Intestinal protozoan infections shape fecal bacterial microbiota in children from Guinea-Bissau

Intestinal parasitic infections, caused by helminths and protozoa, are globally distributed and major causes of worldwide morbidity. The gut microbiota may modulate parasite virulence and host response upon infection. The complex interplay between parasites and the gut microbiota is poorly understood, partly due to sampling difficulties in remote areas with high parasite burden. In a large study of children in Guinea-Bissau, we found high prevalence of intestinal parasites. By sequencing of the 16S rRNA genes of fecal samples stored on filter paper from a total of 1,204 children, we demonstrate that the bacterial microbiota is not significantly altered by helminth infections, whereas it is shaped by the presence of both pathogenic and nonpathogenic protozoa, including Entamoeba (E.) spp. and Giardia (G.) lamblia. Within-sample diversity remains largely unaffected, whereas overall community composition is significantly affected by infection with both nonpathogenic E. coli (R2 = 0.0131, P = 0.0001) and Endolimax nana (R2 = 0.00902, P = 0.0001), and by pathogenic E. histolytica (R2 = 0.0164, P = 0.0001) and G. lamblia (R2 = 0.00676, P = 0.0001). Infections with multiple parasite species induces more pronounced shifts in microbiota community than mild ones. A total of 31 bacterial genera across all four major bacterial phyla were differentially abundant in protozoan infection as compared to noninfected individuals, including increased abundance of Prevotella, Campylobacter and two Clostridium clades, and decreased abundance of Collinsella, Lactobacillus, Ruminococcus, Veillonella and one Clostridium clade. In the present study, we demonstrate that the fecal bacterial microbiota is shaped by intestinal parasitic infection, with most pronounced associations for protozoan species. Our results provide insights into the interplay between the microbiota and intestinal parasites, which are valuable to understand infection biology and design further studies aimed at optimizing treatment strategies.

In the present study, we explored alterations in the fecal bacterial microbiota due to intestinal parasitic infections in a large cohort of children from the capital of one of the poorest countries in the world, Guinea-Bissau, Western Africa. From here on the bacterial microbiota studies is referred to simple as the microbiota. Using this cohort, we have recently investigated the prevalence of intestinal parasitic infections in both health-care seeking children (referred to as cohort I) and children from the background population (referred to as cohort II), all aged 2-15 years and found that infections were highly prevalent in both cohorts [27]. In 566 children in cohort I and 708 in cohort II, we found that prevalence of intestinal helminths was 13.8% and 9.6%, respectively (Fisher's exact test between groups, P = 0.021), whereas prevalence of intestinal protozoa was 41.5% and 46.0%, respectively (Fisher's exact test between groups, P = 0.112).
Helminth infections were mainly due to hookworms, and protozoan infections were dominated by both pathogenic and non-pathogenic species, including Entamoeba coli, Entamoeba histolytica/dispar and Giardia lamblia. Upon fecal parasitological investigation by microscopy, fecal samples were applied to lter papers, and kept at ambient temperature. By high-quality 16S rRNA gene sequencing (> 10,000 reads per sample) of fecal samples from 1,204 of these children, we demonstrate that the fecal microbiota signi cantly associates with intestinal parasitic infection, and that the association is stronger for children infected with protozoa compared to helminths. Further, we here demonstrate that the fecal microbiota from samples stored at ambient temperature on lter (fecal occult blood test, FOBT) papers for up to 1,000 days can facilitate largescale microbiome studies in remote areas. We previously validated the use of lter paper in a benchmark study that show high resemblance of fecal microbiotas from lter paper with conventionally stored samples, with satisfying DNA yield and sequencing quality [28]. Thereby, we demonstrate that long-term storage on FOBT papers is an applicable approach for large-scale sample collection in eld settings, where immediate freezing of samples is not possible.
To our knowledge, this is the largest study to date investigating the relationship between intestinal parasites and alterations of the fecal microbiota. This exploratory study should enable us to move forward with targeted questions towards understanding the role of the microbiota in intestinal parasitic infections and infection-associated complications. This may lead to new therapeutic strategies, as manipulation of the microbiota by the administration of probiotics might be an effective way to enhance the host's immune system to control the parasite. Furthermore, expanding the knowledge may help to differentiate between bene cial and pathogenic intestinal parasites.

