Assessing the Relationship Between Nitrate-Reducing Capacity of the Oral Microbiome and Systemic Outcomes

Abstract

The significance of the oral microbiome in the generation of the nitric oxide (NO) via the enterosalivary nitrate-nitrite-nitric oxide pathway is increasingly recognised, directly linking the oral microbiome to cardiometabolic outcomes influenced by NO. The objective of this chapter is to outline a strategy of identifying pathway-specific bacterial taxa or predicted genes of interest from 16S rRNA data, specifically in the enterosalivary pathway of nitrate reduction, and analysing their relationship with cardiometabolic outcomes using multivariable regression models.

1. INTRODUCTION

Nitric oxide (NO) is an important signaling molecule involved in many physiological processes and its deficiency has been implicated in the pathogenesis of hypertension and insulin resistance^1–3. Thus, NO bioavailability has garnered much attention as an increase in NO production can potentially reduce high blood pressure and blood glucose levels. As NO was thought to be produced only by NO synthases in the endothelium, immune cells and other tissues; previous research has focused on enhancing NO bioavailability produced through this synthase pathway.

However, nitrate resulting from NO oxidation metabolism or from dietary nitrate consumption has since been found to provide an important storage pool for NO. The physiological recycling of nitrate to produce NO takes place via the enterosalivary nitrate-nitrite-nitric oxide (NO3-NO2-NO) pathway. In this alternative pathway, oral bacteria play a crucial role by reducing salivary nitrate to nitrite, which is then swallowed and made systemically available for further reduction into NO in the blood vessels and tissues¹. The direct role of oral bacteria in this NO₃-NO₂-NO pathway was demonstrated by several experimental studies that use antibacterial mouthwash to reduce oral bacteria, resulting in decreased nitrate-reducing capacity and a corresponding increase in blood pressure and plasma glucose^4–8. Not all oral bacteria contribute to the reduction of salivary nitrate, and specific taxa with nitrate-reduction capacity have been identified^9,10. More recently, studies have observed a correlation between baseline nitrate-reducing bacteria abundance and differential blood pressure responses to dietary nitrate supplementation¹¹. Associations of bacteria taxa and genes coding for bacterial enzymes involved along the NO₃-NO₂-NO pathway with blood pressure levels have also been observed^12,13, further emphasizing the importance of oral microbiome composition in NO generation.

The enterosalivary pathway of NO generation has gained significant attention in recent years¹⁴, and presents an alternative target for manipulation to improve NO bioavailability-associated cardiometabolic outcomes. While many have examined this pathway through the effects of dietary nitrate supplementation (i.e., increasing the storage pool of nitrate) on cardiometabolic outcomes^15,16, fewer population-based studies have explored the association of the oral microbiome involved in the NO₃-NO₂-NO pathway with cardiometabolic outcomes.

Most microbiome analyses seek to identify differentially abundant taxa between disease states. The highly dimensional oral microbiome, with a large number of taxa to be compared, results in multiple comparisons and an increased false discovery rate¹⁷. The identification of a pathway-specific hypothesis linking the oral microbiome and cardiometabolic outcomes, such as the NO₃-NO₂-NO pathway, allows us to narrow our focus on specific bacteria or bacterial genes of interest, resulting in a priori hypothesis-driven analyses.

While whole genome shotgun metagenomic sequencing can directly yield information on the functional capacity of nitrate metabolism pathways in the oral microbiome, the relatively high cost of metagenomic sequencing and data handling may be prohibitive. Moreover, there are currently available 16S rRNA sequencing data within large cohorts that will yield valuable prospective data on the incidence of cardiometabolic outcomes and it would be a missed opportunity to not fully capitalize on those data.

The aim of this chapter is to provide explicit step-by-step examples demonstrating the identification of pathway-specific taxa or genes of interest, and their operationalization from 16S rRNA sequencing data. This exposure construct can then be leveraged in traditional statistical analysis workflows exploring the association between microbial nitrogen metabolism capacity and cardiometabolic outcomes.

