a comprehensive MeSH-based search string, yielding 1,921 records (Supplementary Methods A, Step 1). After programmatic filtering for original research articles published between 2000 and 2025, 963 unique citations remained for analysis (Supplementary Methods A, Step 2). To efficiently prioritize articles for full-text retrieval, an LLM-based triage system was developed using the Google Generative AI API with the Gemini 1.5 model. This classifier analysed titles and abstracts to categorize papers as ‘Highly Likely’, ‘Possibly’, or ‘Unlikely’ to contain relevant culture information, identifying 674 potentially relevant papers (Supplementary Methods A, Steps 3–4). The automated classification achieved 90% recall against expert review in a validation study of 50 randomly selected abstracts (Supplementary Methods A, Step 5). Manual retrieval of full-text PDFs was performed for the 560 articles classified as ‘highly likely’, resulting in 459 successfully downloaded files. A custom Python script utilizing the pdfplumber library converted each PDF into LLM machine-readable plain text file (Supplementary Methods A, Steps 6–7). For core data extraction, the full text of each file was submitted to the Gemini 1.5 API with prompts designed to perform targeted, verbatim extraction of sentences describing red blood cell sources for culture (Supplementary Methods A, Step 8). Extraction fidelity was validated against a random sample of 50 full-text articles, where expert review confirmed correct identification and extraction of contextually relevant information (Supplementary Methods A, Step 9). This validation quantified both recall (90%) and false positive rates (23.4%) in the extraction step. To assess whether these over-inclusions compromised downstream analysis, a second validation was conducted on a separate sample of 146 articles, confirming 96% classification accuracy across predefined blood source categories (Supplementary Methods A, Step 11). The extracted verbatim descriptions were then programmatically classified into five predefined source categories through an iterative process requiring significant prompt refinement (Supplementary Methods A, Step 10). Finally, aggregated data was analysed and cross-validated using multiple LLM interfaces (GPT-4o, Claude 3 Opus, and Gemini 1.5) to produce the final summaries (Supplementary Methods A, Step 12). Experimental investigation examining inter-individual growth patterns Study participants Malaria-naive adult volunteers were recruited frome Faculty of Medicine, University of Lisbon, including students, faculty members, clinicians, and researchers. Individuals with a history of malaria or relevant comorbidities, such as hereditary RBC disorders, were excluded. Study design In Phase 1, volunteers completed a questionnaire and provided venous blood samples for P. falciparum in vitro culture. Based on parasite growth after 96 h of culture, individuals were classified into ‘low-’, ‘intermediate-’, or ‘high-growth’ cohorts. Glycaemia, haemoglobin concentration, and ABO/Rh blood groups were also measured. In Phase 2, 24 volunteers, primarily from the ‘low-’ and ‘high-growth’ cohorts, were invited for a second blood donation to assess the reproducibility of parasite growth patterns. In vitro cultures were repeated under identical conditions, and participants underwent an extended panel of haematological and biochemical tests (see Supplementary Methods B, Supplementary Table S11). Questionnaire Each volunteer completed a structured questionnaire to identify factors potentially influencing P. falciparum growth. Data collected included demographic information (age, sex, and birthplace), history of blood donation, dietary habits (including composition and timing of the last meal), previous malaria exposure, history of anaemia, current medications, comorbidities, and family history of red blood cell-related disorders. The full questionnaire is provided in the Supplementary File 1. Blood collection, on-site measurements, and laboratory testing During Phase 1, venous blood was collected from each volunteer for RBC isolation and subsequent culture. RBC were washed, resuspended in RPMI 1640 medium (Gibco, Thermo Fisher Scientific, USA), and stored at 4 °C for up to 24 h. Capillary glycaemia was measured on site using the OneTouch Select® Plus (LifeScan, Gubelstrasse, Switzerland). ABO and Rhesus (Rh) blood groups were determined by rapid slide agglutination using Seraclone™ antisera (Bio-Rad, Germany), following the manufacturer instructions. Haemoglobin concentration was determined spectrophotometrically by mixing 20 µL of EDTA-anticoagulated blood with 5 mL of Drabkin’s reagent (supplied by the clinical laboratory at Hospital de Santa Maria) and measuring absorbance at 540 nm using a cuvette-based spectrophotometer Spectronic™ Helios™ Gamma (Thermo Electron, Massachusetts, EUA) – detailed method in the Supplementary Methods B. In Phase 2, selected volunteers provided additional venous blood samples for comprehensive haematological and biochemical analyses, performed at a certified clinical laboratory. Haematological