Urobiome composition varies with vesicoureteral reflux grade and history of febrile urinary tract infection

Abstract

Vesicoureteral reflux (VUR) is a risk factor for urinary tract infection (UTI). However, data describing the urobiome in children with VUR are limited. Here, we characterized the urobiome in children with and without VUR to better understand its potential role in VUR pathogenesis. Catheterized urine samples were obtained from children who presented for voiding cystourethrogram (VCUG) between 12/1/2020 and 10/3/2023. Children were excluded if they had any surgical procedures of the genitourinary tract in the 30 days prior to VCUG or were on treatment antibiotics at the time of VCUG. 16S rRNA gene sequencing and taxonomical classification identified the urobiome. Statistical models using distance, abundance and distribution-based approaches were used to associate the taxonomic profiles with clinical variables. In children with VUR, there is a positive association between history of febrile UTI and increasing abundance of Anaerococcus and decreasing abundance of Streptococcus, Ezakiella, and Corynebacterium. When examined together with age, sex, antibiotic prophylaxis, and highest grade of VUR, addition of the urobiome improved prediction of history of febrile UTI. Additionally, the urobiome is responsive to VUR grade, but does not predict VUR, and beta-diversity varies with higher grade VUR.

Supplementary Information

The online version contains supplementary material available at10.1038/s41598-026-53345-2.

Introduction

Urinary tract infections (UTIs) are one of the most common bacterial infections in children, occurring in an estimated 8% of symptomatic children¹. Vesicoureteral reflux (VUR), a condition characterized by the back flow of urine from the bladder up the ureters, is a risk factor for UTI, with higher degrees of reflux conferring greater risk². Children with higher grades of VUR are also more likely to have a non-Escherichia coli (E. coli) organism as the causative bacteria for their UTI, which is often associated with antibiotic resistance^3,4. Neither the exact mechanism of the increased risk of UTI in VUR, nor the reason as to why non-E. coli UTIs are more common in higher grades of VUR is known, but they are hypothesized to be related to the retrograde flow of urine from the bladder.

The urobiome refers to the community of bacteria within the bladder, which has been reported in children as young as 2 weeks of age⁵. Emerging data has described changes in the urobiome during acute UTI^6,7, although the exact relationship between the native urobiome and UTI risk is not yet elucidated⁸. There is a paucity of data describing the urobiome in children with different grades of VUR, raising the question of whether the baseline urobiome in children with VUR is different from those without VUR. Further, children with VUR are often managed with antibiotic prophylaxis^9,10. However, the effect of prophylactic antibiotics on the urobiome is also unclear.

The primary objective of this study is to compare the beta-diversity and overall composition of the urobiome between children with no VUR to those with differing grades of VUR. The secondary objectives are to assess the effect of prophylactic antibiotics and history of febrile UTI on both the composition and diversity of the urobiome. We proposed three linear models to examine the relationship between febrile UTI, VUR grade, and the urobiome (Fig. 1 and Methods). We hypothesized that the urobiome would be different based on both VUR status as well as history of febrile UTI and use of antibiotic prophylaxis.

Fig. 1.

Linear models fit integrating covariates, vesicoureteral reflux (VUR), urobiome and febrile urinary tract infection (UTI). The covariates of age, sex, and prophylaxis were controlled for as predictors in all models. Fitting Model 1 (Predisposing Febrile UTI) identifies whether VUR Grade and the components of the urobiome can predict febrile UTI as the response variable (y). Model 2 (Markers of Having Had Febrile UTI) identifies which components of the urobiome (y) can be predicted by VUR Grade and febrile UTI (x). Model 3 (Mediators of VUR Grade) identifies whether components of the urobiome (x) predicts VUR Grade (y). Although significant associations do not prove causations, the relative strengths of associations may support the plausibility of directionality.

Results

We collected 399 urine samples, of which 214 were excluded. A small number of samples (n = 31) were misnumbered and ultimately excluded as we could not verify the sample ID. The remaining 185 were sequenced, with a resulting urobiome identified in 182 (98.4% had a detectable urobiome). Of the children who were ultimately included and had a detectable urobiome, 93 (51.1%) children had no VUR (Fig. 2).

Fig. 2.

Flow diagram of collection sample sizes and subsequent distribution of vesicoureteral reflux (VUR) grade. Low-grade VUR includes grades I-III. High-grade VUR includes grades IV and V. CAKUT congenital anomaly of the kidney and urinary tract.

