Ocular community state types reveal distinct microbial compositions among microbiomes with implications for trachoma control

Abstract

Trachoma, a chronic ocular disease caused by Chlamydia trachomatis (Ct), is the leading infectious cause of blindness worldwide. Despite WHO’s SAFE (Surgery, Antibiotics, Facial cleanliness, Environmental improvement) strategy, ~100M are at risk of blindness. Using metagenomic shotgun sequencing, we characterized the ocular microbiome of 680 villagers in Amhara Ethiopia, identifying 10 Community State Types (CSTs) associated with different population characteristics. Children with the highest prevalence of inflammatory trachoma and Ct were in CST10, dominated by Haemophilus influenzae and four other Haemophilus spp. Adults with the highest prevalence of scarring trachoma were in CST3 and CST6, dominated by Corynebacterium macginleyi. CST5, dominated by Mesomycoplasma hyorhinis and Staphylococcus aureus, had the lowest prevalence of Ct and trachoma, and was the only CST without zoonotic Chlamydia spp. Both M. hyorhinis, a zoonotic porcine bacterium, and S. aureus are capable of forming biofilms, which may competitively prevent/down-regulate chlamydial infections. Other CSTs were dominated by environmental species like Vibrio. This is the first microbiome study to develop CSTs for trachoma. Pathogenic and potentially protective microbes showed distinct associations with demographic, clinical, and chlamydial characteristics, which will guide the design of microbial therapeutics as alternatives to antibiotics and strategies for WHO’s global elimination of blinding trachoma.

Introduction

Trachoma is the leading cause of infectious blindness in the world with nearly two million people blind and over 100 million at risk of visual impairment or blindness (1). Recurrent conjunctival infection with Chlamydia trachomatis (Ct), an obligate intracellular bacterium, is responsible for chronic disease (2). The socioeconomic consequences of trachoma include billions of dollars lost annually due to decreased global productivity and reduced quality of life for individuals and their families in affected communities (2).

The World Health Organization (WHO) developed the SAFE (Surgery, Antibiotics, Facial cleanliness, Environmental improvement) strategy originally to eradicate blinding trachoma by 2020, but this has now been extended to 2030 (3). SAFE entails surgery to correct trachomatous trichiasis (in-turned eyelashes), antibiotics to treat Ct infection, and facial cleanliness and environmental improvement to prevent infection and reduce Ct transmission (4). To identify which countries require SAFE, the bilateral upper tarsal conjunctivae are assessed for trachomatous inflammation-follicular (TF) using the simplified trachoma grading scale; a TF prevalence of >10% triggers mass drug administration (MDA) with azithromycin (4). Other grades include trachomatous inflammation–intense (TI), trachomatous scarring (TS), trachomatous trichiasis (TT), and corneal opacity (CO). TF and TI are active signs of trachoma, while TS and TT represent chronic disease (5). As of November 2025, trachoma remains a public health concern in at least 30 countries (6). About 55% of the known global trachoma population resides in Ethiopia (7). The Amhara Region of Ethiopia has the highest burden of infection and disease despite SAFE interventions that commenced in 2001 and included all districts by 2010 (8, 9). The persistently high prevalence of Ct infection and trachomatous disease in Ethiopia warrants further research into the drivers of trachoma pathogenesis that may include non-Ct ocular commensals and pathogens, as well as acquired zoonotic microbes that promote inflammation and disease progression.

Little is known about the ocular microbiome—the collection of commensal and pathogenic microorganisms found in the conjunctivae—and its influence on the acquisition or prevention of Ct infection or role in trachoma pathogenesis (10, 11). A few studies have used culture or 16S rRNA sequencing to describe conjunctival microbiota in Africa. One study found that women in Ethiopia were more likely than men to have Streptococcus pneumoniae and Haemophilus influenzae, and individuals with TT had more frequent conjunctival Ct infection than those with TS (12). In the Gambia, children under age 5 with S. pneumoniae and H. influenzae were more likely to have TF (13). Based on 16S rRNA sequencing of conjunctival samples in the Gambia, adults with TS had an increased relative abundance of Corynebacterium and Streptococcus genera compared to those with no scarring (14, 15).

While these studies provide some data on the microbiota associated with trachoma, metagenomic shotgun sequencing (MSS) has not been used to interrogate the ocular microbiome in trachoma-endemic populations. MSS provides species and often strain-level taxonomic data on culturable and unculturable DNA organisms (16), and has been used in dry-eye disease (DED), blepharitis, keratitis, meibomian gland dysfunction, and cataracts in developed countries (17, 18).

In this study, MSS data were used to identify ocular Community State Types (CSTs)—a classification system for microbiomes based on microbial composition. CSTs were first described for the vaginal microbiome, providing insights into the association between anaerobic bacteria and bacterial vaginosis across different ethnicities (19). Similar to CSTs, enterotypes have been developed for the gut microbiome in Crohn’s Disease, inflammatory bowel disease, and obesity (20, 21). We found that ocular CSTs have distinct microbial signatures associated with different population characteristics, which will facilitate the management of trachoma within the SAFE program and the development of novel ocular therapeutics to prevent and treat disease.

Results

Population characteristics.

A total of 1057 participants were enrolled from three village regions in Amhara. The metadata for each participant are shown in Table S1. Table 1 summarizes bivariate associations between Ct infection (see Methods for definition) and age group, sex, trachoma grade, including a lack of trachoma that we termed T0, occupation, village region, and zoonotic Chlamydia spp. The overall prevalence of Ct infection was 22.2% (235/1057). Adults and older adults had a significantly lower odds of Ct infection than children under 10 years of age. Active trachoma was associated with an increased odds of Ct. Participant occupations and daily household duties were grouped into seven categories (Table S2) with nearly half of the population working in agriculture. Those performing compound duties had significantly lower odds of Ct infection, whereas pre-school children and students had significantly higher odds. Zoonotic or non-Ct Chlamydia spp. were found in 7.5% (51/680) of the metagenomes (Table S1; see Chlamydia taxonomy section below) and were not significantly associated with Ct infection.

Table 1.