Cohort characteristics and parasite prevalence
The dataset includes microscopic investigation for intestinal parasites from 1,274 children, included between August 2015 and April 2017 in urban Bissau, Table 1 Cohort characteristics and intestinal parasite prevalence. Characteristics of cohort I (n=529) and cohort II (n=675), which were included in the microbiota analysis. Between-group differences are calculated using Wilcoxon rank sum test, Fisher's exact test or Kruskal-Wallis equality-ofpopulations rank test, when appropriate.  25.6%, P = 0.041, respectively). Infection with multiple species was equally common in the two cohorts; 14.0% of children from cohort I and 13.6% from cohort II were infected with two parasite species, and 3.0% in cohort I and 2.2% in cohort II were infected with three or more parasite species.
The major intestinal helminth species found in the study participants were Ancylostoma duodenale (hookworm, n = 94) and Hymenolepis nana (dwarf tapeworm, n = 36). The most common intestinal protozoan species found in the participants were Entamoeba coli (n = 98), Entamoeba histolytica/dispar (n = 214), Giardia lamblia (n = 285) and Endolimax nana (n = 94). A number of other intestinal parasite species were found at lower prevalence, but were not included in the present study due to lack of statistical power. Prevalence and distribution of the major intestinal parasites in the two cohorts and statistical differences between groups are provided in Table 1.

Identi cation of confounding variables
Several host phenotypic and environmental variables have previously been linked to differences in the gut microbiota composition, including age, diet, vitamin supplementation and antibiotic treatment (as reviewed in [29]). Thus, we excluded all individuals with a history of antibiotics use three months prior to inclusion from the analysis, and adjusted for the potential confounding effect of age, history of vitamin A supplementation (binary variable, ((i) yes or (ii) no), toilet source (binary variable, either (i) no private toilet/latrine or (ii) access to private latrine/toilet) and tropical season for sample collection (binary variable, (i) rainy or (ii) dry season). Further, as samples were kept at room temperature on FOBT cards prior to DNA extraction, we adjusted for sample storage time (in days). As there were some differences in characteristics and parasite prevalence between the two cohorts (Table 1), all analyses were performed for both cohorts separately, and jointly, the latter adjusted for cohort status. In the following sections, the reported results are from the joined analysis if not stated otherwise.

Alpha diversity is largely unaffected by intestinal parasite infection
Relative abundance of bacterial phyla across all samples revealed that the composition of the microbiota in all study participants was dominated by Firmicutes and Bacteroidetes, as expected (Fig. 2).
Three different measures of alpha diversity were calculated to explore possible diversity alterations due to intestinal parasitic infections: Shannon entropy; ACE as a measure of species richness; and phylodiversity as a measure of total unique phylogenetic branch length (Table 2). We compared alpha diversity measurements for individuals with each of nine infection variables (either overall parasite positive, helminth positive, protozoa positive or positive for one of the six speci c species, i.e. Ancylostoma duodenale, Hymenolepis nana, Entamoeba coli, Entamoeba histolytica/dispar, Giardia lamblia or Endolimax nana) against non-infected individuals. Effects on all three diversity indices were limited for all infections, as illustrated in Fig. 3 for phylodiversity (Table 2 and Fig. 3). A signi cant decrease in ACE diversity was seen in participants in cohort I with Giardia lamblia infection (β= -3.42; P = 0.0470), which was not found in cohort II. Increase in phylodiversity index was seen in cohort II with any intestinal parasite and protozoa (β = 7.00; P = 0.0153 and β = 6.73; P = 0.0266, respectively). Further, all three alpha diversity indices were increased in cohort II with Entamoeba spp., which were not observed in cohort I. In the pooled dataset containing both cohorts, a signi cant increase in phylodiversity was seen upon infection with Entamoeba spp. (β = 11.35; P.adj.=0.0115 for Ent. coli and β = 8.15; P.adj.