Only Child

The non-redundant SILVA SSURef release 106 was downloaded in ARB-format from the SILVA website at http://www.arb-silva.de. Using the ARB software package [23] (link), we removed all sequences with a pintail score below 75, alignment quality score below 75 or length below 1,200 bp, in order to retain only high quality sequences. Further, we revised the taxonomy of several bacterial and archaeal taxa. The most significant improvements update the taxonomy of the Archaea to include the proposed phylum Thaumarchaeota[42] (link), [43] (link), the Actinobacteria to comply with Bergey’s Taxonomic Outline [44] , the Acidobacteria to incorporate proposed subgroups [45] and the Cyanobacteria to comply with the CyanoDB [46] and in some cases specific studies (details given in Supplementary Table S1). Other added taxa include the Zetaproteobacteria[47] (link), Rubritaleaceae[27] and Armatimonadetes[48] (link). In addition, we identified a number of taxa whose taxonomic annotation disagreed strongly with the topology of the SSURef alignment-based tree and appeared poorly supported by phylogenetic studies. These were either re-assigned to existing parent taxa or novel ones labeled incertae sedis. Unique taxon names were always used and to this end we added the name of the only child taxon to several unlabeled or undetermined taxa, or removed them.
Annotations of the eukaryotic taxa using the NCBI Taxonomy were taken from the SSURef database and manually verified in order to remove all sequences where taxonomical affiliation was in clear conflict with the topology of the alignment-based tree. Selection of fungal reference sequences was done according to recent phylogenetic work [49] (link), [50] (link).
All manual changes are listed in Supplementary Table S1, which can also be downloaded as a text file from http://services.cbu.uib.no/supplementary/crest/and is using an unambiguous format that can be parsed by the nds2CREST script (see below). In total, 82 new taxa were added, 123 were renamed and 17 deleted. All sequences remaining after curation were exported in FASTA format. During this procedure, sequences were cropped so as only the part corresponding to the SSU rRNA gene was saved. This was achieved by applying the Escherichia coli positional filter in ARB, selecting alignment column 1,o00 and 43,183. A tab-separated text file listing the accession numbers and taxonomic placements of each sequence was exported (using “NDS export”).
We developed the python script nds2CREST distributed together with the CREST LCAClassifier in order to convert the exported sequence and taxonomic data from ARB into configuration files for MEGAN [20] (link) and the CREST LCAClassifier. This script also reads a text version of the Manual Changes File (MCF; Supplementary Table S1). For each change specified in the MCF, it confirms that the change was properly carried out. In addition, the script removes all sequences without valid taxonomical annotation or specified to be removed in the MCF. After this procedure, it assigns taxonomic ranks for each taxon based primarily on the NCBI Taxonomy, where such information is available; secondarily on the name of the taxon using the suffices “-ales” and “-acaea” to indicate family or order level, respectively; and lastly based on the parent rank. The output of nds2CREST is (1) a tree-file in Newick format describing the topology of the taxonomy, (2) a tab-separated “mapping file” specifying the name and rank for each taxon, and (3) a reference sequence database in FASTA-format. In addition to SilvaMod, we also prepared such files from the Greengenes Taxonomy [21] (link) using the same procedure, however without manual curation or positional filtering.

