Food desert residence has limited impact on veteran fecal microbiome composition: a U.S. Veteran Microbiome Project study

ABSTRACT Social and economic inequities can have a profound impact on human health, particularly on the development and progression of chronic disease. For military veterans, exposure to unique environments and circumstances may further impact their health. There continues to be limited work regarding the influence of mental health within the context of socioeconomic inequities. In this cross-sectional study, we hypothesized that veterans residing in food deserts (e.g., places in which there is a lack of access to sufficient and/or nutritious food) would have decreased gut microbial species (α-diversity), different microbiome community compositions, and poorer quality of diet and mental health compared to non-food desert residents. The fecal microbiome of 342 military veterans was sequenced, and microbiome diversity and community composition were evaluated. Although dietary quality and α-diversity did not significantly differ by food desert status, resident status (food desert versus non-food desert) accounted for a moderate influence on β-diversity (2.4%). Factors such as race and psychiatric diagnoses accounted for greater proportions of β-diversity influence (7% and 10%, respectively). Moreover, more participants with current post-traumatic stress disorder lived in food deserts (P < 0.04), and there were significantly more participants in the non-food desert group diagnosed with substance use disorders (P = 0.002) and current alcohol use disorder (P = 0.04). These findings suggest that living in a food desert, in combination with additional associated risk factors, may influence gut microbial diversity and composition. To increase ecological validity, researchers investigating the influence of inter-related biopsychosocial factors over time may benefit from adopting a life-course perspective. IMPORTANCE Social and economic inequities can have a profound impact on human health. The inequities could result in alterations to the gut microbiome, an important factor that may have profound abilities to alter health outcomes. Moreover, the strong correlations between social and economic inequities have been long understood. However, to date, limited research regarding the microbiome and mental health within the context of socioeconomic inequities exists. One particular inequity that may influence both mental health and the gut microbiome is living in a food desert. Persons living in food deserts may lack access to sufficient and/or nutritious food and often experience other inequities, such as increased exposure to air pollution and poor access to healthcare. Together, these factors may confer a unique risk for microbial perturbation. Indeed, external factors beyond a food desert might compound over time to have a lasting effect on an individual’s gut microbiome. Therefore, adoption of a life-course approach is expected to increase the ecological validity of research related to social inequities, the gut microbiome, and physical and mental health.

IMPORTANCE Social and economic inequities can have a profound impact on human health.The inequities could result in alterations to the gut microbiome, an impor tant factor that may have profound abilities to alter health outcomes.Moreover, the strong correlations between social and economic inequities have been long under stood.However, to date, limited research regarding the microbiome and mental health within the context of socioeconomic inequities exists.One particular inequity that may influence both mental health and the gut microbiome is living in a food desert.Persons living in food deserts may lack access to sufficient and/or nutritious food and often experience other inequities, such as increased exposure to air pollution and poor access to healthcare.Together, these factors may confer a unique risk for microbial perturbation.Indeed, external factors beyond a food desert might compound over time to have a lasting effect on an individual's gut microbiome.Therefore, adoption of a life-course approach is expected to increase the ecological validity of research related to social inequities, the gut microbiome, and physical and mental health.health than their non-food desert counterparts.We additionally hypothesized that veterans residing in food deserts would have an overall poorer quality of diet than their non-food desert counterparts.

Participants
The U.S. Veteran Microbiome Project (US-VMP) is an ongoing longitudinal study.Veteran participants in the greater Denver metropolitan area provide past history, complete clinical interviews and psychiatric surveys, and provide biological samples at their baseline visit (29).For this paper, baseline measures were used for participants who provided a home address and a gut microbiome sample.Participants completed clinically administered interviews and self-report survey measures (e.g., demographic characteristics, military history) at their baseline visit.

Surveys and assessments
Sample characteristics were assessed using the MIRECC Demographics questionnaire, which includes items regarding military service history and standard demographic information.Participants completed the Harvard Food Frequency Questionnaire (Harvard FFQ) 2007 booklet to assess dietary patterns (30).The Harvard FFQ is a validated, 101-item survey that quantifies specific food intake frequencies over the previous 12 months.Veterans also participated in a structured clinical interview, the DSM-5-TR (Diagnostic and Statistical Manual of Mental Disorders, 5th Edition, Text Revision) Axis I Disorders, Research Version, Patient Edition with Psychotic Screen (SCID-5-I/P W/PSY SCREEN), a validated method for assessing lifetime diagnoses of psychiatric disorders, and past-month symptoms of PTSD (31).Results were collapsed into broad catego ries of disorders (e.g., mood disorders, alcohol use disorders, substance use disorders, anxiety disorders, nonaffective psychotic disorders, and "other" DSM-5 diagnoses).From participants' medical records, we extracted data on veterans' receipt of benefits, termed "service connection, " related to injuries sustained or conditions developed while serving in the military.This service connection disability rating is represented as a percentage from 0 to 100 and is applied in increments of 10%.