=0.0115 for Ent. histolytica/dispar), however only nominally signi cant in cohort II. Therefore, changes in alpha diversity was predominantly observed in cohort II, and predominantly for phylodiversity. The changes were mainly an increased diversity in infected individuals in agreement with ndings reported in [25]. Table 2 Intestinal parasitic infections have limited effects on alpha diversity. Three different alpha diversity indices were analyzed for association with infection status using robust regression (Methods). The table show summary statistics: P value for analysis within participants from cohort I, within participants from cohort II and across all participants (Total), Benjamini-Hochberg adjusted P value for analysis across all individuals (P.adj), and association coe cient (Beta Highly signi cant association between protozoa infection and microbiota composition The association between the parasite infection variables and microbial community structure (Bray-Curtis dissimilarity) was evaluated using a permutational multivariate ANOVA-like approach (adonis in R package vegan with 9999 permutations) ( Table 3). For overall variables, the most pronounced association was seen for protozoa (R 2 = 1.03⋅10 − 2 ; P = 1.00⋅10 − 4 ) followed by any parasite (R 2 = 9.73⋅10 − 3 ; P = 1.00⋅10 − 4 ) while association for any helminth was insigni cant (R 2 = 2.40⋅10 − 3 ; P = 6.61⋅10 − 2 ). Therefore, the association between overall parasite infection and the microbiota may be driven by protozoan infections, as further supported by the similar appearance of the ordination-based visualization of overall parasite and protozoa infections in Fig. 4. Infection load was determined by microscopy and ranked by the number of different parasite species identi ed in each sample. Most individuals were infected with either none (0) or one (1) parasite species, and with decreasing prevalence two, three or four different species (Table 1). Ordination based evaluation demonstrated that increasing infection load showed an increasing shift in the microbiota community away from the composition of the uninfected individuals (infection load 0), suggesting that multispecies infections induced more pronounced alterations than mild ones (Fig. 4), and supported by highly signi cant association with microbiome composition (adonis, R 2 = 0.012, P = 0.0001).

A large portion of tested bacterial taxa are associated with intestinal protozoan infection
Associations between the parasite infection variables and the relative abundance of individual taxa was evaluated for the phylogenetic level phylum to genus ( Table 4, Fig. 5). As mentioned, only minor effects on the fecal microbiota beta-diversity were observed in individuals with helminth infections. Accordingly, we observed only few associations regarding speci c taxa of the fecal microbiota and helminth infections; for the overall helminth variable no associations remained signi cant at genus level (P.adj.<0.05) while the Epsilonproteobacteria-Campylobacterales-Campylobacteraceae branch was increased in abundance for individuals infected with Ancylostoma duodenale. However, Campylobacter abundance was broadly increased across individuals infected with protozoa, thereby causing the association with Ancylostoma duodenale to be non-speci c. Beyond Campylobacter, the only genus associating with helminth infection was Collinsella that associated with Hymenolepis nana infection (β= -0.48; P.adj.= 3.22⋅10 − 2 ). Table 4 Alterations of individual genera due to intestinal protozoa infections. A total of 31 genera from the four major phyla of the gut microbiota were either found to regression, P.adj< 0.05). The Campylobacter genus was found at increased abundance due to any of the protozoan infections. The From the 43 genera analyzed, a total of 31 genera from the four major gut microbiota phyla associated with protozoa infection, either overall protozoa positive, or for individual species (P.adj.<0.05) ( Table 4, Fig. 5). Most genera were found to be altered by overall protozoa infection (10 with increased abundance, 16 with decreased abundance), followed by infection with Entamoeba histolytica/dispar (6 with increased abundance, 15 with decreased abundance), Entamoeba coli (5 with increased abundance, 10 with decreased abundance), Endolimax nana (6 with increased abundance, 7 with decreased abundance) and Giardia lamblia (3 with increased abundance, 7 with decreased abundance).
Of the four tested genera within the Gammaproteobacteria class, three were found to be less abundant in individuals with protozoan infections, namely Escherichia/Shigella, Klebsiella and Haemophilus, while the fourth, Succinivibrio, was increased. In summary, a large portion of the analyzed genera associated with protozoa infection, while only few taxa were signi cantly associated with helminth infection. All major phyla normally found in the human gut microbiota were represented across associated genera.
Storage time associates with small but signi cant differences in microbiota composition Due to challenges associated with collecting fecal samples in Guinea-Bissau, including unstable electricity supply and limited infrastructure to secure adequate storage at freezing temperatures, fecal samples were collected on FOBT paper and stored at ambient temperature for a longer time period. Existing studies of this procedure has reported a general good performance of the FOBT papers [30][31][32][33][34]. We have recently demonstrated that the storage method used in this study provide microbiota results that are very similar to conventionally stored samples kept at -80°C [28]. Within the present study, both overall DNA and sequencing quality from FOBT paper samples was satisfactory, and microbiota data resembled that of a microbiota from a conventionally stored sample.
