The Roman colony of Mariana, in the north-east of the island of Corsica, was established at the start of the 1st century BCE. It considerably evolved over time with the development of wetlands at the gates of the city, as demonstrated by recent geoarchaeological studies. The coastline, which is now 3 km from Mariana, has been subjected to a 400 m progradation, creating lagoon areas behind the dunes from late Antiquity onward. The Golo River, which now flows south of the site, has moved several times. During the early Middle Ages, one of its arms – if not its main course – ran alongside the southernmost buildings and a quay was built at the eastern end of the settlement. Finally, the Biguglia lake, a 1,450 ha body of brackish water located 3 km to the north, dried up during the 10–12th centuries, causing significant silting (20, 21) (Figure 2). As the only episcopal seat on the island, the city features several Christian buildings of worship. The southern basilica, erected at the beginning of the 5th century, has been restructured a number of times. It served as a cathedral until the construction of a second church, and was consecrated in 1,119 (22). A cemetery developed around the southern building from the late 9th or 10th century. The site was completely abandoned in the second half of the 15th century, in favor of the city of Bastia [(22), p.82 and 188–190]. Eighty-eight tombs, all located in and around the southern border of the church, were excavated in the late 1950s and early 2000s, yielding the archaeological materials investigated here. Tombs 72 and 18 were located in the central nave and the southern aisle, respectively. Tombs 40, 41, 44 were located in the vestibule (Figure 3). These two sectors were occupied by the episcopal palace from the second half of the 12th century [(22), p. 184–187]. All tombs were dug from the floor of the church within a large chronological range between the end of the 9th century and the second half of the 12th century. Tomb 18 was radiocarbon dated to between 1,034 and 1,215 (age calibrated to 2 sigma with 95% probability) (Lyon 2385, Age 14c BP: 900 ± 30) and a date between the second third of the 11th and the middle of the 12th century was accepted. Tombs 31, 34 and 35 were located outside and south of the church, in earth and building materials that were radiocarbon dated to the 10th and 11th centuries (T.17 and T.20) and covered by silt deposited before the 16th century [(22), p. 82]. The endemicity of malaria in this part of Corsica was historically attested at the beginning of the 20th century by a doctor who stated: “Only the sanitation of the plain which extends from Bastia to the Golo could reduce the frequency and even get rid of the malaria attacks which return periodically” (23).

The sex of adults was determined using the probabilistic sex determination (PSD) method based on metric data taken from the coxal bone (24). When this method could not be implemented, methods based on observing the morphological characters of the coxal bone were used (25). To estimate the age at death of immature individuals, different indicators were observed, including dental maturation (26–28). For adolescents, this relied on stages of bone maturation and/or stages of epiphyseal synostosis (29). For young adults, the stage of fusion of the epiphyses and clavicle sternal end were observed (30).

A total of 23 teeth (one tooth per individual) were selected from a total of 45 teeth taken from 27 different individuals, after discarding teeth in poor condition (Supplementary Table 1). These samples were a gift by the “Laboratoire d’Archéologie Médiévale et Moderne en Méditerranée” (Dr Corbara and Pr Istria) and “Anthropologie bio-culturelle, droit, éthique et santé” (Pr Costedoat and Pr Signoli). All teeth were handled under a hood in a paleomicrobiology laboratory, where no Plasmodium had previously been handled. The exterior of the tooth was disinfected using a sterile compress soaked with 100% ethanol followed by 9% w/v sodium hypochlorite (NaClO) to avoid any risk of contamination from the tooth surface. After decontamination, a fracture line was made along the length of the tooth, without extending to the pulp cavity, using a sterile diamond disc and the tooth was fractured into two using a sterile straight osteotome. The dental pulp was then scraped out using a dental excavator into a sterile Eppendorf tube and stored at −20°C until used (31).

A tooth collected from an unrelated archaeological site of Boulogne-sur-Mer (dated from the 18th century) devoid of any historical and anthropological pieces of evidence for malaria was used as negative control.