Participant characteristics and bivariate associations of C. trachomatis infection with age group, sex, trachoma grade, occupation, and village.

		C. trachomatis Results
	Total	Negative	Positive	OR	95% CI	P-value
	(n=1057)	(n=822)	(n=235)
Age (median, IQR)	50 (30–65)	50 (35–65)	47 (25–69)
Age Group (years)
Children (<10)	82 (7.8%)	48 (5.8%)	34 (14.5%)	0	ref	-
Adolescents (10–19)	64 (6.0%)	40 (4.9%)	24 (10.2%)	0.847	0.43–1.66	0.63
Adults (20–49)	326 (30.8%)	263 (32.0%)	63 (26.8%)	0.34	0.2–0.57	4.13 × 10−5
Older Adults (>49)	578 (54.7%)	464 (56.4%)	114 (48.5%)	0.35	0.21–0.56	1.86 × 10−5
Missing	7 (0.7%)	7 (0.8%)	0 (0.0%)
Sex
Male	525 (49.7%)	397 (48.3%)	128 (54.5%)	0	ref	-
Female	530 (50.1%)	423 (51.5%)	107 (45.5%)	0.786	0.59–1.05	0.1
Missing	2 (0.2%)	2 (0.2%)	0 (0.0%)
Trachoma Grade
T0	136 (12.9%)	103 (12.5%)	33 (14.0%)	0	ref	-
TF or TI	75 (7.1%)	41 (5.0%)	34 (14.5%)	2.59	1.42–4.71	1.9 × 10 −3
TS or TT	834 (78.9%)	667 (81.2%)	167 (71.1%)	0.78	0.51–1.2	0.26
Missing	12 (1.1%)	11 (1.3%)	1 (0.4%)
Occupation
Agriculture	509 (48.2%)	392 (47.7%)	117 (49.8%)	0	ref	-
Cooking/Childcare	100 (9.5%)	85 (10.3%)	15 (6.4%)	0.59	0.33–1.06	0.08
Compound Duties	60 (5.7%)	55 (6.7%)	5 (2.1%)	0.31	0.12–0.79	1.3 × 10 −2
Professional/Technical	39 (3.7%)	35 (4.3%)	4 (1.7%)	0.38	0.13–1.1	0.07
Service, Trade & Manual Labor	152 (14.4%)	123 (15.0%)	29 (12.3%)	0.79	0.5–1.24	0.31
Pre-School Children	57 (5.4%)	34 (4.1%)	23 (9.8%)	2.27	1.28–4	4.75 × 10 −3
Student	83 (7.8%)	52 (6.3%)	31 (13.2%)	2.02	1.24–3.3	4.98 × 10 −3
Other	57 (5.4%)	46 (5.6%)	11 (4.7%)	-	-	-
Village
Village Region 1	382 (36.1%)	284 (34.5%)	98 (41.7%)	0	ref	-
Village Region 2	319 (30.2%)	247 (30.0%)	72 (30.6%)	0.85	0.60–1.2	0.34
Village Region 3	356 (33.7%)	291 (35.5%)	65 (27.7%)	0.65	0.45–0.92	1.59 × 10 −2
Zoonotic & other non-Ct Chlamydia spp.A
Absent	629 (92.5%)	409 (91.9%)	220 (93.6%)	0	ref	-
Present	51 (7.5%)	36 (8.1%)	15 (6.4%)	0.77	0.41–1.45	0.42

IQR, Interquartile Range; OR, Odds Ratio; 95% CI, 95% Confidence Interval; T0, no trachoma; ref, reference for calculating P-value and OR; Chi-square tests were used for this analysis; P-values highlighted in bold are statistically significant

^AOnly samples retained for downstream microbiome analyses were screened for non-Ct Chlamydia spp. (see Methods).

A multivariable logistic regression model, adjusted for potential confounders, is shown in Table 2. Age group, trachoma grade, occupation, and village were included as independent variables. Only active trachoma and compound duties remained significant.

Table 2.

Multivariate associations of C. trachomatis infection with age group, sex, and trachoma grade.

Variables	UOR	95% CI	AOR	95% CI	P-value
Age Group (years)
Children (<10)	0	ref	0	ref	-
Adolescents (10–19)	0.847	0.43–1.66	0.851	0.32–2.24	0.74
Adults (20–49)	0.338	0.2–0.57	0.357	0.11–1.21	0.10
Older Adults (>49)	0.347	0.21–0.56	0.36	0.10–1.24	0.11
Trachoma Grade
T0	0	ref	0	ref	-
TF or TI	2.588	1.42–4.71	2.51	1.24–5.1	1.08 × 10 −2
TS or TT	0.781	0.51–1.2	1.281	0.72–2.27	0.40
Occupation
Agriculture	0	ref	0	ref	-
Cooking/Childcare	0.591	0.33–1.06	0.554	0.31–1.01	5.2 × 10−2
Compound Duties	0.305	0.12–0.79	0.303	0.12–0.78	1.28 × 10 −2
Professional/Technical	0.383	0.13–1.1	0.4	0.14–1.17	0.09
Service, Trade & Manual Labor	0.79	0.5–1.24	0.752	0.47–1.20	0.23
Pre-School Children	2.27	1.28–4	0.486	0.12–1.89	0.30
Student	2.02	1.24–3.3	0.854	0.33–2.23	0.75
Other	-	-	-	-	-
Village
Village Region 1	0	ref	0	ref	-
Village Region 2	0.845	0.60–1.2	0.92	0.64–1.33	0.66
Village Region 3	0.647	0.45–0.92	0.701	0.48–1.02	0.06

UOR, Unadjusted Odds Ratio; AOR, Adjusted Odds Ratio; 95% CI, 95% Confidence Interval; T0, no trachoma; ref, reference for calculating P-value and OR; Chi-square tests were used for this analysis; P-values highlighted in bold are statistically significant

Figure 1 shows the association between age, trachoma grade, and Ct infection in male and female study participants. Active trachoma showed a bimodal distribution with a dominant peak in young children and a smaller peak in adults, while chronic trachoma showed a broad distribution, skewed towards older ages. T0 was concentrated in late adolescence and early adulthood. In males and females, Ct infection was distributed across all ages with peaks in childhood (both sexes), middle adulthood (females only), and late adulthood (males only). Among Ct-positive individuals, active trachoma was more prevalent in male than female children, whereas chronic trachoma peaked in women in middle adulthood and in men in late adulthood.