testing included complete blood count, reticulocyte count, iron studies (serum iron, ferritin, and transferrin), biochemical analyses included vitamin B12, folic acid, lactate dehydrogenase (LDH), total and direct bilirubin, zinc, glycaemia, glycated haemoglobin (HbA1c), lipid profile, and haemoglobinopathy screening by high-performance liquid chromatography (HPLC). Test details and reference ranges are provided in the Supplementary Methods B, Supplementary Table S11. In vitro parasite culture with volunteer RBC Plasmodium falciparum 3D7 parasites were cultured at 37 °C in a 5% CO₂ atmosphere using Malaria Complete Medium (MCM). MCM consisted of RPMI 1640 medium withou L-glutamine, HEPES (1 M stock), gentamicin (50 mg/mL), (Gibco, Thermo Fisher Scientific, USA) L-glutamine (200 mM) (Merck, Germany), and 0.5% Albumax II. Detailed medium composition and reagent concentrations are provided in the Supplementary Methods B. Plasmodium falciparum cultures were maintained at 5% haematocrit and under 2% parasitaemia, with daily medium replacement and dilution using fresh RBC from the Portuguese Blood and Transplantation Institute (IPST). In selected experiments, Albumax was replaced with 10% human serum obtained from volunteers, requiring adjustments in glucose and other components – detailed method in the Supplementary Methods B. For each volunteer assay, 20 μL of P. falciparum-infected RBC from a continuous culture was combined with 180 μL of washed volunteer RBC and 3.8 mL of prewarmed MCM, to achieve a 5% haematocrit. Cultures were established in 24-well plates (Corning, USA) and incubated at 37 °C and 5% CO₂ for 96 h, with daily medium changes. Parasitaemia was assessed at baseline (0 h) and after 96 h of incubation. Fold-change in parasite growth was calculated as the ratio of final parasitaemia to initial parasitaemia. Volunteers whose parasite growth ratios deviated by at least one standard deviation from the cohort mean were classified as low or high growers and invited for Phase 2. In this second phase, RBC from selected volunteers were processed and cultured under identical conditions to validate parasite growth dynamics. Parasitaemia was primarily assessed by light microscopy of Giemsa-stained smears. At least 1,000 RBC were counted per sample. Parasitaemia was determined using ImageJ software (NIH, USA), using photographic images of stained smears. Full details of image acquisition and quantification methods are provided in the Supplementary Methods B. Data management and statistical analysis Data were entered in Microsoft Excel (Microsoft Office 2019) and analysed using GraphPad Prism version 9 (GraphPad Software, San Diego, CA, USA). Descriptive statistics were reported as frequencies for categorical variables and means ± standard deviation or medians with ranges for numerical variables, depending on distribution. Parasite growth was expressed as the fold-change in parasitaemia (96 h divided by baseline). Normality was assessed with the Shapiro–Wilk or Kolmogorov–Smirnov tests. For normally distributed data, comparisons were performed using Student’s t test (two groups) or one-way ANOVA (multiple groups). For non-normally distributed data, the Mann–Whitney U test or Kruskal–Wallis test was applied. Categorical variables were compared using Fisher’s exact test. Effect sizes were calculated using Cohen’s d, and correlations were expressed as the coefficient of determination (R2) with 95% confidence intervals. Given the limited sample size in Phase 2 (n = 24), nonparametric methods were prioritised to account for distributional assumptions and ensure robustness. Results LLM- assisted literature review The initial Python-based MeSH query yielded 1,921 PubMed citations (Fig. 1). After restricting to original research articles published between 2000 and 2025, the dataset was reduced to 963 records (Supplementary Methods A, Step 2). Of these, an LLM-based classifier (Gemini 1.5) identified 674 (70.0%) as likely to contain blood-source information (classified as “highly likely” or “possible”) (Supplementary Methods A, Step 4). Validation against a randomly selected expert-reviewed subset (n = 49) showed 90% recall, with the model recovering 27 of 30 relevant papers. Notably, no study classified as ‘highly likely’ by expert review was missed (Supplementary Methods A, Step 5). Full-text PDFs were successfully retrieved for 459 of the 674 likely studies (68.1%). Manual screening of these PDFs revealed that 314 (68.4%) contained analysable information on RBC sources, whereas 145 (31.6%) did not provide usable details. The 314 informative studies spanned diverse peer-reviewed journals, including high-impact general publications (9.2%), specialist malaria and parasitology journals (19.7%), general microbiology journals (17.2%) and journals in other medical fields (53.8%). Full-text validation showed 90% recall for RBC source sentence extraction (18/20), with a high false positive rate (11/47, 23.4%) due to contextual over-inclusion. However, subsequent validation of the classification step (n = 146) confirmed that these over-inclusions did not