2. MATERIALS

2.1. Next-Generation High-Throughput 16S rRNA Sequencing

Sequence-based microbiome analysis consists of several steps: sample collection, storage, DNA extraction, library preparation, next-generation high-throughput microbial sequencing, quality control, sequence identification, and finally statistical analysis¹⁸. Briefly, bacteria from the samples collected are lysed and the DNA extracted. In library preparation, the 16S rRNA gene – the most commonly used marker gene in oral microbiome studies and the gold standard for sequence-based bacterial analyses – is amplified from the extracted DNA. These amplicons are then sequenced on the sequencing platform of choice (e.g., Illumina) to produce sequence reads. Quality control is then carried out to filter out short reads or sequences with lower quality scores before assigning the sequences to taxonomic classifiers. Several useful papers and references are available discussing the best practices and considerations at each stage to reduce the potential biases introduced^17–22.

2.2. Taxonomic Classification of Sequence Reads

In general, sequence reads obtained from sequencing are referenced against known microbial reference databases (e.g., SILVA, RDP, or Greengenes database) to assign a taxonomic identifier. A description of the different methods for sequence identification is beyond the scope of this chapter, and other papers provide an overview of the process¹⁷. Our previous analysis¹² utilized the Human Oral Microbiome Identification using Next Generation Sequencing (HOMINGS) methodology specifically designed for the oral microbiome to generate species-level information, which uses a customized BLAST program (ProbeSeq for HOMINGS)^23,24. Other methods of taxonomic classification of 16S rRNA sequence reads are available, such as using operational taxonomic units (OTU) clustering²⁵, or the newer DADA2-corrected amplicon sequence variants (ASV)²⁶. The HOMINGS methodology has been shown to be largely equivalent to the tree-based OTU clustering approach²⁷, but increasingly the use of ASVs is recommended as the standard unit of marker-gene analysis and reporting^17,26. Researchers have a choice of bioinformatics pipeline and reference database, and are recommended to document the software versions used and all commands run¹⁷.

2.3. Required data for this Chapter

For this chapter we will assume the following:

1. Standard 16S rRNA next-generation sequencing has been carried out on microbial DNA.

2. The necessary bioinformatics quality controls have been performed (e.g., filtering and trimming)^19,22

3. An appropriate technique for sequence inference and taxonomic alignment has been used

4. A final table of OTUs/ASVs relating to taxa at the species level and their relative abundances in each sample has been produced. If using predicted gene abundances from 16S rRNA data, we will assume that PICRUSt2 analysis has been successfully conducted, using 16S data to infer KEGG ortholog abundances.

2.4. Required Datasets

1. Taxonomic table with relative abundances, such as shown in Table 1.

2. Metadata of participants including the clinical outcomes of interest, e.g., systolic blood pressure, insulin resistance, such as shown in Table 2.

3. (Optional) Output from PICRUSt2 with KEGG ortholog functional abundance. (See Table 3 example)

Table 1.

Truncated example of OTU/ASV output table after 16S rRNA sequencing and taxonomic alignment in the long format

Taxa	Relative_abundance	ID
Actinomyces_johnsonii	0.00001	1
Actinomyces_massiliensis	0.00389	1
Actinomyces_meyeri	0.00184	1
Actinomyces_naeslundii	0.05540	1
Actinomyces_odontolyticus	0.00001	1
...	...	...
Actinomyces_johnsonii	0.000008	2
Actinomyces_massiliensis	0.000823	2
...		2

Note the taxa is different in each row but with the same ID, and each ID will have relative abundance data for each individual taxon.

Variable Key: Taxa refers to OTUs/ASVs relating to taxa at the species level; ID refers to the unique participant or sample ID. This dataset has one sample per participant. Relative abundance is calculated from absolute counts of that taxa sequence divided by the total counts across all taxa in the individual sample.

Table 2.

Example of participant metadata containing cardiometabolic outcomes of interest in wide format.

ID	Age	Sex	Glucose	MeanSBP	...
1	25	F	85	95.5
2	31	F	78	116.5
3	41	F	94	111.0
4	30	M	78	123.5
5	22	F	90	117.5
...

Study participants are not repeated across rows and each new variable (e.g., age, sex, mean systolic blood pressure) is represented as a new column variable.

Summary scores (taxa or predicted gene-based) are added as a new column in the final dataset used in linear regression analyses.