Our entire cohort had a mean age of 3.9 (standard deviation 3.9) years and was 74.1% female (Table 1). The indication for voiding cystourethrogram (VCUG) was mostly UTI (73.6%) followed by a history of antenatal hydronephrosis (15.4%) and other (11.0%). Most children (68.9%) were on antibiotic prophylaxis at the time of VCUG. Of the 182 children with a detectable urobiome, 89 (48.9%) had VUR. Of those with VUR, 14.6% were grade I, 20.2% were grade II, 32.6% were grade III, 25.8% were grade IV, and 6.7% were grade V. Children with high-grade (grades IV and V) VUR were younger than those without VUR [mean (standard deviation): 2.7 (2.4) vs. 4.4 (4.4) years, p < 0.05]. There was also a lower percentage of females in the high-grade group compared children with either low-grade or no VUR (55.2% vs. 93.3% vs. 67.7%, p < 0.01). While the indication for VCUG was primarily UTI in all groups, the highest proportion of children with UTI as indication for VCUG was in those with low-grade VUR group (88.3% for low-grade VUR versus 65.6% for no VUR and 68.9% for high-grade VUR, p = 0.02). Patients with high-grade VUR had the highest proportion of patients with antenatal hydronephrosis as an indication, although this difference was not significant (24.1% for high-grade VUR vs. 8.3% for low-grade VUR vs. 17.2% in no VUR, p = 0.12). There were also differences in history of febrile UTI and use of antibiotic prophylaxis between groups, with 100% of patients (for whom data was available) with high-grade VUR having history of febrile UTI, and antibiotic prophylaxis being most common in patients with high-grade VUR.

Table 1.

Patient demographics and clinical characteristics.

	All samples (n = 182)	VUR (n = 89)	No VUR (n = 93)	Low grade VUR (n = 60)	High grade VUR (n = 29)	p-value
Mean age in years (SD)	3.9 (3.9)	3.3 (3.3)	4.4 (4.4)	3.8 (3.6)	2.7 (2.4)*	*p < 0.05 vs. no VUR
Female (%)	135 (74.1)	72 (80.1)	63 (67.7)	56 (93.3)	16 (55.2)	< 0.01
Indication for VCUG
UTI (n (%))	134 (73.6)	72 (80.9)	61 (65.6)	52 (88.3)	20 (68.9)	0.02
Antenatal hydronephrosis (n (%))	28 (15.4)	12 (13.5)	16 (17.2)	5 (8.3)	7 (24.1)	0.12
Other (n (%))	20 (11.0)	5 (5.6)	15 (16.1)	3 (5.0)	2 (6.9)	0.19
Bilateral VUR (n (%))	–	48 (53.9)	–	26 (43.3)	22 (75.9)	< 0.01
Highest Grade of VUR	–		–
1 (n (% of all VUR)	–	13 (14.6)	–	13 (21.7)	–	–
2 (n (% of all VUR)	–	18 (20.2)	–	18 (30.0)	–	–
3 (n (% of all VUR)	–	29 (32.6)	–	29 (48.3)	–	–
4 (n (% of all VUR)	–	23 (25.8)	–	–	23 (79.3)	–
5 (n (% of all VUR)	–	6 (6.7)	–	–	6 (20.7)	–
History of febrile UTI (n(%))a	137 (94.5)	75 (87.2)	62 (92.5)	53 (96.4)	22 (100.0)	0.02
Antibiotic prophylaxis (n(%))b	122 (68.9)	68 (76.4)	54 (61.4)	43 (71.7)	25 (86.2)	0.02

(a) This data available for 145 children, (b) This available for 177 children.

VUR grade and the urobiome as predictors of febrile UTI history (Model 1)

The inclusion of the 25 most abundant taxa improved this model’s identification of children with history of febrile UTI over covariates and VUR grade alone (reduced McFadden’s 0.336; full McFadden’s 0.532; difference = 0.196; p-value = 0.08). Examining the coefficients and p-values of the covariates (age, sex, antibiotic prophylaxis, VUR grade) identified only sex (female) as a positive predictor of history of febrile UTI (p < 0.001). Of the included taxa, Ezakiella (p < 0.05), Corynebacterium and Streptococcus (p < 0.1) were negatively associated with history of febrile UTI, while Anaerococcus (p < 0.1) was positively associated with history of febrile UTI (Table 2). In contrast, when examining model 1 using diversity as the quantification of the urobiome (via the Tail statistic or Shannon diversity index) the inclusion of neither diversity statistic improved the prediction of history of febrile UTI when compared to the inclusion of covariates and VUR grade alone. When VUR grade was dichotomized into low (0–3) and high (4 and 5) the improvement in the prediction of a febrile UTI history was negligible. The additive log ratio (ALR) dichotomized VUR model had a p-value of 0.057, versus the ordered categorical p-value of 0.076. Similarly for diversity measures, the dichotomization of VUR grade did not improve its predictive strength.