Paleoserum was extracted from the dental pulp as previously described (32). Briefly, remaining 100 μL of supernatant and debris pellet from lysis treatment for aDNA extraction were mixed with 200 μL of ethylenediaminetetraacetic acid (EDTA) was sonicated 5 × 1 min at 6,000 J and incubated overnight at 4°C under agitation. The mixture was then centrifuged for 30 min at 17,000 g at room temperature and the supernatant (supernatant no. 1) was stored at −20°C. The remaining pellets were washed twice with 200 μL of sterile distilled water and incubated for 48 h at 75°C in 100 μL of 50 mM ammonium bicarbonate solution, pH 7.4. The supernatant (supernatant no. 2) was then collected by centrifugation at 17,000 g for 30 min at room temperature and stored at −20°C. The two supernatants (nos. 1 and 2) were pooled and dialyzed using the Slide-A-Lyzer MINI Dialysis Device, 2 K MWCO (Pierce Biotechnology, Rockford, IL, USA) with one liter of dialysis solution (50 mM ammonium bicarbonate) for 4 h in the first round, and overnight in the second round. The resulting solution contained ancient proteins, including immunoglobulins (paleoserum), and was freeze-dried before being suspended in 30 μL of phosphate-buffered saline (PBS). Protein quantification was performed using a Qubit assay (Thermo Fisher Scientific, Waltham, MA, USA) for each paleoserum.

The dental pulp was scraped with a sterile scalpel into Eppendorf tubes and rehydrated for 24 h at room temperature in 500 μL of 1% formalin 50 vol, 96% ethanol, 30 vol and 5% ammonium bicarbonate, 20 vol, [adapted from Sandison et al. (33)]. Hydrated samples were homogenized and 100 μL of the suspension was cytocentrifuged for 7 min at 800 revolutions per minute (Shandon Cytospin 4 Cytocentrifuge, Thermo Electron Corporation, Waltham, MA, USA), to prepare four slides for each dental pulp sample (Supplementary Figure 1). One positive control slide was incubated with 200 μL of pooled of four leftover serum specimens diluted at 1:500 collected from patients exhibiting positive Plasmodium spp. parasitemia, one patient with 1.3% P. falciparum parasitemia, one with 0.8% P. vivax parasitemia, one with 0.2% P. malariae parasitemia, and one with 0.04% P. ovale parasitemia. A negative control slide was incubated with 200 μL leftover serum sample collected from a malaria-free patient at 1:500 dilution. A third slide was incubated with 200 μL paleoserum collected from the same individual as the investigated dental pulp at 1:20 dilution. A fourth slide was observed using electron microscopy (Supplementary Figure 1). The slides were incubated for 15 min with a permeabilization buffer containing 10% foetal bovine serum (FBS), 0.1% Triton X-100 and 0.01% saponin in PBS, washed three times with PBS, and incubated for 30 min with blocking buffer (1 XPBS, 10% FBS) followed by a PBS wash. One slide was then incubated overnight with a positive control serum mix diluted in 5% FBS-PBS at 1:500, a second slide with a malaria-free negative control serum diluted in 5% FBS-PBS at 1:500, and a third slide incubated with homologous paleoserum diluted in 5% FBS-PBS at 1:20. The slides were washed twice with 10% FBS-PBS for 5 min and incubated with secondary anti-human IgG Alexa-647 (Sigma-Aldrich, Saint-Quentin-Fallavier, France) diluted at 1:100 in 5% FBS-PBS and Phalloïdin Alexa-448 (Biotium, Fremont, CA, USA), diluted at 1:250 in 5% FBS-PBS for 1 h, and washed three times with PBS. After incubation at 20°C in a wet chamber, the slides were stained with Hoechst 3342 (Ref. 62249, Thermo Fisher Scientific) at 1:500 dilution in H₂O for 30 min to reveal nuclei before a final wash with sterile water. The slides were observed and analyzed by confocal laser scanning microscopy (Zeiss LSM800, Zeiss, Jena, Germany) using 405 nm, 488 nm, and 640 nm lasers to reveal Hoechst-3342-stained nuclei, Alexa-448 and Alexa-647 stained actin and anti-human IgG, respectively. An additional 561-nm laser observation was made to control for non-specific fluorescence.

Immunochromatographic assays detected the PfHRP-2 (histidine rich protein-2) antigen specific to P. falciparum, the pLDH (lactate dehydrogenase) antigen specific to P. vivax, and the PAN-pLDH antigen common to all Plasmodium species (pan species) (ACRO Biotech, Inc., Malaria, Montclair, CA, USA). Strips were extracted for quantitative reading using the FUSION FX chemiluminescent imaging system and the ImageQuant TL software (Witec AG, Sursee, Switzerland).