Variable Key: ID refers to the unique participant ID; Glucose refers to the fasting plasma glucose levels of the participant; MeanSBP refers to the mean systolic blood pressure of the participant.

Table 3.

Truncated example of PICRUSt2 output predicted absolute gene abundances using KEGG Orthologs (KOs) conducted on 16S rRNA sequencing data, with KOs in rows and Subject IDs in columns.

KEGG_orthology	ID1	ID2	ID3	...
K00367	34	205	29
K00370	10141	25010	31940
K00371	10388	7487	10450
...

Variable Key: Column heads have ID numbers, referring to the unique participant ID. Each row has one KEGG Ortholog (KO), defined as functional orthologs containing groups of genes. Cells contain absolute counts of that KO in the participant’s samples. Relative gene/KO abundances will be calculated before summing into a summary score.

2.5. Software

1. Statistical software to be used for analysis, e.g., SAS and R.

2. (Optional) PICRUSt2 or Piphillin software packages, if using predicted gene abundance from 16S rRNA sequences.

3. METHODS

An important statistical issue in microbiome analysis is the high dimensionality of microbiome data which includes thousands of taxa. With the enterosalivary nitrate-nitrite-NO pathway of interest, the selection of certain bacteria a priori is possible based on existing knowledge^9,10. While individual taxa can be modelled one-by-one in regression models, statistical hypothesis testing may need to be adjusted for the false discovery rate, which reduces statistical power. In addition, since individual taxa analysis may fail to capture the many complex interactions between bacteria co-existing in a microbial community, a summary score can be a useful feature to give an overall picture of the microbiome community’s nitrate-reducing capacity.

There are two general approaches that can be used to create a summary score from 16S sequencing data: 1) a taxa-based score, using taxa a priori identified to be associated with nitrate-reducing capacity; and 2) a predicted metagenomic (gene)-based score in which scores are based on the estimated number of genes relevant to nitrate reduction.

An advantage of the method of creating a summary score of bacteria taxa previously identified in the literature to be of importance is the specificity of the taxa selected. Numerous oral bacteria are thought to contain nitrate reductase genes, and incorporation of all bacteria containing any nitrate reductase gene into the exposure summary score may result in greater variability and noise, thus masking the effect of the important species and biasing the effect estimate towards the null.

An alternative to using taxa already associated with a pathway of interest is available. In situations where key bacterial species are not in the literature, but a functional gene(s) of interest has been identified (e.g., nitrate reductase), the use of predicted metagenomic content to operationalize the exposure may be useful. It should be noted however, that both methods still rely on taxonomic classification from 16S rRNA marker gene sequencing and that microbial traits, such as horizontal gene transfer between bacteria or strain-level variation within species (e.g., differential nitrate-reduction capacity between strains), make misclassification of the individual’s true nitrate-reducing capacity possible.

3.1. Summary Score of Bacterial Taxa Associated with Nitrate-Reducing Capacity

3.1.1. Identification of bacteria associated with nitrate-reducing capacity

To identify the bacterial species of interest and create a summary score, a literature review was used to identify bacterial species associated with the nitrate-reducing capacity of oral microbiome samples¹². Two reference papers were used to identify the bacterial species of interest in nitrate-reduction^9,10. Doel et. al used culture-based techniques to isolate and identify only nitrate reductase-positive bacteria⁹, and all bacteria identified regardless of rate of nitrate-reduction was included. Additionally, Hyde et al. compared samples with high versus low nitrate-reducing capacity and used next-generation sequencing methods to identify species that were differentially abundant in the highest nitrate-reducing sample¹⁰. The latter sought to provide a whole community picture, including species indirectly helping in nitrate reduction; therefore, while most of the candidate species identified have a nitrate-reductase gene, some like P. melanogenica do not but contain a nitrite reductase gene instead. As our goal was to optimise the measurement of nitrate-reducing capacity, all species identified as candidate species, whether directly or indirectly contributing to nitrate-reducing capacity as “helper” species, were included in the summary score.

3.1.2. Operationalization using the arcsine-square root transformation of taxa relative abundance

From a taxonomic table of OTU/ASVs with absolute counts (i.e., counts of sequence hits), the relative abundance of each taxa is calculated by dividing the number of counts observed for that taxa sequence by the total counts across all taxa in the individual sample. The resulting relative abundance measure is a proportion (i.e., compositional) that is highly skewed and constrained to the range of zero to one with many zeros present (i.e., zero-inflated).