Table 2.

Predictors of history of febrile UTI (Model 1).

Variable	Estimate	Standard error	P-value
Age (years)	0.0734	0.0969	0.449
Female	4.2950	1.0217	< 0.01
Antibiotic prophylaxis	0.5125	0.6744	0.447
Highest grade of VUR	− 0.0242	0.2185	0.912
Anaerococcus	0.4023	0.2256	0.075
Streptococcus	− 0.2911	0.1740	0.094
Ezakiella	− 0.5780	0.2801	0.039
Corynebacterium	− 0.3560	0.2017	0.078

*Only the 4 taxa with significant p-values included in table, remaining 21 taxa that were not significant were left off table.

Covariates as predictors of the urobiome (Model 2)

With the 25 most abundant taxa as the multivariate response, multivariate analysis of variance (MANOVA) revealed that all variables except history of febrile UTI [age and sex, (p < 0.001); prophylaxis and VUR grade (p < 0.01)] contributed to predicting the urobiome as a collective. Univariate analysis identified multiple associations with age, sex, VUR grade, and febrile UTI history, and one significant positive association between antibiotic prophylaxis and Finegoldia (p < 0.1) (Table 3).

Table 3.

Coefficients of urobiome as a response (Model 2).

Variable	Age (years)	Female	Prophylaxis	Highest Grade of VUR​	History of febrile UTI
Acinetobacter​	− 0.09**	− 0.97**	0.14	− 0.08	− 0.01
Actinobacillus ​	− 0.05**	0.10	− 0.34	− 0.06	0.16
Afipia​	− 0.05	− 0.19	− 0.07	− 0.12	0.02
Anaerococcus​	0.12**	0.58	0.19	− 0.03	0.66
Bacillus​	0.07**	− 0.39	0.33	0.09	− 0.46
Campylobacter​	0.07	0.96*	0.11	− 0.13	0.54
Corynebacterium​	0.01	− 0.21	− 0.18	− 0.15*	− 0.87**
Dialister​	0.11***	1.72***	0.17	− 0.16*	0.17
Enterobacter​	− 0.05*	0.32	0.01	− 0.10	− 0.82***
Enterococcus​	− 0.14**	− 0.18	− 0.45	0.35***	0.23
Escherichia​	0.09	− 0.60	0.61	− 0.04	− 0.68
Ezakiella​	0.11**	1.99***	0.13	− 0.16	− 0.37
Fenollaria​	0.16***	2.00***	0.48	− 0.14	0.14
Finegoldia​	0.09	− 0.23	0.90*	− 0.06	− 0.22
Hoylesella​	0.09*	1.61***	− 0.07	− 0.23*	− 0.28
Klebsiella​	− 0.03	− 0.42	− 0.23	− 0.03	− 0.21
Lacticaseibacillus​	− 0.04	− 0.02	0.15	0.06	0.13
Lactobacillus​	0.14***	0.52	0.19	0.16	− 0.24
Peptinophilius​	0.13**	1.09*	0.63	− 0.17	0.61
Porphyromonas​	0.06	1.88***	− 0.65	− 0.16	− 0.48
Prevotella​	0.10*	1.69***	0.59	− 0.21*	0.48
Staphylococcus​	− 0.07	− 2.25***	− 0.45	− 0.26**	− 0.24
Streptococcus​	0.03	0.77	− 0.27	− 0.07	− 0.93*
Varibaculm​	0.08*	1.70***	− 0.53	− 0.14	− 0.13
Veillonella​	− 0.07	0.44	− 0.37	− 0.01	− 0.16

Significant coefficients are indicated with asterisks. *** p<0.01; ** p<0.05; * p<0.1.

As a response to the included covariates, the Tail statistic was negatively associated with increasing VUR grade (coeff = − 0.255, p < 0.01) and positively associated with female sex (coeff = 0.659, p = 0.10). There was also a significant negative association between Shannon diversity index and increasing VUR grade (coeff = − 0.122, p = 0 < 0.01). Results from the distance-based (beta diversity) permutational multivariate analysis of variance (PERMANOVA) suggested higher grade VUR (p < 0.01), sex (p < 0.01) and age (p = 0.02) had significant associations with, and thus plausible effects on, the urobiome (Fig. 3). In PERMANOVA, the dichotomization of VUR Grade reduced its effect size (from η² = 0.0137 to η² = 0.0089) and its statistical significance (from p-value = 0.0070 to p-value = 0.0789), suggesting the ordinal categorization of VUR Grade is more informative.