However, to limit potential confounding effects of storage time, days of storage was included as a covariate in the analyses.
We explored changes in phylogenetic diversity, beta diversity and compositional alterations of selected bacterial taxa with increasing storage time. We observed a relative increase in Bacteroidetes abundance and a corresponding decrease in Firmicutes with increasing storage time (Fig. 6A). The decrease in Bacteroidetes seemed to be driven by a decreased abundance of Prevotellaceae at family level (Fig. 6B), and by Prevotella at genus level (Fig. 6C). Alpha and beta diversity associated with storage time (Spearman rho − 0.097, P = 0.00077, and adonis R 2 = 0.048, P = 0.007, respectively). Furthermore, correlation analysis of the relative abundance of selected taxa demonstrated a general signi cant association with storage time, however with limited average change in abundance (strongest coe cient (rho) = 0.23 observed in the Firmicutes phylum, P-values ranging from 4.0⋅10 − 9 to 6.80⋅10 − 1 for the investigated taxa) (Supplementary g. S1).

Discussion
In the present study, we analyzed fecal microbiota composition from 1,204 children from Bissau, Guinea-Bissau, with an overall high prevalence of intestinal parasitic infections, predominantly caused by protozoans including Entamoeba spp. and Giardia lamblia. We demonstrated that microbial alpha diversity was largely unaffected by helminth infections, and that protozoan infections had moderate effects on alpha diversity. We demonstrated that beta diversity associates with infection status for both pathogenic and non-pathogenic protozoa, and that the abundance of a total of 32 bacterial genera were altered due to parasite infections. We proved that the microbiota from fecal samples stored at room temperature on FOBT papers resembled that from conventionally stored samples, and that FOBT papers were useful in eldwork with lack of freezing capacity.
We found a total of 31 genera from four different phyla, out of 43 genera analyzed, to signi cantly associate with intestinal protozoan infection. These include a decrease of the Collinsella genus in individuals infected with Entamoeba histolytica/dispar and Giardia lamblia. To our knowledge, no previous studies have associated this genus with intestinal parasitic infections, however, the Actinobacteria phylum has recently been demonstrated to be increased upon infection with Trichuris trichiura in humans [24]. Collinsella spp. have been demonstrated to regulate levels of circulating insulin in pregnant women [35], and a reduced abundance has been associated with symptom severity in patients with irritable bowel syndrome [36]. A common and debilitating feature of Giardia infection is the post-giardiasis syndrome after complete elimination of the parasite, with a symptomatology very much alike irritable bowel syndrome [37,38]. One possible explanation for these long-lasting post-infectious symptoms could be an altered microbiota, including decreased abundance of Collinsella spp. However, as Giardia lamblia resides in the small intestine only [39], an effect seen in the fecal microbiota may be immunological derived, rather than by local interactions. In mice, Giardia infection has been reported to increase the abundance of Proteobacteria in the fore-and hindgut [40]. This is contradictory to our observations, where Proteobacteria itself was not associated and two of three genera in the clade were decreased upon Giardia infection.
We found a decreased abundance of the Bacteroides genus due to infection with Entamoeba histolytica/dispar. There are con icting results regarding gut microbiota alterations due to Entamoeba histolytica infection, as both increased [25] and decreased [41] abundance of Bacteroides has previously been observed. The two studies in question are conducted in Zimbabwe and India, respectively, and the con icting ndings could be due to geographical changes in gut microbiota. Further, there are differences in the applied technique, as one is based on sequencing, and the other on targeted PCR. We further found a decrease in Lactobacillus spp. due to infection with Entamoeba histolytica/dispar, which is consistent with previous ndings [41].