The arcsine-square root transformation has been widely used on taxa relative abundance to examine differentially abundant taxa between groups ^28–31. This transformation reduces the skewed distribution, creating a more normally distributed continuous variable that can range in the negative, stabilizing the variance, and allowing it to be effectively used in linear regression models. Unlike studies which use the microbiome as the dependent variable of interest, we use microbial relative abundance as the exposure or independent variable. Therefore, we do not employ statistical methods that model absolute counts, instead of relative abundance, with zero-inflated Poisson³² or negative binomial models³³ as others have. The arcsine-square root provides a simple transformation that can be easily performed in all software and is often used as a baseline comparison with the newly developed methods ^34,35.

3.1.3. Standardization and Creation of a Summary Score for Bacteria

Before summing the selected taxa in a summary score, standardization is first carried out. This gives equal weight to each taxon in the score, preventing very high relative abundance taxa from dominating the score. This is important especially when it is unknown whether the actual nitrate-reducing capacity of each taxon is directly correlated with its relative abundance. For example, it is plausible that nitrate-reducing capacity might vary by taxa. A sum of the relative abundances of taxa of interest without standardization would simply represent the total relative abundance of the selected bacteria in the sample. Therefore, the arcsine square-root transformed relative abundance of each taxa for an individual is standardized via division by the taxon’s standard deviation across all the samples. The standardized values for the selected bacteria are then summed to create a summary score for each individual.

3.2. Summary score of Predicted Metagenomic genes from 16S rRNA sequencing on the NO₃-NO₂-NO Pathways

3.2.1. Prediction of metagenomic content from 16S rRNA sequencing

Bioinformatics tools can be used to predict metagenomic content from 16S rRNA marker gene sequencing. Piphillin and PICRUSt2 are the two most well-known tools for inferring metagenomic content^36–38.

These tools use the taxonomic identification, the relative abundances of the taxa, and a reference database of known bacteria genomes. The output is a functional-gene-count matrix, providing an estimate of the count of each functional gene in each sample. Comprehensive tutorials are available on how to use these tools³⁹.

3.2.2. Identification of functional gene orthologs on the enterosalivary pathway

Genes of interest can be identified by searching the Kyoto Encyclopaedia of Genes and Genomes (KEGG) Pathways for the pathway of interest, in this case the bacterial nitrogen metabolism pathway (Figure 1). Enzymes involved in each step of the NO₃-NO₂-NO pathway are mapped out. From Figure 1, nitrate reduction involves nitrate reductase enzymes EC 1.7.7.2, 1.7.5.1, 1.9.6.1, NR and 1.7.99. Selecting the respective enzyme, for example EC 1.7.5.1, will show the associated KEGG Ortholog (KO)s – functional orthologs containing a group of bacterial genes coding for that molecular-level function. These groups of genes include nitrate reductase genes such as narG, narH, narI, napA, napB, narB, nasA, and nasB. More details on the structure, organization and uses of the KEGG encyclopaedia are available elsewhere⁴⁰.

Figure 1.

KEGG pathway map00910 – Nitrogen metabolism in bacteria. Publicly accessible from https://www.genome.jp/kegg-bin/show_pathway?map00910. Nitrate-reduction enzymes are highlighted in yellow and include EC 1.7.7.2, 1.7.5.1, 1.9.6.1, NR, and 1.7.99. Selecting nitrate reductase enzyme “EC 1.7.5.1” for example brings up K00370, K00371 and K00374 as KO gene groups, containing narG, narZ, nxrA, narH, narY, nxrB, narI, and narV genes. KOs are defined as functional orthologs, and may consist of a single bacterial gene or genes from closely related species.

3.2.3. Operationalization of predicted gene abundance summary score

From bioinformatics tools PICRUSt2 and Piphillin, absolute counts of all possible KEGG KOs in each sample are output. As with taxa, the absolute counts of specific KOs can be converted into relative KO gene abundance by division with the total counts across all possible KOs in that individual sample. The relative gene abundances of interest are then added into a summary score and normalized using the arcsine-square root transformation as performed on the taxa relative abundance in Section 3.1.2.