Gut microbiome sample collection
Fecal samples were collected with double-tipped polyurethane swabs (BD BBL Cul tureSwab EZ II, Cat No. B220144, Fisher Scientific, Pittsburgh, PA, USA).Participants either collected fecal swabs during an inpatient evaluation at the research facility that were subsequently stored at −80°C or received a prepackaged sampling kit for home use with instructions for sample collection.Fecal swab samples collected at home were mailed back to the research facility, and upon receipt, all the samples were stored at −80°C.For a more in-depth description of sample collection, please refer to Brenner et al. (29).

Gut microbiome sample processing
Sample DNA was extracted from fecal samples using the PowerSoil DNA Extraction Kit (Cat.No. 12955-4, Qiagen, Valencia, CA, USA) (for further details, see Brenner et al. [29]).Marker genes in isolated DNA were PCR-amplified in duplicate using GoTaq Master Mix (Cat.No. M5133, Promega, Madison, WI, USA) as described in Caporaso et al. (32) using primers 515F-808R.Successful amplification was visualized on an agarose gel prior to pooling.Pooled amplicons were sent to the University of Colorado Anschutz Research Genetic Organization for normalization and sequencing on an Illumina MiSeq.

Food desert status
Participant residential addresses at the time they participated in the study were geocoded to census tracts via ArcGIS Pro (Version 2.6.0,Esri, Inc.) (33).The USDA Food Access Research Atlas (34) specifies food desert areas by census tract: "Low Access [LA] 1 and 10" or "tracts in which at least 500 people or 33% of the population lives farther than 1 mile (urban) or 10 miles (rural) from the nearest supermarket." Participants residing in census tract LA 1 and 10 locations were designated into the food desert group, whereas those who fell outside of these locations were designated into the non-food desert group.Participants who reported being unhoused at the time of data collection provided the address for the location at which they were staying that day.

Data analysis
For dietary quality, we applied the Healthy Eating Index (HEI) to Harvard FFQ data collected for this cohort (35).The HEI-2015 assesses adherence to the USDA's 2015-2020 U.S. Dietary Guidelines for Americans and is used to evaluate energy-adjusted intake of 13 different dietary groups (e.g., legumes, whole grains, added sugars) and propor tional intakes (e.g., density of total protein per 1,000 kcal).HEI analysis yields individual subscales for each variable and an aggregate score ranging from 0 to 100, with 0 being no adherence to dietary guidelines and 100 being perfect adherence.
The HEI was calculated in SAS 9.4 using a macro provided by the National Cancer Institute (36).We examined differences in total HEI and in each dietary subcategory by food desert residence using independent samples t-tests or Wilcoxon rank-sum tests, as appropriate.We also examined differences in demographic characteristics, mental health diagnoses, and service connection by food desert residence using independent samples t-tests, Wilcoxon rank-sum tests, χ tests, or Fisher's exact tests, as appropriate.All analyses assumed a two-sided test of hypothesis with a significance level of α = 0.05 and were conducted using SAS 9.4.
Sequencing data were processed using the Quantitative Insights into Microbial Ecology open-source software (QIIME2 v. 2021.8)(37).The Deblur algorithm (38) was used to denoise demultiplexed sequences.SATé-enabled phylogenetic placement (39) analysis was performed to improve the phylogenic tree used in β-diversity measures.Developer-recommended inputs included removal sequences that were not 75% similar to any record in the tree, but no amplicon sequencing variants in the present study met these criteria.Quality-filtered sequences were assigned taxonomic classification based on the silva_128 database (40).Samples were shipped to the facility by 237 participants (69.3% of the samples in the study) with instructions to freeze samples overnight and ship with the provided ice pack overnight.We observed that cold shipping was not consistently used, and/or there were mail issues that prevented samples from remaining cold.Therefore, we used a normalization of the community due to growth approximation and as previously described by Amir et al. (38).Downstream analysis and visualization were performed with the open-source statistical package R v.4.2.1 (41).
For α-and β-diversity and taxonomic evaluations, samples were rarefied to a level of 10,000 sequences per sample.All statistical analyses were adjusted for known covariates that can impact the microbiome, including age (grouped into decades), ethnicity, alcohol use, and gender (42).As the majority of participants who had an alcohol use disorder diagnosis also had a concurrent substance use disorder (78%), we did not specifically adjust for the latter.We also adjusted for housing history, specifically a previous history of homelessness, which we have observed is associated with microbial composition (43).
The α-diversity metrics for observed operational taxonomic units (OTUs), evenness, and Shannon diversity index were calculated on rarefied samples.Each of the α-diver sity metrics was skewed, so Tukey's transformation was conducted in the rcompanion package (44) prior to statistical modeling.Statistical analyses were conducted in R using the "glm" function for generalized linear modeling.β-Diversity was determined using the phyloseq package (45) in R for weighted and unweighted UniFrac distance metrics.β-Diversity was measured using the "adonis2" function of the vegan pack age (46) after 1,000 permutations.In order to determine the strongest correlates of fecal microbiota composition, we used a stepwise model-building tool for constrained ordination methods based on adjusted R 2 , using the UniFrac distance metrics with the "envfit" function of the vegan package.As the majority of participants were male (81.2%), α-and β-diversity measures were calculated with a subset of only males.The observed results altered P-values slightly but did not change significance to levels of under or over 0.05, so further stratification of the data was not pursued.
Differential abundance testing was analyzed using the Analysis of Compositions of Microbiomes with Bias Correction (ANCOM-BC) package with the ancombc2 function (47) (grouped by food desert, function of food desert plus five covariates, P adjustment with holm, and taxonomic prevalence >25%).A correlation matrix was developed between the top 20 genera and 25 metadata categories through the microbiomeSeq package (48) in R. Correlations were calculated using the Pearson correlation coefficient with Bejamini-Hochberg adjusted P-values for repeated measures of metadata categories and food desert status groups.The reporting of fecal microbiome data in this report is consistent with the Strengthening The Organization and Reporting of Microbiome Studies guidelines for human microbiome research (49).