Figure 1.

Distribution of trachomatous disease and Ct infection by age among (A) males, (B) females, and (C) all participants.

Metagenomic shotgun sequencing results and Chlamydia taxonomy.

Of the 684 MSSs, ~63B raw reads were generated, ranging from 6.3M to 348.7M per sample (Table S3) with a mean of 98% of reads derived from host DNA. Kraken2 identified 13,687 unique species, of which 3,235 were flagged as potential contaminants (see Methods) and removed (Table S4). Samples with at least 1,000 classified reads were retained (680/684), and species with fewer than two reads in less than 10% of samples were removed, yielding 551 unique species (Table S4).

Ct reads were initially identified in 247 of the 680 MSSs using Kraken2 or CZ ID. However, reads assigned to Ct did not always align to Ct genomic or plasmid sequences using Megablast. Only 183 (74.1%) of 247 MSSs had Ct-specific genomic or plasmid reads. Of these, 59 (32.2%) were also positive by qPCR. The latter can be explained by the fact that the qPCR assays target the ompA, omcB and/or plasmid pgp3 genes, and not any other Ct genes or genomic regions. Ct infection was therefore defined as ≥1 genome copy number by qPCR (see Methods), and/or ≥1 MSS reads with 99–100% match to the Ct genome or plasmid.

Based on prior trachoma studies in Guinea and Nepal that found zoonotic or non-Ct Chlamydia species were associated with inflammation (22–24), we screened each metagenome for these species and identified 51 samples (7.5%) with at least one read mapping to these species. (Table S1). These were of avian, mammalian and reptilian origin, but did not significantly associate with Ct infection, a particular trachoma grade or any other participant metadata (Table S5).

Trachomatous disease is associated with significantly reduced species evenness, especially among older adults.

Figure 2 shows species alpha diversity by Ct infection status and trachoma grade. Ct infection was associated with significantly higher species richness (P=0.00042) with no significant change in the Shannon Index. (Fig. 2A–B). Active trachoma microbiomes had a significantly lower Shannon Index (P=0.041) but no difference in richness compared to T0 (Fig. 2C–D). Chronic trachoma microbiomes were similarly associated with lower Shannon Indices (P<0.0001) without significant changes in richness compared to T0 (Fig. 2C–D). Species richness was significantly higher in those with chronic versus active disease (P=0.0081), while the Shannon Index was lower (P=0.047) (Fig. 2C–D). Older adults had significantly lower Shannon Indices than children (P<0.0001), adolescents (P=0.00038), and adults (P<0.0001) with no significant changes in richness between groups. There were no differences by sex.

Figure 2. Alpha diversity of ocular microbiomes by Ct infection, trachoma grade, age, and sex.

Alpha diversity was measured by species richness and the Shannon Index. Box plots display distributions, and P-values were computed using the Wilcoxon rank-sum test and adjusted for multiple comparisons using the Benjamini-Hochberg (B-H) method. (A) Species richness and Shannon Index for individuals with and without Ct infection. (B) Species richness and Shannon Index for trachoma grade. (C) Species richness and Shannon Index by age group. (D) Species richness and Shannon Index by sex.

Beta diversity is distinct across trachoma grades and age groups.

Significant differences were found between Ct and non-Ct microbial communities using the PERMANOVA test, but not ANOSIM (Fig. 3A); this likely reflects marginal shifts in group centroids despite considerable sample overlap. However, microbiome composition differed significantly by trachoma grade for both tests, although total variation (R²) from PERMANOVA and effect sizes (R) from ANOSIM were small (Fig. 3B). Beta diversity analyses were extended to include age group and sex to help explain the differences in composition. PERMANOVA showed significant results for both metrics across age groups and sex, while ANOSIM was significant only for age (Fig. 3C, D). However, the R² values across all test results were similarly low, suggesting that other population attributes not captured by this study may explain differences between ocular microbiomes.

Figure 3. Beta diversity of ocular microbiomes by Ct infection, trachoma grade, age, and sex.

Beta diversity was assessed using Bray-Curtis dissimilarity and Jaccard Indices and visualized using principal coordinate analysis (PCoA). Group differences were tested using permutational analysis of variance (PERMANOVA) and analysis of similarities (ANOSIM). (A) Bray-Curtis and Jaccard indices between Ct-positive and Ct-negative microbiomes. (B) Bray-Curtis and Jaccard indices between T0, TF/TI, and TS/TT. (C) Bray-Curtis and Jaccard indices between age groups. (D) Bray-Curtis and Jaccard indices for sex.

Haemophilus influenzae predominates in active trachoma, Corynebacterium macginleyi in chronic trachoma, and Mesomycoplasma hyorhinis in healthy microbiomes.

We identified differentially abundant and prevalent species using MaAsLin 3 (Fig. 4B–E). A total of 206 significant associations were found based on individual or joint q-values 2264z0.05 (Table S6). When comparing Ct-positive with Ct-negative microbiomes, Ct was the most abundant and prevalent species (Fig. 4B). Most species significantly associated with Ct infection were identified by presence/absence, and only seven differed by abundance (Fig. 4B). Mesomycoplasma hyorhinis was the most enriched species in Ct-negative microbiomes (Fig. 4B). Compared to T0, H. haemolyticus, H. influenzae, and H. parainfluenzae were significantly more abundant in active trachoma, while C. macginleyi and C. accolens were most abundant in chronic trachoma (Fig. 4C; Fig. S1). Comparing TF/TI with TS/TT, H. haemolyticus, H. influenzae, and H. parainfluenzae were significantly more abundant in active trachoma, while C. macginleyi, C. accolens, and S. epidermidis were more abundant in chronic trachoma (Fig. S1).

Figure 4. Ocular Microbiome Differential Species Abundance and Prevalence.