Examining the NO₃-NO₂-NO pathway using predicted metagenomic content can be operationalised in several ways. As a start, the summary score containing all nitrate-reducing genes can be calculated. Our ongoing methodological work explores incorporating competition from other bacterial genes, such as nitrite-reductase into the summary score¹⁰, and creating summary scores based on bacterial metabolic pathways (e.g., respiratory denitrification), to further examine the different parts of the NO₃-NO₂-NO enterosalivary pathway in relation to cardiometabolic outcomes. It should however be emphasized that estimating metagenomic content from 16S rRNA sequencing does not directly measure the bacterial genes in the microbiome, and the specific strains present in the samples may not have the same functions as mapped in the bacteria reference database.

3.3. Analysis of the association between the taxa and the outcome of interest

Multiple methods for analysing microbiome data have been developed and the standards for microbiome analyses are rapidly evolving^17,41,42. Developments include the use of alternative parameters, such as the change in ratios of taxa to address biases introduced by comparing relative abundances between samples^43–45. In this chapter, we analyse microbial relative abundance as the exposure or independent variable in linear regression models, adjusting for potential confounders. Patient selection criteria may be broader in population-based cross-sectional studies. Importantly, those who took antibiotics less than 30 days before recruitment were excluded, and microbiome studies often exclude those who report taking medications such as proton pump inhibitors. The patient selection criteria would likely be more restricted for smaller studies with a different study design, where control of confounding through design is necessary, or where smaller sample sizes limit the power for multivariate regression analysis.

3.4. Examples of datasets used in the analysis

See Tables 1, 2, and 3.

3.5. Code for running bacterial taxa summary score analyses in SAS

In this section, we provide step-by-step instruction for operationalizing a nitrate-reducing taxa summary score as discussed in Section 3.1 above. We also provide the SAS code for analysing the created summary score with a cardiometabolic outcome of interest, as in our previous work using multivariable regression models^12,46. We have used CAPS for SAS keywords and mixed case for user-supplied text in keeping with typographical conventions used in SAS textbooks.

1. /* Import the OTU/ASV output from 16S rRNA sequencing and sequencing analysis with relative abundances */ (see Note 1).

   PROC IMPORT DATAFILE= “C:\Users\CG\16SOTUrelativeabundance.csv” OUT=microbiome DBMS=CSV REPLACE;

   GETNAMES=YES;

   RUN;

2. /*Import the participants metadata including Subject ID, socio-demographics, and clinical outcomes. Example of such a dataset is provided in Table 2 */

   PROC IMPORT DATAFILE= “C:\Users\CG\Participantsclinicaldata.csv” OUT=metadata DBMS=CSV REPLACE;

   GETNAMES=YES;

   RUN;

3. /*Arcsin-square root transformation on the relative abundance of each taxa*/

   DATA noarc;

   SET microbiome;

   transra= ARSIN (SQRT(relativeabundance));

   RUN;

4. /*Creating a variable called ztransRA, which will be standardised in the next step*/

   DATA arcsinz;

   SET noarc;

   ztransRA=transRA;

   RUN;

5. /* SAS function that carries out the z score standardization, creating a mean of 0 and standard deviation of 1 across all samples. The dataset has to first be sorted by taxa in order for the standardization to be performed correctly. */

   PROC SORT DATA= arcsinz;

   BY taxa;

   RUN;

   PROC STANDARD DATA=arcsinz MEAN=0 STD=1 OUT=zscore;

   VAR ztransRA;

   BY taxa; /*see Note 2*/

   RUN;

6. /*Creating the nitrate-reducing taxa summary score “sumarcz” by including only the a priori identified taxa*/

   PROC SORT DATA= zscore;

   BY id;

   RUN;

   PROC MEANS data=zscore NOPRINT;

   WHERE taxa in (“Actinomyces_naeslundii”, “Actinomyces_odontolyticus”, “Actinomyces_viscosus”, “Capnocytophaga_sputigena”,