Participant characteristics
Demographic characteristics are listed in Table 1.Our final cohort consisted of 342 U.S. veterans, 108 residing in food deserts (32%), and 234 in non-food deserts (68%).The majority of participants identified as male (81.2%) and Caucasian (67.4%), with a mean (± standard deviation) age of 47.9 y (±13.2).Ninety-eight percent of the participants resided in urban areas.Compared to those in the non-food desert group, veterans residing in food deserts were significantly younger (P < 0.03).Among participants who identified as being of Hispanic ethnicity, a greater proportion resided in food deserts relative to non-food deserts (P < 0.04).There was also a significantly higher percentage of students enrolled in post-secondary education among those residing in food deserts (P = 0.02).A significantly greater proportion of veterans living in non-food deserts reported having previously experienced homelessness (P < 0.0001) than veterans living in food deserts.In terms of military service history, food desert residents had experienced more deployments (P = 0.0006) and more deployments to combat zones (P < 0.0002).A significantly higher percentage of veterans living in food deserts had a Veterans Health Administration (VHA) service connection than their non-food desert counterparts (P < 0.0002).

Dietary outcomes
HEI scores were calculated for food and non-food desert groups and are presented in Table 2.There were no significant differences in dietary quality between groups (P = 0.24), with food desert groups having an average HEI score of 66.6 (±10.9) and non-food desert groups having an average HEI of 68.2 (±11.9).Within HEI subcategories, those residing in food deserts reported significantly higher intakes of total fatty acids (P < 0.001) than non-food desert participants, but no other statistically significant differences were found in any other HEI subcategories.

Mental health outcomes
Mental health results are presented in Table 3.There was a significantly higher percent age of participants with past-month PTSD symptoms residing in food deserts than non-food desert participants (P < 0.04), whereas there were significantly more partici pants in the non-food desert group diagnosed with substance use disorders (P = 0.002) and current alcohol use disorders (P = 0.04).No other significant differences were found in mental health outcomes by food desert status.