Fig. 3.

Beta-diversity ordination plots of urobiomes of children with different demographic and medical characteristics. Small glyphs represent individual samples. The larger glyphs represent the centroid of the groupings. Samples were compared between children who (A) were male vs. female, (B) were on antibiotic (ABX) prophylaxis vs. those not on ABX prophylaxis, (C) did or did not have history of febrile urinary tract infection (UTI), or (D) had highest grade of VUR ranging from 0–5. Sex (A) and VUR grade (D), as well as age (data not shown), were significant predictors of inter-sample compositional urobiome distance.

Urobiome as predictor of VUR grade (Model 3)

The addition of the urobiome using taxonomic ALR as the predictor, did not significantly improve prediction of VUR grade as measured by Akaike Information Criterion (AIC): (reduced model without the urobiome AIC = 677.2, full model with the urobiome AIC = 681.7, difference in AIC = 4.5). Examining the coefficients and p-values of the covariates (age, sex, and antibiotic prophylaxis) identified only age as a negative predictor of VUR grade (p < 0.1). Of the twenty-five most abundant taxa, negative associations were identified with Enterobacter (p < 0.01), Prevotella, and Staphylococcus (p < 0.1), and a positive association with Enterococcus (p < 0.01). In contrast, when representing the urobiome through diversity, the comparison of the full vs. reduced model (with vs. without diversity), indicated that including either the Tail statistic or Shannon diversity index, improved model prediction. With VUR grade as the response, univariate statistics identified a negative association with both the Tail statistic (coeff = − 0.0946, p = 0.02) and Shannon diversity index (coeff = − 0.4889, p < 0.01) as predictor variables. When controlling for the Tail statistic, the covariates that made a significant contribution included: a negative association with age (coeff = − 0.0758, p = 0.02) and a positive association with antibiotic prophylaxis (coeff = 0.6202, p = 0.02) (Table 4). Note that when associations can be made with the Tail statistic but not with the Shannon diversity index, then the taxa contributing to differences in diversity will tend to be those that are less abundant. The Tail statistic is more sensitive to the diversity that is contributed by lower abundant taxa¹¹.

Table 4.

Coefficients of VUR as a response (Model 3).

Variable	Estimate	Standard error	P-value
Age (years)	− 0.0792	0.0322	0.02
Female	0.0767	0.2925	0.80
Antibiotic prophylaxis	0.6685	0.2652	0.01
Shannon Diversity Index	− 0.4889	0.1472	< 0.01
Age (years)	− 0.0758	0.0327	0.02
Female	0.0181	0.2957	0.95
Antibiotic prophylaxis	0.6202	0.2715	0.02
Tail	− 0.0946	0.0393	0.02

Identifying new clinical phenotypes with hierarchical clustering with multinomial logistic regression (HCAMLR) and cluster influencers

Cluster analyses of individuals’ urobiome identified new subject phenotypes. In contrast to PERMANOVA, where clinical variables are used to predict the distancing between samples by fitting a linear model, HCAMLR first defines clusters via hierarchical clustering, then iteratively uses tree cutting and multinomial logistic regression on resultant clusters to evaluate whether the clinical predictors can be associated with any of these clusters. The clusters are subsequently examined for taxonomic “cluster influencers,” thus implicating them, via the cluster as a proxy, as associated with the clinical variables. Clusters derived from urobiomes showed significant associations with differences in febrile UTI history, antibiotic prophylaxis, VUR grade, sex, and age (Fig. 4). The taxa driving formation of clusters with significant clinical associations varies. Associations with febrile UTI history identified two clusters: Cluster 1 (37% of samples) and Cluster 2 (63% of samples) (p = 0.057) that were enriched and sparsely represented, respectively, by samples from subjects with a history of febrile UTI. Cluster 1 was also significantly enriched in females. A diverse set of cluster influencer taxa were identified for Cluster 1 (enriched in those with a history of febrile UTI) including Peptoniphilus, Anaerococcus, Fenollaria, Hoylesella, Porphyromonas, and others. No specific cluster influencing taxa were identified for Cluster 2 (diminished in those with history of febrile UTI) indicating multiple taxa making smaller contributions to variation. A small cluster (Cluster 3; 0.5% of samples), in addition to holding fewer subjects with febrile UTI history. was also enriched in subjects with the higher grades of VUR and diminished in those on antibiotic prophylaxis. Enterococcus was identified as the taxon driving this cluster formation, which is consistent with the identification of the urobiome as a predictor of VUR grade (Model 3) where increasing the abundance of Enterococcus was also identified as a significant predictor of VUR grade (Supplemental Table 1).