   “Corynebacterium_durum”, “Corynebacterium_matruchotii”, “Eikenella_corrodens”,

   “Haemophilus_parainfluenzae”,

   “Neisseria_flavescens”,

   “Neisseria_sicca”,

   “Neisseria_subflava”,

   “Prevotella_melaninogenica”,

   “Prevotella_salivae”,

   “Propionibacterium_acnes”,

   “Rothia_dentocariosa”,

   “Rothia_mucilaginosa”,

   “Selenomonas_noxia”,

   “Veillonella_dispar”,

   “Veillonella_parvula”,

   “Veillonella_atypica”); /*see Note 3*/

   VAR ztransRA; OUTPUT OUT=sumzscore (drop= _TYPE_ _FREQ_)

   sum(ztransRA)=sumarcz;

   BY id;

   RUN;

7. /*Merging the datasets to add the nitrate-reducing taxa summary score to the participant metadata */

   PROC SORT data=sumzscore; BY id; RUN;

   PROC SORT DATA= metadata; BY id; RUN;

   DATA final;

   MERGE sumzscore (IN=a) metadata (IN=b);

   IF a AND b;

   BY id;

   RUN;

8. /*Multivariable regression models regressing systolic blood pressure outcome (MeanSBP) on the exposure summary score of nitrate-reducing capacity (sumarcz) controlling for other covariates age, sex, ethnicity, body mass index (BMI), smoking status, etc. */

   PROC GENMOD DATA=final; /* see Note 4*/

   CLASS sex(ref=last) ethnicity(ref=“d Hispanic”) smoking(ref=“never”);

   MODEL MeanSBP=sumarcz age sex ethnicity bmi smoking /LINK=identity DIST=normal;

   RUN;

3.6. Code for running bacterial taxa summary score analyses in R

The following sections generate results identical to those in Section 3.5 which creates a bacterial taxa summary score as described in Section 3.1; only now we are using R software, and provide the R code to perform the analyses.

1. # Import the OTU/ASV output from 16S rRNA sequencing and sequencing analysis with taxa relative abundances (see Note 1).

   microbiome <- read.csv(“C:\Users\CG\16SOTUrelativeabundance.csv”)

2. # Import the participants metadata including Subject ID, socio-demographics, and clinical outcomes.

   metadata <- read.csv(“C:\Users\CG\Participantsclinicaldata.csv”)

3. # Conduct the arcsin-square root transformation on the relative abundance of each taxa, using the function created above

   microbiome$arsin <- asin(sqrt(microbiome$relativeabundance))

4. # Conduct taxa-specific z score standardization, creating a mean of 0 and standard deviation of 1 across all samples (see Note 5)

   microbiome$taxaZscore <- ave(x = microbiome$arsin, group = microbiome$taxa, FUN = scale)

5. # Create vector with the selected taxa, and subset to include only these taxa (see Note 3)

   taxaNames <- c (“Actinomyces_naeslundii”,

   “Actinomyces_odontolyticus”,

   “Actinomyces_viscosus”,

   “Capnocytophaga_sputigena”,

   “Corynebacterium_durum”,

   “Corynebacterium_matruchotii”,

   “Eikenella_corrodens”,

   “Haemophilus_parainfluenzae”,

   “Neisseria_flavescens”,

   “Neisseria_sicca”,

   “Neisseria_subflava”,

   “Prevotella_melaninogenica”,

   “Prevotella_salivae”,

   “Propionibacterium_acnes”,

   “Rothia_dentocariosa”,

   “Rothia_mucilaginosa”,

   “Selenomonas_noxia”,

   “Veillonella_dispar”,

   “Veillonella_parvula”,

   “Veillonella_atypica”)

   taxaSubset <- microbiome[microbiome$taxa %in% taxaNames, ]

6. # Add the z scores for each sample, and create new data set with appropriate column names

   sumZscore <- as.data.frame(aggregate(taxaSubset$taxaZscore, by = taxaSubset$ID, FUN = sum)

   names(sumZscore) <- c(“ID”, “sumarcz”)

7. # Merge dataset with the nitrate-reducing taxa summary score to the metadata

   final <- merge(metadata, sumZscore, by = “ID”)

   13. # Multivariable regression model, regressing systolic blood pressure outcome (MeanSBP) on the exposure nitrate-reducing taxa summary score (sumarcz) controlling for other covariates

   fit <- lm(MeanSBP ~ sumarcz + age + sex + ethnicity + bmi + smoking, data = final)

   summary(fit)

3.7. Code for running predicted gene abundance summary score analyses on SAS

In this section, we provide step-by-step instruction using SAS software to operationalize the PICRUSt2 output predicted gene abundance data, based on 16S rRNA data as discussed in Section 3.2 above. A nitrate-reducing gene abundance summary score is created and used as an exposure in multivariable linear regressions with a cardiometabolic outcome of interest-mean systolic blood pressure.