lead to relevant misclassification, with an overall classification accuracy of 96%. Detailed error types and adjudication are provided in Supplementary Methods A, Steps 9 and 10. Among these 314 studies (Supplementary Methods A, Steps 8), 192 (61.1%) lacked sufficient information to fully characterize the blood source (Table 1). Direct volunteer recruitment was reported in only 23 studies (7.3%), while 48 studies (15.3%) used blood bank sources. Blood group information was frequently missing: ABO/Rh typing was absent in 228 of the 314 studies (72.6%). Reporting frequency varied by source category, with blood type specified in 29.7% of blood bank-derived samples, 21.7% of volunteer donations and 5.6% of patient samples (Table 1). Among the 86 studies that reported blood group, O Rh-positive was most common (47/86, 54.7%), followed by A Rh-positive (19/86, 22.1%) (Table 2). Only 19 studies (6.0%) specified the anticoagulant used: heparin in 15 cases and EDTA in four of them. No other anticoagulants were reported across the dataset. Table 1Primary blood sources in the screened literature (n = 314)Source category(n)Blood group mentionedVolunteer2322%Patient (clinical)366%Blood bank4842%Human–unspecifieda13145%Non‑humanb150%No information610%aHuman unspecified sources (n = 131) mentioned human erythrocytes, RBCs, or blood groups without documenting acquisition methods, donor characteristics, or institutional sourcesbNon-human sources consisted primarily of rodent malaria models (P. berghei, P. yoelii) for liver-stage assays and transgenic parasite development, primate blood for P. vivax and P. knowlesi research, and specialized culture systems requiring species-specific blood sourcesTable 2ABO/Rh distribution among studies that reported a blood group (n = 86)Blood group(n)(%)O Rh-positive4754.7A Rh-positive1922.1Multiple/pooled67.0A Rh-negative22.3O Rh-negative22.3Othera1011.6Total specified86100.0aother categories: A1; A Rh1; O +/A −; A −/A +; ORh +; ARh variants; O +/Multiple; B +; AB Table 1Primary blood sources in the screened literature (n = 314)Source category(n)Blood group mentionedVolunteer2322%Patient (clinical)366%Blood bank4842%Human–unspecifieda13145%Non‑humanb150%No information610%aHuman unspecified sources (n = 131) mentioned human erythrocytes, RBCs, or blood groups without documenting acquisition methods, donor characteristics, or institutional sourcesbNon-human sources consisted primarily of rodent malaria models (P. berghei, P. yoelii) for liver-stage assays and transgenic parasite development, primate blood for P. vivax and P. knowlesi research, and specialized culture systems requiring species-specific blood sources Primary blood sources in the screened literature (n = 314) aHuman unspecified sources (n = 131) mentioned human erythrocytes, RBCs, or blood groups without documenting acquisition methods, donor characteristics, or institutional sources bNon-human sources consisted primarily of rodent malaria models (P. berghei, P. yoelii) for liver-stage assays and transgenic parasite development, primate blood for P. vivax and P. knowlesi research, and specialized culture systems requiring species-specific blood sources Table 2ABO/Rh distribution among studies that reported a blood group (n = 86)Blood group(n)(%)O Rh-positive4754.7A Rh-positive1922.1Multiple/pooled67.0A Rh-negative22.3O Rh-negative22.3Othera1011.6Total specified86100.0aother categories: A1; A Rh1; O +/A −; A −/A +; ORh +; ARh variants; O +/Multiple; B +; AB ABO/Rh distribution among studies that reported a blood group (n = 86) aother categories: A1; A Rh1; O +/A −; A −/A +; ORh +; ARh variants; O +/Multiple; B +; AB Experimental investigation examining inter-individual growth patterns A total of 69 malaria-naïve volunteers participated in the P. falciparum in vitro culture experiments. The cohort was predominantly female (55/69, 79.7%) and of Portuguese nationality (63/69, 91.3%), with a mean age of 26 years (range 18–56). Following 96 h of culture, parasite growth varied substantially among volunteers, with fold-change values ranging from 3.0 to 19.3 (mean 10.0, SD 3.8) (Fig. 2A). Based on a ± 1 standard deviation threshold, participants were stratified into three groups: 12 volunteers (17.4%) were classified as ‘low-growers’ (fold-change < 6.2), 13 (18.8%) as ‘high-growers’ (fold-change > 13.8), and 44 (63.8%) as ‘intermediate-growers’. These proportions closely mirrored those expected under a normal distribution. The observed growth patterns were stable over time. A strong positive correlation was found between the parasite fold-change in the first and second experiments on paired samples (Fig. 2B). Furthermore, longitudinal testing of blood from one ‘low-grower’ and one ‘high-grower’ volunteers over a three-year period consistently reproduced their respective growth phenotypes.Fig. 2Inter-individual variability in Plasmodium falciparum growth in red blood cells from different volunteers and its reproducibility. A The plot displays the fold-change in parasite growth after 96 h of in vitro culture using red blood cells from 69 healthy volunteers. Each dot