aDNA was extracted from ancient dental pulp along with two negative controls, one experimental negative control consisting in PBS in each batch, and one biological negative control consisting in one tooth from the 18th century. The aDNA extraction used in this study was previously described (34). In brief, 1 μL of TISS which is a synthetic extraction control (35) was added to each sample as an external extraction control. The dental pulp was resuspended in 1 mL of lysis buffer (0.45 M EDTA, pH 8.0 mixed with 0.25 mg/mL proteinase K in nuclease-free water) and incubated at 37°C for 16 h using a thermomixer (500 rpm). After 2 min of centrifugation at 16,000 g, 900 μL of supernatant was transferred into a 50 mL Falcon tube. Then, 7 mL of freshly prepared binding buffer (5 M guanidine hydrochloride), 40% (vol/vol) isopropanol, 0.05% Tween-20 and 90 mM sodium acetate (pH 5.2)) were added before vortexing. The remaining 100 μL was stored with the pulp debris to be used for further protein extraction, as above. A Qiagen MinElute Silica Spin column (QIAGEN, Hilden, Germany) was placed in a vacuum system to which 700 μL of mixture was added. The same volume was added every time the tube emptied, until the full 7 mL of sample had been adsorbed on the column. To remove the rest of the liquid, a 1 min 16,000 g centrifugation was carried out. After that, 700 μL of wash buffer (PE buffer, QIAGEN, Hilden, Germany) was added to the MinElute column to be washed following a 1 min 5,300 g centrifugation, and the contents of the collection tube were discarded. An additional 2-min 16,000 g centrifugation was performed to remove any trace of washing buffer. The column was placed into a sterile 1.5 mL LoBind Eppendorf tube and 12.5 μL of pre-warmed nuclease-free water was added directly to the membrane of the column, followed by 1 min of centrifugation at 16,000 g to elute DNA. The last step was repeated a second time, resulting in a final 25 μL volume of purified aDNA.

A first sequencing run incorporated pooled aDNA from the 23 teeth, normalized using the Qubit dsDNA High Sensitivity Assay Kit (Life Technologies, Villebon-sur-Yvette, France) and a dilution was performed to obtain 1 ng/uL. Sequencing was performed following the Illumina Nextera-XT paired-end protocol on the MiSeq instruments (Illumina Inc., San Diego, CA, USA). After tagmentation, a limited 12-cycle PCR amplification was performed to complete tag adapters and introduce dual-index barcodes. Genomic aDNA was then purified on AMPure XP beads (Beckman Coulter Inc., Fullerton, CA, USA). In addition, the libraries were normalized manually. We then pooled all the libraries into one for aDNA sequencing on MiSeq. The size of the library was determined using a 2100 Bioanalyzer and a High Sensitivity DNA Kit (Agilent, Santa Clara, CA, USA) and the concentration was determined by a Qubit. A sequencing blank was carried out throughout the preparation of the library as negative controls. A second sequencing run incorporated a selection of 14 individual aDNAs for NovaSeq™ 6000 (Illumina Inc.) sequencing, after a library was prepared as mentioned above. Library pooling (3 nM) and denaturation were performed following standard protocol A from the NovaSeq 6000 System Denature and Dilute Libraries Guide (document #1000000106351 v03), with a 15-h sequencing run performed with a NovaSeq 6000 SP Reagent Kit (100 cycles) v1.5 (Illumina, Inc.) Pre-processing sequencing data was decontaminated using BBduk tools from Galaxy Europe online software (Galaxy)¹ by removing the sequencing blank data, after adapter trimming using CLC Genomics Workbench. Taxonomic assignment was then performed using Kraken2, which allowed for the rapid taxonomic classification of metagenomic data based on the k-mer approach, with standard database plus protozoa and fungi 2021-05 (36). Kraken 2 results were then converted on Galaxy Europe with convert kraken data tools to translate the results of the Kraken metagenomic classifier to the full representation of NCBI taxonomy. Results were visualized with the Krona pie chart from the converted Kraken2 data. Data generated from the sequencing blank were analyzed using the same approach as a control, except they were decontaminated using BBduk tools. The same process was carried out with metagenomic datasets assembled with metaSPAdes on Galaxy Europe online software. We further investigated reads generated from NovaSeq™ 6000 instruments using a mapping approach against three reference genomes, P. falciparum, P. vivax and P. malariae, using CLC Genomics Workbench software (version 7.5) incorporating a 0.95 similarity fraction, a 0.8 length fraction, a deletion cost of 3, an insertion cost of 3, and a mismatch cost of 2. Mapped reads were then generated and blasted against the NCBI database to confirm their identity. The resulting reads were then decontaminated using BBduk tools from Galaxy Europe online software (Galaxy, see text footnote 1) by removing the blank sequencing data. The data has been deposited and is available on the European Nucleotide Archive website under the accession number PRJEB63329.