We found limited (single taxa and beta-diversity) or no (alpha-diversity) effects on gut microbiota composition due to helminth infections. This is contradictory to previous studies on the subject, in which several bacterial taxa have been associated with infection. Regarding hookworm infection, previous studies have demonstrated an increase in Bacteroidetes and a decrease in both Lachnospiraceae and Firmicutes [24,42]. The number of hookworm-infected individuals in these studies vary between 8 and 55, compared to a total of 94 in the present study, and the differences could thus be explained by a lack of power in the previous studies, as these hold an increased risk of false positive ndings. Furthermore, helminth prevalence is minor in the present study, compared to protozoa, and may partly explain why less pronounced effects due to helminth infections are observed. Further, regional and geographical differences in gut microbiota composition is another plausible explanation for the lack of uniform results. To our knowledge, no other studies have investigated gut microbiota alterations due to infection with Hymenolepis nana.
Historically, all protozoa and helminths were considered parasitic, and assumed to be pathogenic. As re ected by the number of prevalent cases and related morbidity worldwide, this is indeed true for some species. A distinctive feature of many intestinal parasitic infections is that they cause signi cant morbidity, and less pronounced mortality. For instance, STH infections, which especially affect children, may cause nutritional de ciency, which may lead to anemia and ultimately reduced growth and cognitive development [2,3,43]. With regards to some intestinal parasites, infection can be life-threatening and even fatal, as seen in hyperinfection syndrome of Strongyloides infection, bowel obstruction in Ascaris infection, or invasive amebiasis by Entamoeba histolytica infection [44][45][46]. Although some intestinal parasites may cause pronounced pathology in humans, evaluation of the existing literature indicates that many common eukaryotic species within the human gut, originally identi ed as pathogenic parasites, are actually commensals or even bene cial, at least in part, and could be regarded as pathobionts, only causing disease in certain contexts [22,47,48]. Some even extend this to state that eukaryotic members of the microbiota (termed the eukaryome or parasitome) are crucial in maintaining gut homeostasis and shaping host immunity [49], and that consequently, absence of e.g.
helminths may result in a dysfunctional immune system [14], which partially explain the rise in autoimmune diseases seen in the industrialized world [50]. The most prevalent intestinal parasites found in the present study are pathogenic, with the exception of the amebic species Entamoeba coli and Endolimax nana, which are generally accepted as being non-pathogenic. We demonstrate that there are only minor differences between the two cohorts (children seeking medical attention and children found in the background population, respectively) with regard to parasite prevalence, and this difference is due to helminth infections, which are dominant in cohort I. By so, our ndings support that the presence of intestinal parasites does not necessarily cause individuals to seek medical attention, even though the microbiota of these individuals is altered.
One major limitation of our study is the somewhat old-fashioned and rough methods for detecting intestinal parasites. In the industrialized world, conventional light microscopy has largely been replaced by molecular diagnostics including qPCR, which has proven to be superior to microscopy with increased sensitivity and speci city [51]. However, in developing countries and eld settings, laboratory access is sparse, and microscopy remains a cheap, fast and reproducible method for parasite examination, with acceptable sensitivity and speci city [52]. One could argue that an increased sensitivity by qPCR would result in overdiagnosis and detect intestinal parasites at much lower abundance, thereby limit the effects seen in the fecal microbiota. As the parasites are detected by relatively insensitive light microscopy, it is safe to assume that the parasite burden in positive samples is clinically relevant, which increases the value of our ndings regarding fecal microbiota alterations, as these are not driven by clinically irrelevant infections.
Another methodological limitation by microscopy diagnosis is the failure to differentiate between pathogenic and potentially fatal Entamoeba histolytica and non-pathogenic Entamoeba dispar, as the two are indistinguishable by microscopy [53,54]. The high prevalence of Entamoeba histolytica/dispar found in our study may very likely resemble Entamoeba dispar, and the described associations with fecal microbiota may not be due to Entamoeba histolytica. Microbiota alterations due to infection with Entamoeba histolytica has previously been investigated in Cameroon, where an infection-dependent increase in Bacteroidetes was reported [25].
Due to the sequencing approach to investigate the fecal microbiota, we were unable to detect alterations on strain-level. Furthermore, as the two genera Escherichia and Shigella (within the Gammaproteobacteria class) have very similar 16S rRNA gene sequences, we were unable to differentiate between the two. Both genera are related to gastrointestinal pathology, and differentiation between the two by other approaches would possibly yield interesting aspects.
The use of samples stored on FOBT paper at room temperature could be considered both as a strength and a weakness of the present study. Optimal sample storage has been reported to be among the most important issues in microbiota research, and compositional changes might occur over a relatively short time.