MaAsLin 3 was used to identify species associated with participant characteristics. Species were ordered based on abundance β coefficients and have either individual or joint (a combination of B-H-corrected abundance and prevalence P-values) q-values ≤0.05. For some species, prevalence probabilities could not be computed, so only abundance results are shown. Significant species were identified after controlling for covariates based on (A) Ct infection (reference: Ct-negative); (B) active and chronic trachoma (reference: T0); (C) chronic trachoma (reference TF/TI); (D) sex (reference: females); and (E) age group (reference: children).

Sex and age are associated with distinct microbial abundance profiles.

Cutibacterium granulosum, Corynebacterium singulare, and Cutibacterium avidum were more abundant and prevalent in males compared to females, where Lactiplantibacillus plantarum and S. oralis dominated (Fig. 4D). H. influenzae abundance was significantly lower in all age groups compared to children and declined with age (Fig. 4E; Fig. S1). A similar decline in the abundance of H. parahaemolyticus, H. haemolyticus, H. aegyptius, and Kocuria sp. KD4 was observed in adults and older adults. Only the prevalence of Moraxella bovis and abundance of Streptococcus pneumoniae were significantly lower in adults compared to children (Fig. 4E; Fig. S1). Conversely, S. epidermidis was more abundant in adults and older adults than in children (Fig. 4E; Fig. S1). Five additional Corynebacterium species were more abundant in older adults, but only C. kefirresidentii was more prevalent.

Ten distinct ocular Community State Types (CSTs) segregate villagers microbial composition and population characteristics.

Dirichlet Multinomial Mixtures (DMM) partitioned samples into 10 CSTs that recapitulate the abundance and compositional patterns described above, offering a model-based framework for characterizing ocular microbiome variation in Amhara (Fig. S2). The top 10 species for each CST and their associations with microbial diversity, Ct status, trachoma grade, and age are shown in Figure 5A. C. macginleyi was a top species in six of the 10 CSTs with CST3 and CST6 having the highest contributions or values, which reflect species relative abundance and cluster homogeneity (see Methods). CST3, also dominated by S. epidermidis, had the oldest mean age and a 95% prevalence of chronic trachoma. CST6, which had the highest C. macginleyi contribution, the lowest mean Shannon index, and the highest chronic trachoma prevalence (96%), also included five additional Corynebacterium species.

Figure 5. Characteristics of the Ocular CSTs.

(A) Ten CSTs were generated using DMM, and the top 10 species-level contributions or values (see Methods) for each are shown. CST headers in bold highlight the represented species and the villager demographic, clinical, and chlamydial infection characteristics. (B) The microbial composition of each sample was analyzed for each CST using the relative abundance of the top 25 species in the population. (C) Each CST was analyzed by Shannon Index and age, and the Kruskal-Wallis chi-square test identified significant differences between CSTs. (D) The stacked bar plots show the proportions of individuals within each CST stratified by trachoma grade, Ct infection status, sex, and village region, highlighting characteristics associated with CST membership.

Four CSTs had unique dominant species: CST5 had M. hyorhinis and the highest proportion of Ct-negative (92%) and T0 (30%) microbiomes, and was the only CST without any zoonotic Chlamydia spp. (Table S6); CST8 had three Vibrio spp.; CST9 had five Moraxella spp. and a high prevalence (91%) of chronic trachoma; and CST10 had five Haemophilus spp., the highest prevalence of TF/TI (80%) and Ct infection (45%), and the lowest mean age. The remaining four CSTs had lower but relatively equal species contributions.

The distribution of the dominant species in each CST are shown in Figure 5B. and significant differences in Shannon diversity and age by CST for trachoma grade, Ct infection status, sex, and village membership in Figures 5C–D. To evaluate significant associations between participant metadata and CSTs, we fitted Firth-penalized one-vs-all logistic regression models for each CST, adjusting for Ct infection, trachoma grade, sex, age group, occupation, and village (Table S7). CST5 had a significantly lower odds of Ct infection, active trachoma, and chronic trachoma. By contrast, CST7 had an increased odds of Ct infection and decreased odds of having individuals from Village 2 and Village 3. CST9 had a much higher odds for females, and CST10 had a significantly higher odds of active trachoma.

The 10 ocular CSTs highlight distinct microbial patterns in Amhara.

CST1 and CST2 had the highest number of samples (weight ${\overline{\pi }}_1=0.17$; ${\overline{\pi }}_2=0.15$), and the highest variability in microbial composition (precision ${\overline{\theta }}_1=27.7$; ${\overline{\theta }}_2=73.0$) (Table S8). CST5 and 6 had the most homogeneous microbiomes (${\overline{\theta }}_5=414.6$; ${\overline{\theta }}_6=436.2$), each capturing 10% of the population (${\overline{\pi }}_5=0.10$; ${\overline{\pi }}_6=0.10$) (Table S8). Comparing the species composition of each cluster to a reference cluster representing the average ocular microbiome of all participants, the most distinct clusters were CST6 (144%), CST5 (132%), CST9 (121%), and CST10 (117%) (see Methods; Table S8). Nine species accounted for 40% of the cumulative differences between CSTs and the reference: C. macginleyi, H. influenzae, Moraxella sp. K1664, M. bovis, M. hyorhinis, V. kanaloae, S. aureus, H. haemolyticus, and E. cecorum (Table S9).

Figure 6 shows the PCoA results based on Bray-Curtis indices with 3-D representations highlighting CST clustering (Fig. 6B–F; see Methods). The CSTs were significantly different with CST5, 6, and 9 showing the best separation.

Figure 6. PCoA Representation of CSTs.

(A) Two-dimensional PCoA plot of Bray-Curtis dissimilarity indices for each CST. Significant differences in microbiome composition were observed using PERMANOVA (F=51.789, P<0.001, R2=0.4103) and ANOSIM (R=0.468, P<0.001). (B) Three-dimensional visualization of CST1 (light blue), CST2 (dark blue) and CST4 (dark green); (C) CST3 (light green) and CST6 (red); (D) CST5 (salmon) and CST7 (light orange); (E) CST9 (light purple) and CST10 (dark purple); and (F) CST8 (dark orange).