In Mariana, the individuals who tested positive for malaria died between the end of the 9th century and the second half of the 12th century (Supplementary Table 1). On this site, the funerary practices carried out in relation to the positive test subjects were globally similar, with individual primary deposits for the most part in simple pits. Some tombs stand out (T40, T41, T70) (Figure 3), both for their architecture and their use of masonry chests, as well as the fact that they were reused to bury a second individual or even several successive individuals. The position of these tombs in the internal sectors of the cathedral suggested that they housed the remains of people of privileged status. This observation was reinforced by the presence of clothing and finery in some tombs, denoting high social status. For example, in tomb T18, gold filaments associated with a bone pin were found on the skull of a young woman, and were probably part of a headdress (22, 37). In these skeletons, a 47% prevalence of non-specific inflammatory lesions raised the hypothesis of a systemic, chronic infection in these individuals. Periostitis lesions were observed in nine individuals and cribra orbitalia were noted for four individuals (Figure 4; Supplementary Table 1). Given the position of the Mariana site, located in the heart of a vast marshy area, malaria was suspected, especially since it has long been endemic in Corsica (37).

Laser scanning microscopy showed that slides which had been incubated with homologous (from the same individual) paleoserum yielded red after 640-nm laser excitation, consistent with pathogen antigen detection, blue after 405-nm laser, consistent with DNA detection, and green after 488-nm laser excitation, consistent with actin detection (Figure 5). The same observation was made with positive sample slides incubated with a positive control serum reacting with four Plasmodium species. Similarly, positive samples incubated with the paleoserum negative control showed no antibody signals in red and revealed only the presence of nuclei in blue and actin in green. Therefore, the paleo-auto-immunohistochemistry results were strongly indicative of the presence of Plasmodium antigens in Mariana dental pulp specimens (Table 1; Supplementary Table 1; Supplementary Figure 2). There was no presence of Plasmodium antigens in the negative control tooth.

Immunochromatographic assays detected Plasmodium spp. antigens in nine of the 23 individuals (39%) (Table 1; Figure 6). The three Pan-pLDH, PvLDH and PfHRP2 antigens were detected in individual T18, Pan-pLDH and PfHRP2 antigens were detected in individuals T31 and T44, and the PfHRP2 antigen was detected in individuals T9, T34, T35, T40, T41, T72, with all the negative controls remaining negative. Nine individuals were positive for the P. falciparum antigen and only three were positive for Pan-pLDH. This could indicate an active P. falciparum infection at the time of death of the individuals (38). Only one individual was positive for the P. vivax “PvLDH” antigen and only three individuals for the pan Plasmodium “Pan-pLDH” antigen. This could represent a period of replacement of non-falciparum malaria by P. falciparum malaria in this part of southern Europe. The antigenic detection for the tooth control from the 18th century was negative.

While both the experimental negative control (PBS) and the biological negative control (dental pulp) yielded no reads assignable to Plasmodium spp. (Supplementary Figure 3), a total of 71 reads assigned to Plasmodium spp. (Figure 7) were found in pooled dental pulps, while blank sequencing analyzed in parallel to the same sequencing run, did not give any reads assigned to the genus Plasmodium. A taxonomic overview of Novaseq sequencing reads assigned a majority of reads to bacteria (90–96%) while only 0.3–3% of reads were eukaryotic. Metagenomic sequencing results indicated the presence of Plasmodium assigned reads; however, the very low number of reads meant that the thorough identification and the exact assignment of the Plasmodium species were not possible, with reads assigned to Plasmodium spp. being different per sample, ranging from 12 to 176 reads. Regarding the classification of data after assembly by MetaSPades, Kraken2 taxonomic classification of the assembled runs after decontamination demonstrated a very low taxonomic assignment to Plasmodium sp., ranging from one to four, with a maximum of eight assignments for sample T8 and no assignments for two samples, T35 and T18. Mapping against reference genomes resulted a number of reads between the samples and ranged from 5,450 to 208,185 reads (Supplementary Table 2). After decontamination, only 0.26 to 1.28% of mapped reads remained. To obtain the taxonomic assignation of the remaining mapped reads, we performed at first taxonomic classification using Kraken2, followed by a manual Blastn on NCBI with the standard databases of Plasmodium spp. assigned reads. Only three individuals showed indications for Plasmodium DNA after a manual Blastn check, while most mapped reads were taxonomically assigned to humans (Supplementary Table 2).