Fig. 4.

Urobiome clusters and their clinical associations. (A) Dendogram representing the hierarchical clusters constructed for urobiome samples based on Manhattan distance. Six urobiome clusters were identified. (B) Clusters with significant associations with clinical data are annotated with red (positive) and blue (negative) bars. Subjects with antibiotic (ABX) prophylaxis were fewer in Cluster 3 (of note, Cluster 3 is enriched in Enterococcus). Subjects with higher grades of VUR (third bar) were positively associated with Cluster 3. Females (fourth bar) were positively associated with Cluster 1 (enriched in multiple taxa, including Peptoniphilus, Anaerococcus, Fenollaria, Hoylesella, Porphyromonas, and others) and negatively associated with Cluster 4. Older age (bottom bar) was positively associated with Cluster 5 and negatively associated with Cluster 6. Significance: *** p < 0.001, ** p < 0.01; * p < 0.05.

Discussion

Here, we report that in children with VUR, there are associations between certain taxa (specifically the presence of Anaerococcus and absence of Streptococcus, Ezakiella, and Corynebacterium) that, when examined together with age, sex, antibiotic prophylaxis, and highest grade of VUR, improved the prediction of febrile UTI history (Model 1). This suggests the urobiome may play a role in, or be altered as a response of, febrile UTI history. In contrast, the strength of associations in our models suggest that the urobiome was responsive to VUR grade (Model 2), but did not improve prediction of VUR (Model 3). These results suggest that VUR is associated with an environment that is conducive to the establishment and growth of an altered urobiome.

After controlling for age, sex, and the use of antibiotic prophylaxis, we found both a positive association between grade V VUR and the presence of Enterococcus, and a negative association between grade V VUR and Enterobacter. Surprisingly, we also found that use of antibiotic prophylaxis was only marginally associated with a change in urobiome composition. We had hypothesized that use of prophylaxis would be associated with altered community structure, but our data does not support this hypothesis. One possible explanation is that children had been on antibiotic prophylaxis that was discontinued prior to the VCUG, and therefore they were included in our “no prophylaxis” group while effects from the previous antibiotic prophylaxis may have continued to be seen in the urobiome. However, in most clinical scenarios, the decision to stop antibiotic prophylaxis is made after the VCUG results are available, so those children would typically still be on prophylaxis at the time of VCUG. We were unable to collect prophylaxis adherence data on the children, so the detection of significant associations may have been affected by their actual adherence. Conversely, it is also possible that some children in the “no prophylaxis” group had received treatment antibiotics. While we excluded children who had received antibiotics to treat an infection within the two weeks prior of the VCUG, it is possible that this was not reliably recorded in the electronic health records. In addition, there is a lack of data to suggest if and when the urobiome normalizes after antibiotic exposure. Therefore, it is possible that antibiotics that were completed earlier than two weeks before the VCUG still affected the urobiome.

Prior work has shown that children with high grade VUR are more likely to have UTIs due to non-E. coli bacteria³. One study noted that the causative bacteria in infections in children with VUR include Klebsiella and Enterococcus³. Further, data from the RIVUR trial, which included children with a history of symptomatic febrile UTI and VUR, reported that Klebsiella was the most-common non-E. coli bacteria in children with VUR in that cohort, followed by Pseudomonas, and then Enterococcus⁹. Another study showed that, in children with UTI, individuals with VUR had increased abundances of Pseudomonas and Halomonas genera compared to those without VUR¹². However, the mechanism by which children with VUR have higher rates of non-E. coli UTIs is not clear. Friedman et al. hypothesized that the abnormal anatomy allowed the presence of these less virulent bacteria in the genitourinary system¹³, but do not comment on why this occurs. Others have hypothesized that the urinary stasis present in VUR as a result of retrograde flow may allow bacteria that lack the adhesins present on uropathogenic E. coli to ascend to the kidney³. Some authors have investigated potential alterations in the innate immune response, and have found copy number variants in genes associated with innate immunity in children with VUR¹⁴, or alterations in genes encoding for specific antimicrobial peptides¹⁵ or in intergenic regions located near genes involved in immunity and inflammation¹². Together, these data suggest that defects in the innate immune response may play a role in these non-E. coli UTIs. While our cross-sectional data cannot directly uncover causality, the strong association between urobiome and VUR grade do suggest that the environment of VUR is conducive to an altered urobiome, which may also be one that promotes UTI.