represents the mean result for an individual volunteer. The horizontal lines indicate the mean fold-change and one standard deviation (SD) for the entire cohort. Volunteers were stratified into three groups based on these results: ‘low growers’ (orange dots, n = 12), whose RBC consistently supported growth below −1 SD from the mean; ‘high growers’ (blue dots, n = 13), whose RBC supported growth above + 1 SD; and ‘intermediate growers’ (black dots, n = 44). B The plot shows the correlation between P. falciparum fold-change from the first blood sample (Fold-change 1) and a second sample taken later (Fold-change 2) from the same 13 volunteers. Each dot represents one donor, color-coded by their original growth group: ‘low growers’ (orange, n = 4), ‘high growers’ (blue, n = 7), and ‘intermediate growers’ (black, n = 2). The strong linear correlation (R2 = 0.77) demonstrates that the capacity of RBC to support parasite growth is a stable, volunteer-specific trait Inter-individual variability in Plasmodium falciparum growth in red blood cells from different volunteers and its reproducibility. A The plot displays the fold-change in parasite growth after 96 h of in vitro culture using red blood cells from 69 healthy volunteers. Each dot represents the mean result for an individual volunteer. The horizontal lines indicate the mean fold-change and one standard deviation (SD) for the entire cohort. Volunteers were stratified into three groups based on these results: ‘low growers’ (orange dots, n = 12), whose RBC consistently supported growth below −1 SD from the mean; ‘high growers’ (blue dots, n = 13), whose RBC supported growth above + 1 SD; and ‘intermediate growers’ (black dots, n = 44). B The plot shows the correlation between P. falciparum fold-change from the first blood sample (Fold-change 1) and a second sample taken later (Fold-change 2) from the same 13 volunteers. Each dot represents one donor, color-coded by their original growth group: ‘low growers’ (orange, n = 4), ‘high growers’ (blue, n = 7), and ‘intermediate growers’ (black, n = 2). The strong linear correlation (R2 = 0.77) demonstrates that the capacity of RBC to support parasite growth is a stable, volunteer-specific trait To identify potential determinants of growth, a broad set of parameters was compared between the ‘low-grower’ (n = 12) and ‘high-grower’ (n = 13) groups. Four factors were significantly associated with the parasite growth phenotype. ‘Low-growers’ had significantly lower baseline haemoglobin concentrations than ‘high-growers’ (mean 12.3 g/dL vs. 14.8 g/dL; mean difference − 2.5 g/dL, 95% CI − 4.7 to − 0.3; p = 0.03), an effect of large magnitude (Cohen’s d = 1.1). Consistent with this, a personal history of anaemia was reported exclusively among ‘low-growers’ (5/12 vs. 0/13; p = 0.01), corresponding to a risk difference of 41.7% (95% CI 15.2–68.2%). Family history of anaemia was also more prevalent in the ‘low-grower’ group (8/12 vs. 3/13; p = 0.04; OR = 6.7, 95% CI 1.4–31.8). Lastly, ancestry from a malaria-endemic country was more frequent in the ‘low-grower’ group (6/12 vs. 1/13; p = 0.03; OR = 12.0, 95% CI 1.3–111.8). No significant associations were observed for other tested variables, including age, sex, ABO/Rh blood group, dietary intake, glycaemia or concurrent medication use. To investigate underlying mechanisms, a subset of 24 participants, stratified by their growth phenotype, underwent further haematological and biochemical testing. In this subgroup, the significant difference in haemoglobin concentration between ‘low-’ and ‘high-growers’ was confirmed (mean 12.7 g/dL vs. 13.9 g/dL, p = 0.04). However, correlation analysis revealed no significant linear relationships between parasite fold-change and individual laboratory parameters, including haematocrit, mean corpuscular volume, iron, vitamin B12 and folic acid (Fig. 3). Furthermore, comparisons of over 20 additional parameters showed no other statistically significant differences between growth groups. HPLC analysis confirmed normal adult haemoglobin (HbA) profiles in all participants, with no relevant elevation in fetal (HbF) or HbA2 percentages.Fig. 3Correlation between standard haematological parameters and Plasmodium falciparum growth. Scatter plots show parasite fold-change versus six different haematological and nutritional markers for the full subset of 24 volunteers. Each dot represents one individual volunteer, color-coded by their growth group: ‘low growers’ (orange, n = 6), ‘intermediate growers’ (black, n = 9), and ‘high growers’ (blue, n = 9). Linear regression analysis shows no significant correlation for any parameter, with all R2 values below 0.11 Correlation between standard haematological parameters and Plasmodium falciparum growth. Scatter plots show parasite fold-change versus six different haematological and nutritional markers for the full subset of 24 volunteers. Each dot represents one individual volunteer, color-coded by their growth group: ‘low growers’ (orange, n = 6), ‘intermediate growers’ (black, n = 9), and ‘high growers’ (blue, n = 9). Linear regression analysis