Summary details of samples, SNPs, CpGs and covariates included in association tests are presented in Additional file 1: Table S2. Each SNP in the imputed datasets was analysed against all CpG sites in the Illumina Infinium HM450 array with the exception of those failing QC and those reported to map to more than one location (N = 19,834) or to contain a genetic variant at the CpG site (N = 74,182) [34 (link)]. We performed a post hoc test for analysis of non-specific probes which determined that there was no enrichment of cross-hybridising probes in our results. We opted for post hoc annotation of potential probe effects within our results database for the remaining potentially problematic CpG sites (N = 229,983) [34 (link)]. The final number of probes analysed was 395,625. Preliminary association analysis of SNPs with CpG sites was performed using an additive model (rank-normalised CpG methylation on SNP allele count) using Matrix eQTL [39 (link)] in order to perform the computationally demanding task of estimating 16 trillion associations. Analyses were batched by SNP chromosome and sample time point for parallel analysis on the University of Bristol High Performance Computing (HPC) cluster. SNP effects from this analysis that were p < 1 × 10⁻⁷ were then taken forward for re-analysis in PLINK1.07 to perform exact linear regression including covariates. Covariates included in all analyses were age (excluding birth), sex (children only), the top ten ancestry principal components, bisulphite conversion batch and estimated white blood cell counts (using an algorithm based on differential methylation between cell types [33 (link)]). Methylation at each CpG site was regressed on these covariates and residuals taken forward for regression on SNP genotype. To report the number of mQTL, we used a conservative threshold of 1 × 10⁻¹⁴. All associations below 1 × 10⁻⁷ were stored and are available in our online mQTL database (http://www.mqtldb.org/). Analysis on rank-normalised data results in effect sizes that are not directly interpretable on the original scale. Additional file 1: Figure S2 illustrates the effect size distributions observed in our data.
Without access to an appropriate replication sample, we performed all subsequent enrichment and downstream analysis using only cis mQTL with p < 1 × 10⁻¹⁴ (Additional file 1: Table S3) to reduce the possibility of including false positives (unless otherwise stated). We consider results below p < 1 × 10⁻¹⁴ in a single time point to provide informative evidence of association on the basis that this corresponds to a 0.2 % false positive rate after a Bonferonni correction for the number of tests based on directly genotyped SNPs and directly assayed CpG sites. We define replication in additional time points to be associations at p < 1 × 10⁻⁷ because typically we are testing in the order of 30,000 associations for replication and on the basis that these are supported by their combination with the evidence at other time points through the life course.

All febrile (or those with history of fever in the last 48 h) participants admitted at the emergency ward and pediatric emergency ward of all ages were referred to the study team and seen on admission by a clinical staff of the study. Summary data were recorded on a pro-forma sheet. Only patients presenting with positive falciparum malaria test with clinical signs of severe illness (as described above, according to the WHO criteria [8 ]) were enrolled. These patients were examined by a physician and their data captured on a standardized form. The study was authorized by the Director of the hospital and approved by the Gabonese National Ethics Committee (PROTN°23/2019/PR/SG/CNER). Only children whose parents gave informed written consent were enrolled in the study. The rainfall dataset for this study was obtained from the Agence pour la sécurité de la navigation aérienne en Afrique et à Madagascar (ASECNA).Venous blood was collected for all enrolled patients for the following medical exam.

The study recruited participants from a convenience sample of healthy and sick children (8–12 years) and adolescents (13–17 years) in urban Blantyre, Malawi. Children and adolescents attending schools and seeking any health care services through out-patient department at the Queen Elizabeth Central Hospital made up healthy and sick participants, respectively. Written assent and consent was obtained from children and their parents/guardians. For sick participants, the invitation came at the end of clinical care. For healthy participants, invitations were made through the school via a teacher. Participants took the study information leaflets and consent forms home for receipt of consent by their respective parents/guardians and these were brought back to the school the following day. For both sets of participants, once consent was obtained, the questionnaires were distributed by the research team at the end of clinical care or interviews were arranged on a school day. Once the participants completed the questionnaires (in clinic or classroom settings, respectively), the forms were handed over and collected by the study staff. Only children who were literate (as evident from the written consenting process) and therefore able to self-complete the questionnaires were included, but the critically ill were excluded from recruitment. As previous research had revealed a tendency for respondents to avoid the middle responses when completing the adult EQ-5D-5L questionnaire if the EQ-5D-3L is administered first [3 (link)], the EQ-5D-Y-5L was administered before the EQ-5D-Y-3L. This was followed by the self-report Pediatric Quality of Life (PedsQL)™ 4.0 Generic Core Scales for children (8–12 years) or teens (13–17 years). Ethical approval for this study was granted by Ethics Committees at the Malawi College of Medicine (now KUHeS) (P.10/18/2509) and Liverpool School of Tropical Medicine (19-045). A sample size of 200 participants was calculated to provide 80% power, at the two-sided significance level of 0.05, to address the minimum psychometric criteria for convergent and discriminant validity.