Many children with VUR either have resolution of their VUR, and those that do not or have high grade VUR undergo surgical correction. However, some children still experience renal-related adverse events as adults, despite surgical correction of VUR. This suggests that there are additional mechanisms at play beyond the effects of the retrograde flow of urine into the kidneys. We are unable to comment whether the urobiome may play a role in these long-term sequelae as it is well outside the scope of our work. However, future longitudinal studies may elucidate a role for the urobiome in long-term outcomes of children with VUR who undergo surgical correction of their lesion.

Limitations of this study included its cross-sectional design. The lack of both longitudinal data and follow-up information limited our ability to hypothesize causality more definitively. Further, all clinical data was collected by retrospective chart review, which limited our ability to obtain all clinical variables from all patients, as well as the ability to obtain adherence data on antibiotic prophylaxis. Despite collecting almost 400 urine samples, we only observed a limited number of children with grade V VUR without an additional anatomical anomaly. Further, we were unable to include a healthy control group in this study as it is not ethical to obtain catheterized urine specimens from healthy children for the purpose of a research study. Finally, we used 16S rRNA gene sequencing, which although is commonly used for microbiome analyses to provide genera-specific read count profiles, it cannot reliably differentiate taxa at the level of species or strain, nor can it directly measure microbiome function. The low microbial biomass in urine samples may also aggravate PCR bias, thus increasing the variance of assayed read counts. However, despite these limitations, this study represents the largest cohort of children with VUR whose urine samples were all collected by catheterization and appropriately processed for urobiome analysis. The cohort size allowed our analysis to control for several relevant variables such as age, sex, and use of antibiotic prophylaxis, and the catheter-based urine collection avoided post-urethral contamination, thus yielding more accurate urobiome profiles. Together, these strengths contributed towards the robustness of our conclusions.

Methods

Study design and patients

This is a cross-sectional study convenience sample of all patients who presented to a single center for VCUG between 12/1/2020 and 10/3/2023. Children were excluded if they had any surgical procedures of the genitourinary tract in the 30 days prior to VCUG or were on treatment antibiotics at the time of VCUG. We collected all urine samples from children who presented for VCUG and for whom the technologist was able to collect urine during the catheterization process. The Institutional Review Board at the University of Pittsburgh approved this study and informed consent was obtained after sample collection but before retrieval of data from the electronic health record. If a family did not consent, then the sample was destroyed. All research in this study was performed in accordance with the Declaration of Helsinki.

Urine samples

Urine samples were collected by catheterization and immediately refrigerated. Samples that were used for urobiome sequencing had AssayAssure (Sierra Molecular) added in a 1:9 ratio prior to freezing at − 80 °C. Samples were frozen within four hours of collection.

Covariates

We extracted covariates, including demographic information, indication for VCUG, VCUG results, whether any prior UTI was febrile, and use and type of antibiotic prophylaxis at time of VCUG, from the electronic medical record. Grade of VUR was identified as the highest grade on the VCUG as read by the radiologist. If bilateral VUR was present, we recorded the highest grade. We included VUR grade as an ordinal categorical variable in the subsequent analyses, with children with no VUR assigned a “0”.

Genomic DNA extraction

Approximately 5 ml of urine per sample was centrifuged. If less than 5mL of sample was available, the entire volume was centrifuged (range 2-5 ml). Supernatant was removed to leave an approximate ~ 300 µl volume. Pelleted cells were resuspended in the remaining 300 µl of supernatant for extraction. DNA extraction was performed using the Qiagen Powersoil Microbiome Kit EP for automated DNA extraction using an Eppendorf, 5075VTC liquid handling workstation. HEPA filtration was used during sample processing, and the workstation was UV sanitized between batches. Specimens were processed per manufacturer’s protocol with the following modifications: Resuspended specimen was added to individual bead beating tubes to ensure no carryover between samples during the bead beating process. Aliquots from the individual tubes were then transferred to 96-well blocks for completion of the automated genomic DNA extraction process. Reagent blanks were included as negative controls. Our positive controls were cells and genomic DNA from a microbial community of known composition (ZymoBiomics Microbial Community Standards; Zymo Research, Irvine, CA). As a component of the quality control process, positive controls were evaluated across sample batches to evaluate laboratory and sequencing performance and compared to historical performance of 16S rRNA gene sequencing at the Center for Medicine and the Microbiome (CMM). No significant batch deviation was identified in this project.