Gut microbiome results
α-Diversity did not differ by food desert status in measures of observed OTUs (Wilcoxon test, P = 0.96), Shannon diversity index (Wilcoxon test, P = 0.67), or Pielou's evenness (Wilcoxon test, P = 0.48) (Fig. S1).Generalized linear modeling with covariates lowered the P-values, but not to a significant level, for observed OTUs (glm, P = 0.933), Shannon diversity index (glm, P = 0.36), and Pielou's evenness (glm, P = 0.335).For β-diversity, microbial community composition also did not significantly differ for weighted UniFrac distance after adjustment for the five covariates (ADONIS, P = 0.651) (Fig. S2a).In contrast, unweighted UniFrac distance, which accounts for phylogenetic differences but not taxonomic relative abundance, was nearly significantly different after adjustment for covariates (ADONIS, P = 0.058) (Fig. S2b).To explore correlates of β-diversity further, canonical-correlation analysis (CCA) was performed for weighted and unweighted UniFrac distances.In the CCA using an unweighted UniFrac distance matrix, food desert status was significantly associated with phylogenetic differences, accounting for 2.4% of β-diversity among participants (R 2 = 0.024, P = 0.026) (Fig. 1B; Fig. S3).Using a One participant was missing information about sex.Five participants were missing information about their race.Five participants were missing information about ethnicity.
One participant was missing information about education.Three participants were missing information about marital status.Five participants were missing information about employment.One participant was missing information about student status.One participant was missing information about current homelessness.One participant was missing information about the most recent era served.Two participants were missing information about MST question 1.One participant was missing information about MST question 2.
b One participant was missing information for months served on active duty.Two participants were missing information for months of reserve service.c Question: while you were in the military, did you receive any uninvited and unwanted sexual attention, such as touching, cornering, pressure for sexual favors, or inappropriate verbal remarks?d Question: while you were in the military, did anyone ever use force or the threat of force to have sexual contact with you against your will? the unweighted UniFrac distance metric, categories of rurality, HEI total score, mood disorders, current homelessness, race, and service-connected disabilities all significantly differed between food desert and non-food desert participants.In CCA analysis of the weighted UniFrac distance matrix based on food desert status (Fig. 1A), β-diversity was most explained by statistically significant demographic variables (race [R 2 = 0.08, P = 0.02], education level [R 2 = 0.07, P = 0.03], sex [R 2 = 0.05, P = 0.002]), military-related variables (combat frequency [R 2 = 0.04, P = 0.02]), and mental health measures (mood disorder [R 2 = 0.05, P = 0.08], other psychiatric diagnosis [R 2 = 0.04, P = 0.03]).The relative abundances of the most prevalent phyla did not differ by food desert status.The most prevalent phyla (mean [x̅ ] relative abundance ± SD) were Firmicutes (49.3% ± 20.9%), Bacteroidetes (37.6% ± 22.5%), Actinobacteria (6.0% ± 8.0%), Proteo bacteria (4.8% ± 11.8%), and Verrucomicrobia (1.7% ± 5.0%) (Fig. S4a).At the genus level, the taxa with the highest relative abundances were Bacteroides (22.2% ± 18.9%),  Faecalibacterium (7.7% ± 8.2%), Blautia (6.2% ± 7.5%), and Prevotella (6.2% ± 11.0%), none of which differed by food desert status (Fig. S4b).No differentially abundant taxa were observed through ANCOM-BC2 analysis with multiple test corrections for genera and covariates.Prior to adjusting for covariates, we observed significant differential abundance for Barnesiella (x̅ = 0.370% food desert, x̅ = 0.162% non-food desert, P = 0.022), Campylobacter (x̅ = 0.064% food desert, x̅ = 0.185% non-food desert, P = 0.007), and Peptoniphilus (x̅ = 0.813% food desert, x̅ = 0.977% non-food desert, P = 0.048).Finally, a correlation matrix was developed for the 20 taxa with the highest relative abundances in comparison to 25 participant characteristics, separated by food desert status (Fig. 2).Six correlations were statistically significant and positively associated for non-food desert participants while negatively associated for food desert participants: (i) Escherichia-Shigella was associated with race and rurality; (ii) Ruminococcus torques was associated with current homelessness; (iii) Ruminococcus 2 and Blautia were each associated with a history of military sexual assault; and (iv) Ezakiella was associated with nonaffective psychiatric disorder.Lastly, age was negatively correlated with Faecali bacterium in non-food desert participants, and Ruminococcus UCG.002 was positively correlated with nonaffective psychiatric disorder in participants in food deserts.