Discussion

The global commitment to eliminating blinding trachoma by 2030 requires a better understanding of disease mechanisms and drivers of persistence in hyperendemic countries. To that end, we applied MSS to characterize ocular microbiomes among individuals in Amhara, which accounts for nearly half of the known global trachoma cases despite decades of SAFE interventions. Our study is the first to apply MSS to define the ocular microbiome in trachoma. We detected commensal and pathogenic species often missed by culture or 16S rRNA approaches, which is an important advantage for low-biomass conjunctival samples (25, 26). Our findings show that ocular CSTs translate microbiome patterns into actionable insights for trachoma control.

We found that children and those with active trachoma had a higher odds of Ct infection, consistent with prior studies that showed that children are the primary reservoir of Ct (2, 27, 28). Male children had a higher proportion of Ct and TF/TI than female children, likely from greater organism transfer during close-contact play. Adult females had higher infection and TF/TI prevalence than males and developed scarring nearly two decades earlier (Fig. 2). In sub-Saharan Africa, women are the primary caretakers of children, which increases their exposure to Ct and makes them four times more likely than men to develop TS (29, 30). Chronic trachoma, occurring from inflammation and progressive scarring (31), was prevalent in older adults; men developed it later than females, possibly due to declining mobility and/or retirement when household contact with infected children would increase.

These epidemiologic patterns are reflected in the diversity and composition of the ocular microbiome. Although Ct-positive microbiomes had higher species richness than Ct-negative microbiomes, Shannon diversity did not differ, suggesting that the additional taxa were rare and did not alter community evenness. Active trachoma metagenomes had lower species richness and Shannon diversity compared to T0, consistent with a Gambian 16S rRNA study that reported reduced richness but no change in Shannon diversity (15). Chronic trachoma was associated with significantly lower Shannon indices, especially among older adults, without a change in richness, suggesting expansion of a few dominant taxa. Reduced microbial diversity is common in TS (14, 15), and a hallmark of ocular surface dysbiosis in DED, Stevens-Johnson Syndrome, and meibomian gland dysfunction (32). Children had the highest Shannon diversity, likely reflecting greater physical contact, exposure to environmental microbes, and immature hygiene practices.

The high sensitivity of MSS may amplify contaminants in microbiomes, but there is little consensus on how best to address this issue (25, 33–35). Some studies purport to identify ‘known’ contaminants and recommend removing them prior to downstream analyses (36–38). However, contaminant lists are reported at the genus level, which risks excluding legitimate species in the microbiome (39, 40). For example, Corynebacterium has appeared in negative sequencing controls (36). Yet several Corynebacterium spp. (i.e., C. macginleyi, C. accolens, C. pseudodiptheriticum, C. propinquum, C. mastidis, C. striatum, and C. xerosis) have frequently been isolated from the conjunctiva and implicated in infection and disease (41, 42). In our study, we successfully used the program decontam to identify contaminants, but retained species that could reasonably comprise Amhara microbiomes, noting that decontam may flag taxa present in both samples and controls depending on the prevalence threshold used (39).

We also considered that some species classified as contaminants could reflect environmental exposures since the ocular surface is exposed to the outside world (43). Nearly half of our participants worked in agriculture and animal husbandry, and one abundant species in our dataset, Bradyrhizobium sp. SK17, is a common African soil symbiont that nodulates legumes and fixes atmospheric nitrogen (44). Legumes such as faba beans, haricot beans, field peas, and chickpeas are widely cultivated by households in Ethiopia for food security, soil fertility, and export (45–48). Although contamination cannot be entirely excluded, the detection of Bradyrhizobium sp. SK17 is consistent with agricultural exposures and may represent a transient or stable member of the ocular microbiome in Amhara. A 2014 study in The Gambia detected the Bradyrhizobium genus in villagers’ eyes and attributed its presence to poor housing and sanitation (14). Another study comparing 16S rRNA sequences from female hands in the U.S. and Dar es Salaam, a non-trachoma city, found that Tanzanian participants harbored more soil- and water-associated family-level bacteria like Bradyrhizobiaceae, and Rhizobiaceae, microbes the authors linked to contact with soil and untreated surface water (49). In Amhara, limited access to sanitation, safe drinking water, and adequate hygiene (50) likely increases environmental and animal exposures, reflected in the detection of 51 zoonotic and other non-Ct Chlamydia spp. of avian, mammalian, and reptilian origin among the microbiomes. These findings highlight the need to expand studies of low-biomass microbiomes to capture geographic population differences driven by environmental and societal factors.

We defined 10 ocular CSTs that capture distinct microbial profiles in Amhara, providing a framework to study the microbiome’s role in Ct infection and trachoma pathogenesis. A recent 16S rRNA study of healthy Europeans identified nine ocular CSTs dominated by Staphylococcus, Bacillus, and Corynebacterium genera (51). Many participants in our study harbored C. macginleyi. While dominant in six CSTs, this species was only markedly abundant in two CSTs, and both had the highest percentage (i.e., ≥95%) of individuals with TS/TT. Several Corynebacterium species are constituents of the ocular microbiome and may synergize with other commensals to protect the eye from virulent microbes (52). However, these species have also been found in conjunctivitis, keratitis, blepharitis, and endophthalmitis (53–55).

Microbiome shifts are common in DED, a condition associated with aging and characterized by tear film instability due to lacrimal or meibomian gland dysfunction (MGD) (56). MGD disrupts the outermost lipid layer of the tear film and can favor the overgrowth of lipophilic commensals (57). One 16S rDNA study of DED found that a single species (C. accolens-macginleyi or C. kroppenstedtii) accounted for >75% of sequencing reads in 12% of patients (53). Studies of MGD similarly found C. macginleyi-dominant microbiomes, highlighting its potential as a biomarker of disease severity (58). In our study, C. macginleyi had the highest abundance in CST3 and 6, which had the oldest participants and highest chronic trachoma prevalence. TS causes conjunctival goblet cell loss, reduced tear production, and meibomian gland atrophy that promotes DED (59, 60). Further, expansion of lipophilic microbes like C. macginleyi can exacerbate inflammation via increased lipid metabolism, which may explain its differential abundance in chronic trachoma (59, 60). CST6 contains five additional Corynebacterium species, three of which (C. accolens, C. tuberculostearicum, C. kefirresidentii) are also lipophilic. Though C. accolens is associated with the nares (55), it has been detected in DED, likely due to proximity of the nares to the eyes. Both C. accolens and C. macginleyi produce lipases that suppress other bacteria and facilitate their own growth (53), supporting a potential pathogenic role for C. macginleyi in TS and TT.