Bacterial community sequencing

Extracted genomic DNA was amplified for the V4 region of the 16S rRNA gene using Q5 HS High-Fidelity polymerase (New England BioLabs, Ipswich, MA) with inline barcode primers design based on the method of Caporaso et al.¹⁶. V4 primer sequences were: 515f 5ʹ-GTGCCAGCMGCCGCGGTAA-3ʹ and 806r 5ʹ-GGACTACHVGGGTWTCTAAT-3ʹ. Approximately 5–10 ng of each sample were amplified in 25 µL reactions. Cycle conditions were 98 °C for 30 s, then 25 cycles of 98 °C for 10 s, 57 °C for 30 s, and 72 °C for 30 s, with a final extension step of 72 °C for 2 min. Amplicons were purified with AMPure XP beads (Beckman Coulter, Indianapolis, IN) at a 0.8:1 ratio (beads: DNA) to remove primer dimers. Eluted DNA was quantitated on a Qubit fluorimeter (Life Technologies, Grand Island, NY). Sample pooling was performed on ice by combining 40 ng of each purified band. For negative controls and poorly performing samples, 20 µL of each sample was used. The sample pool was purified with the MinElute PCR purification kit (Qiagen, Germantown, MD). The final sample pool underwent 2 more purifications: AMPure XP beads to 0.8:1 to remove primer dimers, and a final cleanup in Purelink PCR Purification Kit (Life Technologies). The purified pool was quantitated in triplicate on the Qubit fluorimeter prior to sequencing.

The sequencing pool was prepared as per Illumina’s recommendations (Illumina, Inc., San Diego, CA), with an added incubation at 95 °C for 2 min immediately following the initial dilution to 20pM. The pool was then diluted to a final concentration of 7pM + 20% PhiX control (Illumina). Sequencing was done on an Illumina MiSeq 500-cycle V2 kit (Illumina).

Bioinformatics

Sequences from the Illumina MiSeq were deconvolved and then processed through the CMM in-house sequence quality control pipeline, which includes dust low complexity filtering, quality value (QV < 30) trimming, and trimming of primers used for 16S rRNA gene amplification, and minimum read length filtering. Using the scripts fastq_quality_trimmer and fastq_quality_filter from Hannon’s Cold Spring Harbor Laboratory’s FASTAX-Toolkit (http://hannonlab.cshl.edu/fastx_toolkit/), reads were trimmed until the QV was 30 or higher. Trimmed reads shorter than 75 bp or those with less than 95% of the bases above a QV of 30 were discarded. Forward and reversed paired reads were merged with a minimum required overlap of 25 bp, proportion overlap mismatch > 0.2, maximum N’s allowed = 4, and a read length minimum of 125 bp. Forward and reverse reads were merged into contigs then processed through the CMM’s Mothur‐based (Mothur 1.48.2)¹⁷ 16S rRNA gene sequence clustering and annotation pipeline. Sequence taxonomic classifications were performed with the Ribosomal Database Project’s (RDP) naïve Bayesian classifier¹⁸ with the SILVA 16S rRNA database (138.2)¹⁹.

16S rRNA gene sequence taxonomic profile analyses

Three general categories of microbiota analyses were applied to the taxonomic profiles: abundance-based, distribution-based, and distance-based. These methods were utilized to associate the profile data with clinical data and test the three VUR/urobiome models.

Abundance-based analyses

The abundance-based analyses provide a means to consider each taxon as an independent and normally distributed entity after they have been transformed with the additive log ratio (ALR) transformation²⁰. This reduces the spurious correlations between taxa resulting from the compositional nature of 16S rRNA gene sequence profiles. The transformation is necessary for each taxonomic category to be considered as independent predictors in a multiple regression or as a multivariate response²¹. The twenty-five most abundant taxa, by average abundance across all samples, were selected for inclusion into the regression models. The remaining taxa not in the top were accumulated into a remainder and used as the denominator of the log ratio.

Distribution-based analyses

The distribution-based analyses consider the entire composition of taxa within a sample with a single metric. Here the Tail statistic²² and Shannon diversity index were examined (with higher values denoting greater diversity), and samples are compared to each other with this computed value. Alpha (i.e. diversity within a single sample) diversity indices are useful quantifications of compositional data represented visually as stacked barplots and rank abundance plots.