shows no significant correlation for any parameter, with all R2 values below 0.11 Discussion In vitro culture of P. falciparum is a cornerstone of modern malaria research, enabling thousands of studies in drug development, parasite genetics and immunology. The human RBC is arguably the most critical component of this system [2]. Despite this centrality, the source and characteristics of RBC are often poorly documented. While research-grade RBC can be obtained from a few commercial suppliers, this is often at considerable cost, and access to it remains limited. For example, in the European Union, regulations prohibit the commercial sale of human blood [32]. This systematic literature review highlights the dimension of this issue. Of 314 studies that provided information on blood sources, a concerning 61.1% failed to provide sufficient detail to identify the RBC source (Table 1). Key parameters were frequently omitted, with only 27.4% of studies reporting the blood group, 6.0% specifying the anticoagulant used, and 7.3% mentioning direct phlebotomy from known donors. Nearly half of the studies did not clarify whether the RBC used were fresh or stored. These findings reveal that RBC sourcing is not only underreported, but also highly heterogeneous, therefore raising serious concerns for reproducibility and comparability of data across studies. Identifying this reporting gap was non-trivial. ‘Culture’ or ‘in vitro culture’ are not standardized MeSH terms, and simple keyword searches for ‘culture’ are confounded by irrelevant microbiological or sociological studies. This study demonstrates that LLM can effectively overcome such limitations. In line with recent evaluations showing LLM can achieve human-level performance in literature screening [33–36], a Python-based, multi-step classification pipeline was applied to screen nearly 1,000 abstracts and extract detailed RBC metadata from 451 full-text articles. The process was iterative; initial prompts often misinterpreted context (e.g., misclassifying fetal bovine serum as a non-human blood source) and required over a dozen revisions to develop a hierarchical strategy that could reliably extract the desired information. While not error-free, the LLM-assisted pipeline was sufficiently accurate to make this large-scale analysis feasible, enabling the extraction of methodological details buried deep within manuscripts. The lack of reporting is particularly concerning because variation in P. falciparum growth across blood donors is well-documented [3, 4, 29–31]. Previous work has shown that RBC from different malaria-naïve adults can yield a 0.5–2.5-fold variation in parasite invasion efficiency, a difference not explained by major surface receptors or ABO group [3]. Mechanistic analyses have linked part of this variability to intrinsic RBC properties such as hydration status and cellular deformability [9], while genomic studies indicate that variation in a small subset of RBC proteins may explain the majority of the differences observed in parasite fitness [37]. This study contributes to this evidence by characterizing P. falciparum growth in RBC from a relatively large and homogenous cohort of 69 malaria-naïve volunteers. Using a standardized 96-h culture protocol with the 3D7 P. falciparum parasites, growth difference ranging from 3.0 to 19.3-fold (mean 10.0, SD 3.8) were observed, representing a degree of variability that exceeds previously reported values. This growth phenotype was stable and reproducible for individual volunteers sampled over several years. Notably, this high variability occurred despite the demographic uniformity of the cohort (predominantly young healthy Portuguese females). This suggests that the biological relevance of inter-individual RBC differences is underestimated, with important consequences for research outcomes. The preferential use of group O RBC persists in many laboratories, despite limited evidence supporting its rationale. In the literature review, 54.7% of the 86 studies that reported blood group used type O, likely a legacy of historical protocols where ABO incompatibility was a genuine concern in human serum-containing media [2]. However, with modern serum-free systems, this concern is obsolete. While some studies report modest, strain-dependent invasion preferences for type O RBC [20, 21], the well-established epidemiological protection conferred by group O is due to reduced rosetting – a cytoadhesion phenomenon not relevant to merozoite invasion in in vitro cultures [20, 25, 26]. Consistent with this, no association was observed between ABO blood group and parasite growth in the present 69-volunteer study. Therefore, indicating that the continued preference for group O RBC is an outdated practice that unnecessarily restricts donor selection. While blood from transfusion services is anonymous, it comes from a relatively standardized, healthy donor pool (Table 3). In contrast, blood from volunteers allows for deeper profiling, but may carry a perceived biological risk as it is not screened for transfusion-transmissible infections. These findings, however, suggest