Several common ocular commensals among our CSTs, such as Kocuria palustris, Kocuria rosea, Dolosigranulum pigrum, and S. epidermidis are also opportunistic pathogens (61, 62). S. epidermidis is predominant in DED (58, 63), and its cholesterol and fatty wax esterases contribute to tear-film lipid degradation (64). S. epidermidis was among the top ten species in eight CSTs and differentially abundant in adults, older adults, and chronic trachoma, suggesting a possible role, alongside C. macginleyi, in disease pathogenesis. Both bacteria form biofilms that may promote DED via quorum sensing, virulence factor expression, and inflammation (56), mechanisms that could worsen trachomatous disease. Conversely, S. epidermidis has probiotic potential—its biofilms and antimicrobial peptides can inhibit colonization by opportunistic pathogens like S. aureus through competitive exclusion (65). However, early S. aureus colonization can overtake a niche and displace S. epidermidis and other protective commensals, as has been shown in the skin (66). Consistent with this, S. aureus occurred only in CST5 at a high relative abundance without S. epidermidis. The median age in CST5 was 28 years (the second lowest), which may reflect age-related microbiome shifts after repeated azithromycin MDA in childhood with accumulation of antimicrobial resistance genes (ARGs) among non-Ct bacteria. Antibiotic exposure can select for pathogens and ARGs, especially in S. aureus (67). A study of azithromycin MDA to reduce childhood mortality in Burkina Faso and Mali reported an increase in S. aureus nasal carriage from 13.8% to 20.1% after one year without a significant change in azithromycin resistance among isolates (68). In The Gambia, nasal azithromycin- and macrolide-resistant S. aureus increased after three annual rounds of MDA (69). Although we did not assess ARGs in this study, elevated S. aureus abundance warrants ongoing surveillance for potential collateral effects of continued MDA in Amhara.

The most abundant bacterium in CST5 was M. hyorhinis, a porcine respiratory commensal that can cause systemic infection (70). Porcine production is an emerging export sector in Amhara, typically carried out by adults aged 21–39 years (71). An ocular microbiome study in China, which accounts for roughly half of global pig production and consumption, found M. hyorhinis enriched in young adults (mean: 27.9 years) versus older adults (mean: 67.1 years) regardless of sex (72, 73). This suggests that porcine husbandry may be a zoonotic source of M. hyorhinis. CST5 also had the highest proportion of Ct-negative and T0 individuals and was the only CST without zoonotic Chlamydia spp. (Figure 5), suggesting some competitive or inhibitory interaction among CST5 species. Like S. aureus, M. hyorhinis may serve as a founder organism in biofilm formation (74, 75), thereby preventing Ct and/or other Chlamydia spp. infections. In vitro studies have shown that various Mycoplasma species, including M. hyorhinis, can inhibit the infectivity and growth of Ct and C. pneumoniae (76). But urogenital Ct strains frequently co-exist with Mycoplasma genitalium in the vagina (77). Interactions among ocular Ct, M. hyorhinis, S. aureus, and other Chlamydia spp. require further investigation.

CST8 was characterized by Vibrio spp., two of which are opportunistic aquacultural pathogens: V. kanaloae and V. diabolicus (78). Vibrio spp. have been isolated from Lake Tana, a major fishing site in Amhara, despite contamination from fertilizers, municipal sewage, and pit latrines (79). V. plantisponsor is associated with wild rice (80). Although these are not established human pathogens, CST8 underscores the need to strengthen the “F” and “E” in SAFE. We also identified other zoonotic and environmental microbes among the CSTs: E. cecorum, an emerging poultry pathogen (81); Dietzia kunjamensis, from desert soil (82); Ornithinimicrobium flavum, a plant microbe (83); and Kytococcus sedentarius, a marine microbe that can cause opportunistic human infections (84).

Villagers in CST10 were the youngest and had the highest TF/TI prevalence and second-highest Ct prevalence. CST10, dominated by H. influenzae, a leading cause of childhood conjunctivitis, also had a high abundance of four other Hemophilus spp. H. influenzae along with Streptococcus species have frequently been associated with active trachoma (13). S. pneumoniae was significantly more abundant in children, while S. oralis was enriched in females. These species colonize the nasopharynx and can be transmitted to the eye via fomites or autoinoculation.

CST7 had the highest prevalence of Ct-positive microbiomes, and in addition to C. macginleyi, was dominated by B. wexlerae, a gut microbe associated with DED (63). B. wexlerae produces acetate, which at low levels suppresses NF-κβ activity and is anti-inflammatory, but in excess can be proinflammatory (85). Excess acetate has been documented in S. epidermidis, which was also present in CST7, and C. acnes infections of the skin (85, 86). Since acetate does not require receptor uptake (85), it spreads systemically and regionally, which may lead to higher concentrations in the eye from adjacent skin or nasopharyngeal infections with S. epidermidis. The presence of both B. wexlerae and S. epidermidis in CST7 may therefore enhance inflammation and damage mucosal barriers that could predispose to Ct infection, persistence and/or spread within the tissue. Host-pathogen interactions among B. wexlerae, S. epidermidis, and Ct in the ocular microbiome require additional study.

Five Moraxella spp. dominated CST9 and are known to elicit proinflammatory responses that contribute to trachomatous disease (87, 88). These species, notably M. catarrhalis, are common in the conjunctivae and nasopharynx of children and can cause conjunctivitis (89). CST9 comprised mostly older adult females with chronic trachoma, which is supported by prior studies in Ethiopia where the Moraxella genus was prevalent in individuals with TT (12, 90). Moraxella spp. were also common among adult females in our study, suggesting that repeated exposure to this organism during childcare increases infection risk among female caregivers as with Ct infection.