1. /* Import the output from PICRUSt2 analysis using KEGG Orthologs (KOs), conducted on 16S rRNA sequencing data. This data set has KOs in the rows and Subject IDs as columns. See Table 3) */

   PROC IMPORT DATAFILE= “C:\Users\CG\PICRUSt2OutputKO.csv” OUT=picrust2_wide DBMS=CSV REPLACE;

   RUN;

2. /* Transpose PICRUSt2 output using the SAS procedure PROC TRANSPOSE. The NAME=ID command is to create a new variable containing all the participants IDs previously listed as columns */ (see Note 6).

   PROC TRANSPOSE DATA=picrust2_wide OUT=picrust2_long NAME=ID;

   ID kegg_orthology;

   RUN;

3. /*Create a variable “TotalCounts” that records the total counts of all KOs per sample. Be sure to check the names of the first and last KO variables that appear in the data set (in this example, K00360 and K15876) */

   DATA picrust2v2;

   SET picrust2_long;

   TotalCounts= SUM (OF K00360--K15876); /*see Note 7*/

   RUN;

4. /*Create a subset of data keeping only Subject ID, TotalCounts, and KOs related to nitrate reduction genes*/

   DATA picrustNO3only;

   SET picrust2v2;

   KEEP ID TotalCounts

   K00367 K00370 K00371 K00374 K02567 K02568 K00372 K00360; /*see Note 8*/

   RUN;

5. /* Calculating relative abundances of the individual predicted genes by division with the “TotalCounts” of the sample*/

   DATA picrustNO3only;

   SET picrustNO3only;

   K00367_rel= K00367/TotalCounts;

   K00370_rel=K00370/TotalCounts;

   K00371_rel=K00371/TotalCounts;

   K00374_rel= K00374/TotalCounts;

   K02567_rel= K02567/TotalCounts;

   K02568_rel= K02568/TotalCounts;

   K00372_rel=K00372/TotalCounts;

   K00360_rel=K00360/TotalCounts;

   RUN;

6. /*Create a NO3 reduction relative gene abundance summary score, “NO3_rel”, by adding the predicted relative abundances of genes involved in nitrate reduction.*/

   DATA picrustNO3only;

   SET picrustNO3only;

   NO3_rel= K00367_rel + K00370_rel + K00371_rel + K00374_rel + K02567_rel + K02568_rel + K00372_rel + K00360_rel;

   RUN;

7. /* Arcsin-square root transformation on the NO3 reduction gene abundance summary score*/

   DATA picrustNO3only;

   SET picrustNO3only;

   NO3_arsin=arsin(sqrt(NO3_rel));

   RUN;

8. /*Import the participants metadata including Subject ID, socio-demographics, and clinical outcomes. Example of such a dataset is provided in Table 2 */

   PROC IMPORT DATAFILE= “C:\Users\CG\Participantsclinicaldata.csv” OUT=metadata DBMS=CSV REPLACE;

   GETNAMES=YES;

   RUN;

9. /*Merging the datasets to add the NO3 reduction gene abundance summary score to the participant metadata */

   PROC SORT DATA=picrustNO3only; BY id; RUN;

   PROC SORT DATA= metadata; BY id; RUN;

   DATA final;

   MERGE picrustNO3only (IN=A) metadata (IN=B);

   IF A AND B;

   BY id;

   RUN;

10. /*Multivariable regression model, regressing systolic blood pressure outcome (MeanSBP) on the exposure, NO3 reduction gene abundance summary score (NO3_arsin), controlling for other covariates */

    PROC GENMOD DATA=final;

    CLASS sex(ref=last) raceethn(ref=“d Hispanic”) cigcurr(ref=“never”);

    MODEL MeanSBP=NO3_arsin age sex raceethn bmi cigcurr /LINK=identity DIST=normal;

    RUN;

3.8. Code for running predicted gene abundance summary score analyses on R

The following section generate results identical to those in Section 3.7, using predicted gene abundance data output from PICRUSt2 to create a nitrate-reducing gene abundance summary score as described in Section 3.2; only now we are using R software, and provide the R code to perform the analyses.