that even extended donor profiling offers limited predictive value for identifying ‘high-growth’ or ‘low-growth’ RBC. The 69-volunteer cohort showed no significant associations between parasite growth and age, sex, blood group, glycaemia or medication use. Even detailed haematological and biochemical testing in a stratified subgroup of 24 individuals revealed no significant differences. Table 3Donor Screening Criteria and Inferred Characteristics of DonorCategoryParameterStandard screening criteriaInferred characteristicDemographicsAge and weightTypically, 18–65 years; weight ≥ 50 kg (110 lbs)Excludes RBC from paediatric, geriatric, and underweight individualsVital signsBlood pressure, pulse, temperatureNormotensive (e.g., 90/50–180/100 mmHg), regular pulse (50–100 bpm), afebrile (≤ 37.5 °C)Excludes donors with acute hemodynamic instability or febrile illnessHaematologyHaemoglobin (Hb) ≥ 12.5 g/dL (female donors); ≥ 13.0 g/dL (male donors)All donors are non-anaemic, with RBC mass within a narrow, healthy rangeInfectious disease screeningTransfusion-transmissible infections (TTIs)Mandatory non-reactive tests for HIV-1/2, HBV, HCV, and Syphilis. Regional screening for HTLV-I/II, T. cruzi, and othersRBC are sourced from donors free of major blood-borne viral and bacterial pathogensImmunohematologyBlood group and antibodiesABO/RhD typing performed on all units. Often screened for irregular antibodiesABO/Rh type is known. Alloimmunized donors are typically excludedMedical and travel historyHealth and exposure questionnaireDeferral for: travel to malaria-endemic regions; use of specific medications (e.g., anticoagulants, teratogens); high-risk behaviours (e.g., IV-drug use); recent live-virus vaccinationsExcludes donors with recent or latent malaria exposure and removes confounding effects of many medications or acute illnessesThis table summarizes typical inclusion and exclusion criteria for volunteer whole blood donors in high-income countries, based on guidelines from organizations like the World Health Organization (WHO) and national transfusion servicesThe screening process is designed to ensure recipient safety and results in a highly selected, healthy, and immunologically ‘clean’ donor pool. However, this standardization does not account for the intrinsic biological variability in red blood cells properties that can influence P. falciparum growth in vitro [41, 42]HBV, Hepatitis B Virus; HCV, Hepatitis C Virus; HIV, Human Immunodeficiency Virus; HTLV, Human T-lymphotropic Virus; RBC, Red Blood Cell Table 3Donor Screening Criteria and Inferred Characteristics of DonorCategoryParameterStandard screening criteriaInferred characteristicDemographicsAge and weightTypically, 18–65 years; weight ≥ 50 kg (110 lbs)Excludes RBC from paediatric, geriatric, and underweight individualsVital signsBlood pressure, pulse, temperatureNormotensive (e.g., 90/50–180/100 mmHg), regular pulse (50–100 bpm), afebrile (≤ 37.5 °C)Excludes donors with acute hemodynamic instability or febrile illnessHaematologyHaemoglobin (Hb) ≥ 12.5 g/dL (female donors); ≥ 13.0 g/dL (male donors)All donors are non-anaemic, with RBC mass within a narrow, healthy rangeInfectious disease screeningTransfusion-transmissible infections (TTIs)Mandatory non-reactive tests for HIV-1/2, HBV, HCV, and Syphilis. Regional screening for HTLV-I/II, T. cruzi, and othersRBC are sourced from donors free of major blood-borne viral and bacterial pathogensImmunohematologyBlood group and antibodiesABO/RhD typing performed on all units. Often screened for irregular antibodiesABO/Rh type is known. Alloimmunized donors are typically excludedMedical and travel historyHealth and exposure questionnaireDeferral for: travel to malaria-endemic regions; use of specific medications (e.g., anticoagulants, teratogens); high-risk behaviours (e.g., IV-drug use); recent live-virus vaccinationsExcludes donors with recent or latent malaria exposure and removes confounding effects of many medications or acute illnessesThis table summarizes typical inclusion and exclusion criteria for volunteer whole blood donors in high-income countries, based on guidelines from organizations like the World Health Organization (WHO) and national transfusion servicesThe screening process is designed to ensure recipient safety and results in a highly selected, healthy, and immunologically ‘clean’ donor pool. However, this standardization does not account for the intrinsic biological variability in red blood cells properties that can influence P. falciparum growth in vitro [41, 42]HBV, Hepatitis B Virus; HCV, Hepatitis C Virus; HIV, Human Immunodeficiency Virus; HTLV, Human T-lymphotropic Virus; RBC, Red Blood Cell Donor Screening Criteria and Inferred Characteristics of Donor This table summarizes typical inclusion and exclusion criteria for volunteer whole blood donors in high-income countries, based on guidelines from organizations like the World Health Organization (WHO) and national transfusion services The screening process is designed to ensure recipient safety and results in a highly selected, healthy, and immunologically ‘clean’ donor pool. However, this standardization does not account for the intrinsic biological variability in red blood cells properties that can influence P. falciparum growth in vitro [41, 42] HBV, Hepatitis B Virus; HCV, Hepatitis C Virus; HIV, Human Immunodeficiency Virus; HTLV, Human T-lymphotropic Virus; RBC, Red Blood Cell These results are consistent with growing evidence that routine clinical variables fail to capture the complex biology underlying RBC susceptibility to infection. Parasite invasion and growth depend on dynamic processes like membrane biomechanics, cytoskeletal remodelling and metabolic state [38], which are invisible to standard laboratory tests. For example, host-parasite interactions are influenced by donor-specific variation in hundreds of metabolites and protein modifications that dictate parasite ability to invade and replicate within the host cell [5–10, 20, 21, 39, 40]. Although most routine parameters were not predictive, a subtle signal related to anaemia did emerge. Volunteers with lower parasite growth rates presented significant lower haemoglobin concentrations (mean 12.3 g/dL vs. 14.8 g/dL; p = 0.03), a personal (5/12 vs. 0/13; p = 0.01) or family (8/12 vs. 3/13; p = 0.04) history of anaemia, or recent ancestry from malaria-endemic regions (6/12 vs. 1/13; p = 0.03). However, extended testing revealed no associations with iron status, folate, or vitamin B12. This suggests that anaemia and ancestry are not causal factors themselves but are likely clinical markers for underlying protective RBC phenotypes that are undetectable by standard haematology investigations. This interpretation aligns with known protective mechanisms. Iron-deficiency anaemia, common haemoglobinopathies (e.g., HbS, HbC), and the Dantu blood group polymorphism all inhibit parasite growth through mechanisms such as impaired invasion or altered membrane tension [6, 11, 37, 43]. These traits are enriched in populations with a history of malaria exposure, which is consistent with the ancestry signal observed in this study. Donor RBC are a core determinant of P. falciparum growth in vitro but remain a critically under-reported and poorly standardized variable. The combined literature review and experimental data demonstrate that donor-dependent growth variability is widespread, reproducible and biologically significant. This oversight has tangible implications, as differences in RBC source have been shown to cause inter-laboratory discrepancies in drug sensitivity assays (IC₅₀ values) and confound metabolic studies [44]. To address this critical gap and align malaria research with best practices in reproducible science, he adoption of the STRuBeC (Standards for Reporting use of (red) Blood Cells in Culture) framework is proposed. As detailed in Table 4, this six-point checklist provides a simple yet comprehensive guide for documenting the most crucial variables. By emphasizing parameters such as donor source, storage duration, and known genetic traits, STRuBeC shifts the focus from legacy variables toward those with a demonstrated impact on experimental outcomes. Analogous community-driven frameworks—such as MIQE for qPCR, ARRIVE for animal research, and CONSORT for clinical trials—have successfully transformed reporting standards in their respective fields with minimal burden on researchers [27, 28, 45]. In a similar spirit, adopting the STRuBeC checklist is a practical step to enhance the rigor and reproducibility of in vitro malaria research. Furthermore, the field would benefit from the reporting of a standardized growth score for each RBC lote, information that would allow researchers to better assess biological suitability and facilitate future meta-analyses [4, 39]. Table 4The STRuBeC Framework: Proposed Minimum Reporting Standards for Red Blood Cells in Blood-Stage Malaria CultureLetterParameterInformation to reportScientific rationale for reportingSSourceState whether RBC were from an anonymous blood bank or identified, consented volunteersDetermines the availability of donor metadata and the feasibility of longitudinal studies or repeat sampling from the same individualTStorage durationReport the time from phlebotomy to use (e.g., ‘used fresh within 6 h’, ‘stored 14 days at 4 °C’)The ‘storage lesion’ causes progressive metabolic and structural decay in RBC, directly impacting parasite viability and experimental outcomesRRelevant traits (if known)Disclose any known RBC traits of the donor (e.g., HbAS, G6PD deficiency, thalassemia trait)Shifts focus from legacy parameters to biologically meaningful variants that are known to significantly alter parasite invasion, growth, and drug sensitivityuUse of anticoagulant/additiveSpecify the anticoagulant or preservative solution used (e.g., CPDA-1, CPD, EDTA, heparin)Different formulations directly affect RBC metabolism, membrane stability, and ion balance, creating a key source of experimental variabilityBNumber of blood donorsState if cells from a single donor were used, or if cells from multiple donors were used across experimentsDonor-to-donor variability in parasite growth is substantial. This