We present the most comprehensive assessment of the ocular microbiome in trachoma to date. The 10 CSTs provide a species-level framework for risk stratification within trachoma-endemic populations like Amhara based on microbiome composition. The CSTs comprise pathogenic and potentially protective microbes and identify meaningful connections with participant characteristics, enhancing our understanding of trachomatous disease. However, these species require further investigation to identify mechanisms that could either drive or prevent Ct infection and trachoma pathogenesis as well as the role that zoonotic Chlamydia may play in both. Furthermore, the CSTs highlight discrete microbiome shifts that can be quantified and linked to Ct infection, trachoma progression, and disease prediction in future studies. Our findings will inform the design of novel microbial therapeutics as alternatives to antibiotics, thereby advancing the WHO SAFE strategy towards the global elimination of blinding trachoma.

Methods

Study population, trachoma grading, and sampling.

The study took place at least one year following the last azithromycin MDA program and is well described in the parent study (91). Briefly, trained ophthalmic personnel screened participants for trachoma by examining the upper tarsal conjunctivae using a 2.5x ocular loop and the WHO simplified trachoma grading scale: TF, five or more follicles with a diameter of at least 0.5-mm; TI, pronounced inflammation that obscures more than half of the underlying vessels; TS, white visible scars; and TT, one or more eyelashes touching the eyeball or evidence of recent eyelash epilation (89). The term T0 was used to denote the absence of clinical signs of trachoma. All trachoma grades were confirmed with high-resolution photographs taken prior to swabbing the upper tarsus of each eye as previously described (91). Swabs were stored in sterile M4 buffer (Remel, Lenexa, Kansas) at −80°C until use. Demographic data included age, sex, employment status, occupation, daily household duties, and village. Individuals were grouped by age: children (<10 years), adolescents (10–19 years), adults (20–49 years), and older adults (≥50 years). Figure 1 details the pipeline used to characterize the ocular microbiomes of study participants.

Sex as a biological variable.

Both male and female villagers from Amhara, Ethiopia, were included in this study. Sex was modeled as a covariate, and sex-dependent differences were investigated in all analyses.

Sample processing, metagenomic shotgun sequencing, and taxonomic profiling.

Conjunctival samples were treated with a lysozyme cocktail followed by DNA extraction as previously described (92, 93). Libraries were prepared from the extracted DNA using Illumina Nextera XT kits and sequenced as 150-bp paired-end reads on an Illumina NovaSeq platform. Raw reads were preprocessed using fastp (v0.23.4) (94), which performs ultrafast quality control, adapter trimming, and base correction on FASTQ data. Human reads were depleted using Scrubby as previously described (95). Taxonomic profiling was performed with Kraken2 using the plusPF database, a combination of RefSeq protozoa and fungi genomes and the Kraken2 Standard database (RefSeq archaea, bacteria, viruses, plasmid, human, and UniVec_Core complete genomes) (96). Kraken2 uses exact k-mer (k=35) matches to classify reads with high sensitivity and speed compared to conventional tools (97, 98), but may generate spurious matches in low-biomass environments such as the conjunctiva where the signal-to-noise ratio is low (39). The accuracy of read classification is influenced by low-complexity and homologous nucleotide sequences, which can result in multi- and cross-mapping. Multi-mapping, common between closely related species, occurs when reads fail to uniquely align to a single species (99). Cross-mapping, often between distantly related species, occurs when reads align to off-target genomes (99). These issues are further complicated by reference databases that may over- or under-represent certain taxa, contain human-contaminated or incomplete genomes, or have inaccurate or sparse taxonomic labels (96, 100).

To help mitigate these concerns, we applied a confidence score threshold in Kraken2 to specify the minimum proportion of k-mers in a read that must map to a taxon for assignment at that taxonomic rank. Confidence scores can help reduce false species identification in metagenomes, but higher thresholds can increase the number of unclassified reads (101). To identify an appropriate threshold, we calculated label scores—the ratio of k-mers mapping to a species to all unambiguous k-mers within a read (102)—for each read classified at the species-level using Kraken2’s default confidence setting of 0. These scores were averaged for all reads assigned to each species across all samples, producing a distribution of confidence scores (Fig. S3). A confidence score of 0.1 was found to improve the precision of read classification, and samples with at least 1,000 classified reads were retained for further analysis.

Identification of contaminants and quality control.

True microbial signals in microbiomes may also be obscured by DNA contamination from laboratory kits, reagents, and/or the environment (36, 103). Decontam (v1.26.0) in R was used to statistically identify and remove contaminating species using the ‘isContaminant’ function, which combines frequency and prevalence modes. The frequency method (threshold=0.1) identifies species inversely correlated with microbial DNA concentration, while the prevalence method (threshold=0.2) detects species that are more abundant in negative controls compared to samples (104). In this study, we processed nuclease-free water, M4 buffer, control swab elutions, and control swabs in M4 buffer as negative controls alongside clinical samples to detect microbes that may have been introduced during sample collection and handling, DNA extraction, and/or sequencing. While combining frequency and prevalence methods is optimal for identifying contaminants in low-biomass samples, the prevalence method may misclassify genuine taxa as contaminants when a species occurs in both clinical samples and controls (39). Additionally, many taxa commonly reported as laboratory and reagent contaminants (36, 105) are found in the environment, but may, in fact, comprise the conjunctival microbiomes of study participants who lead an agrarian lifestyle.