Distance-based analysis

The distance-based analysis computes the beta-diversity by examining the pairwise difference between two samples’ profiles using the Manhattan distance. Ordination plots were generated with multidimensional scaling (MDS). Subsequent analyses focus on the association of variables with clustering or linear models. PERMANOVA²³ was performed with the “adonis2” function from the “vegan” library in R to associate targeted variables with inter-sample distances and test for their statistical significances.

Statistical models

We proposed three linear models to examine the relationship between febrile UTI, VUR grade and urobiome (Fig. 1). Across all three models, age, sex, and antibiotic prophylaxis were included as covariates to control for their potential effects on the outcomes of febrile UTI, the urobiome, or VUR grade. Each of the three models proposes a different hypothesis, and asymmetries in their fit may suggest plausible, although not be proof of, causality. The first model (Model 1) includes covariates, VUR grade and the urobiome to predict history of febrile UTI. This model identifies whether history of febrile UTI is aggravated or mitigated by the urobiome, and if VUR grade is a predisposing factor. The second model (Model 2) includes covariates, VUR grade and history of febrile UTI to predict the urobiome. This model identifies biomarkers in the urobiome that may exist if a subject has a history of febrile UTI. The third model (Model 3) includes covariates and the urobiome to predict VUR grade. This model identifies members of the urobiome that could be plausibly modulating VUR severity through mechanisms that may include chronic inflammation, fibrosis, or other processes. Along with comparing the significance of asymmetric associations between the outcome variables when modeled as predictors or response variables, overall model fit was also assessed with the Akaike Information Criterion (AIC). In addition, McFadden’s R-squared (pseudo R-squared) was also used to quantify the improvement of the fitted (full) models when the urobiome was included, over the covariates-only (reduced) models.

Identifying new clinical phenotypes

Hierarchical clustering with multinomial logistic regression (HCAMLR)²⁴ was used to associate specific clusters of individuals formed by their urobiome with clinical variables. Multinomial logistic regression was iteratively applied to evaluate whether clinical variables (predictors) were significantly associated with clusters (response). Hierarchical clusters were computed with Ward’s Minimum Variance algorithm using the Manhattan distance to represent inter-sample distances. The resultant tree was iteratively cut at increasing cluster counts k, from 2 to max clusters. For each k, a multinomial logistic regression model was fit for each of the k generated clusters to estimate the regression coefficients across the clinical predictors. At each iteration, the most significant regression coefficient p-value among the clusters i = 1 to k, was retained for each of the predictors. After the last iteration (max clusters), each of the predictors’ accumulated p-values were evaluated to determine the “optimal cluster cutoff” by identifying the iteration that produced the smallest p-value for that predictor. The optimal cluster cutoff for a specific clinical predictor refers to the value of k that split the cohort into clusters with the most statistically significant “uneven” representation of the clinical predictors across all the clusters. Note that fitting the logistic regression allows the models to control for other predictors, which would not be possible with a basic chi-squared test.

Cluster influencers

Clusters were examined for “influencer” taxa to explain why each cluster separated from the other clusters at the same cut height²⁵. An analysis of variance (ANOVA)-based R² ratio was computed between pairs of clusters to identify whether excluding a specific taxon would increase their separation. The set of taxa that reduced cluster separation when removed were considered the cluster influencers. These cluster influencers were then used as annotation for the clusters that were identified to be of interest in HCAMLR.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Theresa C. Barrett: Collected and analyzed data and wrote and revised the manuscriptKelvin Li: Conceptualized data analysis plan, analyzed and interpreted data, wrote and revised the manuscriptAdam Fitch: Organized data tracking, analyzed data and contributed data interpretationsAsha Patel: Performed sample and primary data tracking, generated 16S rRNA gene sequencingBhanu Sharma: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptJacqueline Holden: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptKermina Abdelsayed: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptNina N. Kowalewski: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptSridhar Narla: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptDaniel Bushnell: Collected and processed samples, critically edited the manuscript and approved the final version of this manuscriptBarbara A. Methé: Contributed to sequence data generation, conceptualized data analysis plan, analyzed and interpreted data, wrote and revised the manuscriptCatherine S. Forster: Conceptualized and designed the study, collected and analyzed data, and wrote and revised the manuscript.

Funding

CF was funded by 1K23DK129783 and in part by G. Nicholas III and Dorothy B. Beckwith.

Data availability

The original contributions presented in the study are publicly available in NCBI using accession number PRJNA1366551.

Declarations

Competing interests

The authors declare no competing interests.