1. # Import the output from PICRUSt2 analysis using KEGG Orthologs (KOs), conducted on 16S rRNA sequencing data. This data set has KOs in the rows and Subject IDs as columns.

   PICRUSt2_wide <- read.csv(“C:\Users\CG\PICRUSt2OutputKO.csv”)

2. # Transpose PICRUSt2 output.

   PICRUSt2_long <- gather(data = PICRUSt2_wide, key = ID, value = Count, 2:ncol(PICRUSt2_wide))

3. # Record total number of reads per sample as a new data set, and label columns appropriately.

   Total <- as.data.frame(aggregate(PICRUSt2_long$Count, by = list(PICRUSt2_long$ID), FUN = sum)

   names(Total) <- c(“ID”, “TotalCounts”)

4. # Create vector with the KOs of interest. Using the approach outlines above, we will select the KOs corresponding to enzymes involved in nitrate reduction to nitrite (see Note 8).

   KOs <- c(

   “K00367”, #Corresponds to the gene narB

   “K00370”, #Corresponds to the genes narG, narZ, and nxrA

   “K00371”, #Corresponds to the genes narH, narY, and nxrB

   “K00374”, #Corresponds to the genes narI and narV

   “K02567”, #Corresponds to the gene napA

   “K02568”, #Corresponds to the gene napB

   “K00372”, #Corresponds to the gene nasA

   “K00360”, #Corresponds to the gene nasB)

5. # Create a subset of data, including only the KOs selected above

   NO3 <- PICRUSt2_long[PICRUSt2_long$KEGG_orthology %in% KOs,]

6. # Convert NO₃ data set to long format.

   NO3 <- spread(NO3, key = KEGG_orthology, value = Count)

7. Add the total sample counts to the NO₃ data set.

   NO3 <- merge(NO3, Total, by = “ID”)

8. # Transform absolute abundances into relative abundances, by dividing each value by the sample total (see Note 9).

   NO3$K00367_rel <- NO3$K00367/NO3$Total

   NO3$K00370_rel <- NO3$K00370/NO3$Total

   NO3$K00371_rel <- NO3$K00371/NO3$Total

   NO3$K00374_rel <- NO3$K00374/NO3$Total

   NO3$K02567_rel <- NO3$K02567/NO3$Total

   NO3$K02568_rel <- NO3$K02568/NO3$Total

   NO3$K00372_rel <- NO2$K00372/NO3$Total

   NO3$K00360_rel <- NO2$K00360/NO3$Total

9. # Create a NO3 reduction relative gene abundance summary score “NO3_rel”, by adding the predicted relative abundances of genes involved in nitrate reduction

   NO3$NO3_rel <- NO3$K00367_rel + NO3$K00370_rel + NO3$K00371_rel + NO3$K00374_rel + NO3$K02567_rel + NO3$K02568_rel + NO3$K00372_rel + NO3$K00360_rel

10. # Conduct the arcsin-square root transformation on the NO3 reduction gene abundance summary score created.

    NO3$NO3_arsin <- asin(sqrt (NO3$NO3_rel))

11. # Import the participants metadata including Subject ID, socio-demographics, and clinical outcomes.

    metadata <- read.csv(“C:\Users\CG\Participantsclinicaldata.csv”)

12. # Merge data sets with inferred metagenome proportions to the metadata.

    final <- merge(metadata, NO2, by = “ID”)

13. # Multivariable regression model, regressing systolic blood pressure outcome (MeanSBP) on the exposure, NO3 reduction gene abundance summary score (NO3_arsin), controlling for other covariates.

    fit <- lm(meansbp ~ NO3_arsin + age + sex + raceethn + bmi + cigcurr, data = final)

    summary(fit)