To distinguish true microbial community members from contaminants, we statistically compared species read distributions between clinical samples and controls. Species belonging to genera not previously associated with the eye or unlikely to be found in trachoma-endemic agrarian regions like Ethiopia were excluded. To account for species found in both samples and controls that may plausibly belong to the ocular microbiome, we calculated the ratio of the maximum reads in samples compared to controls for species flagged as contaminants. Species with ratios exceeding a threshold based on the number of reads in controls were considered non-contaminants. For example, Staphylococcus aureus had <100 maximum reads in controls (5x ratio), and Corynebacterium kefirresidentii had 100–1,000 maximum reads in controls (10x ratio). Because both showed substantially higher maximum reads in samples, they were retained for downstream analyses. Finally, we excluded species with fewer than two reads in fewer than 10% of samples to control for potential sequencing artifacts from rare taxa (106). We verified the remaining species using CZ ID (formerly IDSeq) Illumina mNGS Pipeline (v8.3), an open-source platform for pathogen detection in metagenomic data (107). CZ ID complements Kraken2’s k-mer approach for taxonomic profiling by using minimap2 (108) to directly align reads to reference sequences in a compressed version of NCBI’s NT database (Index Date 2024-02-06). CZ ID has built-in tools to evaluate species coverage and integrates with the Basic Local Alignment Search Tool (BLAST), which was used to confirm species identified by Kraken2.

C. trachomatis infection screening and identification of zoonotic and other non-Ct Chlamydia species.

Samples were screened for Ct infection using an in-house quantitative (q)PCR assay targeting the single-copy omcB or ompA gene (109). While qPCR is highly sensitive and specific for detecting Ct, targeting only two genes does not detect the full Ct genome or plasmid, potentially underestimating infection prevalence. Additionally, although MSS has been used to clinically diagnose infectious diseases (110), its effectiveness in low-biomass sites like the conjunctiva may be limited when DNA yield is low (39). After species were identified using Kraken2, each participant’s microbiome was queried for the presence of Ct and non-Ct Chlamydia spp. reads using CZ ID and BLAST as described above to confirm their species assignment.

Ct infection was defined as ≥1 genome copy number by qPCR and/or one or more MSS reads with 99–100% match to the Ct genome or plasmid. Similar criteria were used to define the presence of non-Ct Chlamydia spp.

Analysis of microbial diversity, composition, and differential abundance.

To determine population-level changes in conjunctival microbial diversity, the number of observed species (i.e., richness) and the Shannon Index, which accounts for both species richness and relative abundance, or evenness, in a sample (111), were computed for Ct infection and trachoma grade. We additionally assessed differences in species composition between individuals based on Ct status and trachoma grade using Bray-Curtis (B-C) and Jaccard beta diversity indices. Bray-Curtis quantifies dissimilarity between samples based on species abundance, while Jaccard measures dissimilarity based on species presence or absence (112). Principal Coordinate Analysis (PCoA) was used to visualize group differences.

We identified differentially abundant and prevalent species in ocular microbiomes using Microbiome Multivariable Associations with Linear Models 3 (MaAsLin 3) (113) to test associations with Ct infection while controlling for trachoma grade, age group, sex, and sampling depth. MaAsLin 3 accounts for microbiome sparsity and compositionality, analyzes absolute and relative abundances, and adjusts for covariates. It was implemented using default settings, except for an abundance threshold of 1% and a false discovery rate (FDR) of 0.05 using the B-H correction. We further identified differentially abundant and prevalent species across trachoma grades, sex, and age groups using the same model.

Identification of ocular Community State Types (CSTs).

Dirichlet Multinomial Mixtures (DMM), a probabilistic model for microbiome datasets with varying sampling depth, was used to define ocular CSTs. Briefly, DMM uses unsupervised learning and an evidence-based framework to determine the optimal number of Dirichlet mixture components, or clusters, that fit a dataset (114). DMM was applied at the species level, and the weight ($\overline{\pi }$) and precision ($\overline{\theta }$) of each cluster were analyzed to determine its size and homogeneity, respectively. Similar to Holmes et al., we identified the most distinct CSTs by summing the absolute differences in mean species relative abundance between each CST and a reference cluster (114). The reference cluster represented the average microbiome composition across all participants. The sum was converted to a percentage ranging from 0 to 200, with 0 representing completely identical clusters and 200 completely different clusters. We additionally used this result to quantify each species’ contribution to the calculated differences between CSTs and the reference cluster.

The top 10 species in each cluster were assessed using the Dirichlet parameter ($\overline{\alpha }$), which is the product of that cluster’s precision and taxa relative abundances. This quantity, referred to as “value,” reflects how strongly each species contributes to a given cluster. The species composition of each sample was evaluated using relative abundance, and we compared Shannon Indices, age, trachoma grade, Ct infection, and the presence of non-Ct Chlamydia spp. by CST. We built Firth-penalized logistic regression models for each CST to identify associations between participant characteristics and cluster membership, and visualized differences between CSTs using PCOA.

Statistics.

Statistical tests were performed in R using phyloseq (v1.50.0) (115) and vegan (v2.7.1) (116), and plots were generated in R using ggplot2 (v3.5.2) (117) and ggpubr (v0.6.0) (118). Alpha diversity comparisons and differences between read distributions for samples and controls were identified using pairwise Wilcoxon Rank-sum tests (B-H correction for multiple comparisons). Shannon index and age-group differences across CSTs were assessed with the Kruskal-Wallis test, while beta diversity comparisons used Permutational Analysis of Variance (PERMANOVA) and Analysis of Similarities (ANOSIM). Two-sided P<0.05 was considered significant unless otherwise stated.

Study Approval.

As per the parent study (91), villagers aged one and older from Amhara, Ethiopia, provided informed consent prior to enrollment, with parents consenting for minors, per the approved protocols of the Institutional Review Boards of the University of California San Francisco School of Medicine and the Ethiopian Ministry of Health in accordance with the Declaration of Helsinki. Participants were assigned a unique identification number, and only these data were used for all downstream analyses in the present study.

Figure 7.

Pipeline for ocular microbiome data generation, preprocessing, and analysis. Adapted from Bommana et al. (92).

Funding Statement

This research was funded by the NIH and is governed by the NIH Public Access Policy. NIH has the right to make the research available in PubMed, and all authors adhere to this policy.

• National Institutes of Allergy and Infectious Diseases grant R01AI158527 (to DD and TDR)

• National Institutes of General Medical Sciences Medical Scientist Training Program grant T32GM141323 (to Aimee Kao; AU funded through this mechanism)

Data availability.

FASTQ files for the metagenomic data are available at NCBI-SRA under BioProject PRJNA12345.