General recommendations include immediate freezing within 15 minutes [55] to 24 hours [56] from defecation, and storage at -80 °C is regarded as gold standard [57]. However, several studies, including our own benchmark study, have demonstrated that the use of FOBT lter papers is a valid approach, which does not induce major compositional shifts [33]. However, all analysis in the current study were adjusted for storage time, in order to further ensure biologically reliable results. Amir et al. have found that speci c bacterial taxa grow at room temperature, which could induce confusing results. Especially genera within the Gammaproteobacteria class are reported to bloom [58]. However, as we previously demonstrated that this class was unaffected by room temperature storage for ve weeks on FOBT papers [28], we judged that adjusting for storage time accommodated adequately for any potential confounding effect of FOBT storage (Table 4, Fig. 5). Although we demonstrated some signi cant alterations due to room temperature storage, these were minor, and do not argue against the use of FOBT cards in eldwork without electricity. What seems to be important is a uniform sample collection, and that analyses should be adjusted for storage time.

Conclusion
In the present study, we demonstrate that microbiota assemblages are signi cantly associated with intestinal protozoa infections, whereas limited or no effects are seen due to helminth infections. We nd that speci c taxa associated with different protozoan infections, which can lead to improvement of treatment strategies by probiotic supplementation, and further enhance our understanding of the interplay between host microbiota and intestinal parasitic infections.

Sample collection and storage
Stool samples were collected as a part of a prospective no-intervention two-cohort study, investigating the prevalence and potential risk factors for intestinal parasite infections in children from Bissau, Guinea-Bissau, Western Africa. The study area and sample collection procedure has been described in detail previously [27]. In brief, children aged 2-15 years were included between August 2015 and April 2017 at local health centers (cohort I) or at their private address (cohort II). Upon inclusion, participants delivered fresh stool samples in designated sterile containers, which were kept in a refrigerator prior to microscopic analysis for intestinal parasites. Microscopic parasitological analyses were performed following the local routine, and infection load was determined by the number of different species identi ed. Upon microscopic investigation, the fecal sample was manually homogenized within the container, and approximately 0.5 mL of the sample was applied to the two lter paper windows of a fecal occult blood test (FOBT) lter card (Hemoccult®, Beckman Coulter) with a clean wooden spatula. The sample was air-dried under laminar air ow, protected from sunlight, for 1-6 hours, after which the sample was packed in an individual airtight zip lock bag with desiccant (Whatman® desiccant packs, Sigma-Aldrich). Samples were subsequently stored in the dark at ambient temperature, which is approx. 25 °C on average in Guinea-Bissau [59], prior to airplane shipment to laboratory facilities in Germany for DNA extraction and 16S rRNA sequencing. Sample storage time was calculated from day of inclusion to the day of DNA extraction. All samples were stored between 209 and 993 days at room temperature prior to DNA extraction.
The storage on lter paper was chosen due to lack of freezing capacity and further lack of a possibility to transport samples from Guinea-Bissau to central laboratory facilities at stable and constant freezing temperatures. We have recently demonstrated that this particular storage method is applicable in microbiota research, as the fecal microbiota from samples stored on FOBT lter papers at room temperature for up to ve months is comparable to that of a sample frozen and kept at -80°C immediately after collection, with regards to diversity and cumulative abundances [28].
A total of 1,274 fecal samples were collected for microscopic investigation, all from participants with complete questionnaire data. Of these, samples from 1,264 participants were applied to lter paper and underwent DNA extraction as described below. A ow diagram of the study is provided in Fig. 1.
The actual lter from the FOBT cards was cut free from the card using scissors and handled using tweezers. Instruments were cleansed thoroughly between each sample using absolute ethanol (EMSURE®) to avoid cross-contamination between samples. Bacterial DNA ampli cation and pooling The two hypervariable regions V1 and V2 of the 16S rRNA gene were ampli ed using the forward 27F primer and reverse 338R primers and dual MID indexing, as described by Kozich et al. [60]. Bacterial DNA was dually barcoded by unique forward and reverse primers, as described by Caporaso et al. [61], enabling subsequent multiplexing of the PCR product. PCR products were evaluated by gel analysis and normalized using the SequalPrep Normalization plate Kit OTU tables were generated in UPARSE [65], implemented in VSEARCH. After removal of replicates and singletons, reads were clustered based on 97% similarity. Chimeras were once again ltered using VSEARCH in de-novo mode. To generate OTU abundance tables, all reads per sample were mapped to OTU tables using VSEARCH. Using the SINTAX classi er [66] at lowest possible level with minimum 80% bootstrap con dence, one representative sequence for each OTU was annotated. OTUs with identical annotations were grouped into taxonomic bins. Samples from individuals with self-reported antibiotic usage 3 months prior to inclusion were not included in the analysis (n = 31).