information is crucial for interpreting results and assessing their generalizabilityCKey processing stepsBriefly note critical RBC processing steps (e.g., ‘washed 3 × in RPMI’, ‘leukocyte-depleted’)Processing methods can alter the final cell preparation, affecting experimental comparability in ways that are independent of the initial RBC sourceSTRuBeC framework, a six-point checklist proposed to enhance the transparency, rigor, and reproducibility of in vitro malaria research. By encouraging standardized reporting on key red blood cells variables, the framework aims to address the significant but often undocumented donor-dependent variability that can impact experimental outcomes. CPD, Citrate–Phosphate-Dextrose; CPDA-1, Citrate–Phosphate-Dextrose-Adenine; EDTA, Ethylenediaminetetraacetic acid; G6PD, Glucose-6-Phosphate Dehydrogenase; HbAS, Haemoglobin AS (sickle cell trait); RBC, Red Blood Cell; RPMI, Roswell Park Memorial Institute (medium) Table 4The STRuBeC Framework: Proposed Minimum Reporting Standards for Red Blood Cells in Blood-Stage Malaria CultureLetterParameterInformation to reportScientific rationale for reportingSSourceState whether RBC were from an anonymous blood bank or identified, consented volunteersDetermines the availability of donor metadata and the feasibility of longitudinal studies or repeat sampling from the same individualTStorage durationReport the time from phlebotomy to use (e.g., ‘used fresh within 6 h’, ‘stored 14 days at 4 °C’)The ‘storage lesion’ causes progressive metabolic and structural decay in RBC, directly impacting parasite viability and experimental outcomesRRelevant traits (if known)Disclose any known RBC traits of the donor (e.g., HbAS, G6PD deficiency, thalassemia trait)Shifts focus from legacy parameters to biologically meaningful variants that are known to significantly alter parasite invasion, growth, and drug sensitivityuUse of anticoagulant/additiveSpecify the anticoagulant or preservative solution used (e.g., CPDA-1, CPD, EDTA, heparin)Different formulations directly affect RBC metabolism, membrane stability, and ion balance, creating a key source of experimental variabilityBNumber of blood donorsState if cells from a single donor were used, or if cells from multiple donors were used across experimentsDonor-to-donor variability in parasite growth is substantial. This information is crucial for interpreting results and assessing their generalizabilityCKey processing stepsBriefly note critical RBC processing steps (e.g., ‘washed 3 × in RPMI’, ‘leukocyte-depleted’)Processing methods can alter the final cell preparation, affecting experimental comparability in ways that are independent of the initial RBC sourceSTRuBeC framework, a six-point checklist proposed to enhance the transparency, rigor, and reproducibility of in vitro malaria research. By encouraging standardized reporting on key red blood cells variables, the framework aims to address the significant but often undocumented donor-dependent variability that can impact experimental outcomes. CPD, Citrate–Phosphate-Dextrose; CPDA-1, Citrate–Phosphate-Dextrose-Adenine; EDTA, Ethylenediaminetetraacetic acid; G6PD, Glucose-6-Phosphate Dehydrogenase; HbAS, Haemoglobin AS (sickle cell trait); RBC, Red Blood Cell; RPMI, Roswell Park Memorial Institute (medium) The STRuBeC Framework: Proposed Minimum Reporting Standards for Red Blood Cells in Blood-Stage Malaria Culture STRuBeC framework, a six-point checklist proposed to enhance the transparency, rigor, and reproducibility of in vitro malaria research. By encouraging standardized reporting on key red blood cells variables, the framework aims to address the significant but often undocumented donor-dependent variability that can impact experimental outcomes. CPD, Citrate–Phosphate-Dextrose; CPDA-1, Citrate–Phosphate-Dextrose-Adenine; EDTA, Ethylenediaminetetraacetic acid; G6PD, Glucose-6-Phosphate Dehydrogenase; HbAS, Haemoglobin AS (sickle cell trait); RBC, Red Blood Cell; RPMI, Roswell Park Memorial Institute (medium) Several limitations of this work merit consideration. The sample size for the detailed analysis (n = 24) limited the statistical power to detect modest effects. Genotyping for G6PD deficiency or α-thalassaemia was not performed, which are known to influence parasite growth and could contribute to some of the observed variability [6, 7, 46]. However, these conditions are relatively uncommon in the specific study population [47, 48] and are unlikely to account for the full spectrum of the observed growth differences. Finally, a single laboratory-adapted parasite strain (3D7) and one standardized culture medium were used; results may differ with field isolates or alternate culture conditions [49]. Conclusions Donor RBC are a critical yet under-reported determinant of P. falciparum in vitro growth. Based on the combined literature and experimental evidence, adoption of the STRuBeC framework is recommended as a minimum reporting standard to improve transparency and reproducibility in malaria research. Supplementary Information Supplementary Material 1 Supplementary Material 1