Statistical analysis
Between-group differences in baseline characteristics and parasite prevalence were calculated using Wilcoxon rank-sum test, Fisher's exact test and Kruskal-Wallis rank test in STATA 15.1 (StataCorp, College Station, TX, USA). P-values < 0.05 were considered signi cant in these analyses.
Statistical analysis of the microbiota data was performed using the R programming environment v3.2 [67]. All adjusted p-values were obtained using the P.adjust function in R package stats and the Benjamini-Hochberg method (method="BH"). First, data was ltered by excluding samples with less than 10,000 reads and samples possibly affected by overgrowth of facultative anaerobic taxa, which were predominantly found in the Proteobacteria and Firmicutes phyla (the latter in the branch of Streptococci). These samples were identi ed if they fell above the third quartile plus three times inter quantile range (IQR) of phylum abundance (n = 18). Microbial count data on the remaining 1,204 samples was transformed to adjust for deviating sequencing depth by dividing the counts by sample sum and multiplied by one hundred to obtain relative abundances between zero and 100.
As the life cycle, infection route and severity of infection vary considerably between different species of intestinal parasites, we chose to analyze different aspects of the fecal microbiota separately. First, we performed analyses for any parasite, any helminth and any protozoa, and subsequently performed analysis for the most abundant individual pathogenic parasite species (Ancylostoma duodenale, Hymenolepis nana, Entamoeba coli, Entamoeba histolytica/dispar, Giardia lamblia and Endolimax nana). For each analysis, the infected group was compared with the subset of individuals with no detectable parasitic infection (n = 596).
All analyses were adjusted for storage time at room temperature. Alterations associated with helminth infections were adjusted for co-infection with protozoa, and vice versa. To further control for potential confounding factors, all analyses were adjusted for age, usage of vitamin A, toilet source and tropical season of sample collection. As described above, participants were enrolled in the study on a two-cohort basis. While there was no overall difference in infection load between the two cohorts, prevalence of some species differed signi cantly, and we thus adjusted for cohort status in the joint analysis and performed a separate supportive analysis within each group of the two cohorts. Results from the joined analysis is provided in the main text while Supplemental table S1 show results from all three analyses.

Association with alpha-and beta-diversity
Associations between alpha-diversity and each of the nine infection states (overall parasite positive; overall helminth positive; overall protozoa positive; Ancylostoma duodenale positive; Hymenolepis nana positive, Entamoeba coli positive, Entamoeba histolytica/dispar positive, Giardia lamblia positive, Endolimax nana positive) was evaluated using a robust regression (lmRob function in R package robust [68]) and the covariates given above. The alpha diversity measures considered were Shannon entropy (diversity function with index= "shannon" in R package vegan [69]), the measure of species richness ACE (estimateR function using result row four in R package vegan), and phylodiversity as a measure of total unique phylogenetic branch length (calculated using mothur's phylo.diversity function [70] with the phylogenetic tree built using FastTree with --nt and --gtr and the 16S OTU table as input). Evaluation of the association between the parasite infection status and microbial community structure was performed using adonis function in R package vegan [69] with Bray-Curtis dissimilarity and 9999 permutations (remaining settings as default). Furthermore, association with infection load was evaluated with infection load ranging from 0 to 4 by the number of identi ed species within each sample. Covariates were as listed above.

Analysis of single taxa
The relative abundance of single bacteria was evaluated from the taxonomic level of phylum to genus, and was ltered to keep most abundant taxa as follows; ltered to require a mean abundance across all samples of at least 0.05, and an abundance of 0.05 in at least one sample. Further, taxa with ≥40% zeroes across samples were removed. After ltering, 43 genera, 23 families, 15 orders, 11 classes and 5 phyla remained. Associations between the selected taxa and parasite infection status were evaluated using a linear regression with square root transformed taxa abundance and the covariates as listed above.