DISCUSSION
In this study of food desert status, mental health, and the gut microbiome of military veterans, we explored the complexity inherent in evaluating such distinct yet interre lated factors.The most abundant genera in the present study were similar to those observed in the general population (50).A CCA metric indicated that living in a food desert accounted for 2.4% of β-diversity among veterans.While this finding is statisti cally significant, it is possible that there are other as-yet unexamined socioeconomic factors related to food that may account for a larger proportion of β-diversity.A factor worthy of future exploration is food insecurity.A financially driven condition of limited or inadequate access to sufficient nutrition (51), food insecurity is associated with an increased risk of adverse health outcomes, including cardiovascular disease, diabetes, and psychiatric illness (52)(53)(54).Of note, food insecurity differs from food desert status in that it comprises all of the financial and interpersonal resources that shape access to food, independently or in addition to geographic access.Food insecurity is highly prevalent among veterans in general (55), and the inclusion of measures of food insecurity, which can be more nuanced and granular than a binary food desert status, could yield more informative microbiome-related data.Of note, just recently, Zuniga-Chaves et al. (56) observed that food insecurity mediates the association between low socioeconomic status and decreased α-diversity.Nevertheless, it is also important to note that measuring food insecurity presents its own challenges, related to self-stigma in reporting, particularly among veteran populations (57,58).
In the context of dietary quality, previous food desert studies have yielded mixed results (59), with some studies reporting differences in nutritious food intake (e.g., fruits and vegetables) (60), and others suggesting that there are no significant differences in dietary habits between food desert and non-food desert residents (61,62).In our previous assessment of dietary habits among veterans in the US-VMP study (63), we did not observe marked differences in HEI scores.Veterans consumed a largely homogene ous Western-style diet, characterized by high intakes of nutrient-poor refined sugars and processed foods and poor intakes of more nutritious foods (63).Both food desert and non-food desert participants in our study scored higher on the HEI than the national HEI average, which was 57 as of 2018 (36), though the reasons for this are unclear.Total HEI scores did not differ by food desert status, but nevertheless, in the unweighted Unifrac CCA analysis, we observed that total HEI scores accounted for just under 8% of β-diversity among participants living in a food desert.While the reasons for this are also unclear, in addition to geographic proximity to nutritious food, an individual's subjective perception of how accessible healthy food options are may be an important consideration.The perception as to whether one lives in a food desert and/or a "food swamp" (64) area with a high density of unhealthy food options, which varies by race and ethnicity, has been associated with poorer self-reported dietary quality (65).Definitions of food swamps have yet to be formally operationalized, but such a metric may more accurately reflect participants' experiences with obtaining the kinds of foods they value to meet their nutritional needs than a measure of geographic proximity to grocery stores.
Race, education, rurality, and mood disorders were significantly positively correlated with residing in a food desert, as were an increased military service connection and current homelessness.Conversely, non-food desert status was significantly correlated with sex, other mental health disorders, and a greater frequency of service in combat.In terms of specific taxa, results were also mixed.For example, Ruminococcus 2 and Blautia, taxa that have been associated with anti-inflammatory functions (66,67), were both positively correlated with prior history of military sexual assault in non-food desert participants, but not among veterans with a history of sexual assault living in food deserts.Similarly, Faecalibacterium, a beneficial genus of bacteria with probi otic properties, was negatively correlated with age.This finding is supported by prior research (68), but in our cohort, there was no such correlation among veterans living in food deserts.
There are several possible explanations for some of our mixed, counterintuitive findings.Some of our observations may have been happenstance, as in the case of the positive correlation between Escherichia-Shigella, a pathogen, and rurality among non-food desert participants, when only 2% of all participants resided in a rural area.It is also possible that overall socioeconomic status or income are more strongly associated with microbial composition and that factors like food desert status, age, and current student status are simply proxies for participants' relative burden of social inequity (69,70).For example, college students, who were significantly more likely to reside in a food desert in this cohort, are inherently at a stage of life with diminished earning potential and at a financial disadvantage compared to their older counterparts (71).Participants with a prior history of experiencing homelessness were also more likely to reside in a food desert.There is evidence to suggest that a history of homelessness is correlated with reduced long-term earning potential (72,73), once again suggesting that food desert status may have been a proxy for income.
Other demographic factors such as race and ethnicity were associated with food desert status, and race in particular accounted for 7% to 10% of β-diversity, depending on the β-diversity metric.While there is evidence for racial differences in gut micro bial compositions and related health outcomes (74,75), there is also strong evidence that these variations may be due to socioeconomic inequities rather than biological differences (27,76).
We also observed that, as in previous research (77), mental health is significantly associated with microbial composition.While we observed differences in correlations by food desert status (mood disorders among food desert residents, other psychiatric diagnoses among non-food desert residents), we cannot determine that the disorder in question is truly driving the association between microbial composition and food desert status.Psychiatric conditions, whether a mood disorder, a schizoaffective condition, or PTSD, can all exert a profound influence on how people experience social inequities (78), and differentiating by food desert status did not allow for parsing out how precisely these conditions manifest or how they shape veterans' behavior as they experience their environments.
Finally, in taking a broader view of our findings, we suggest that military veterans' unique exposures and characteristics constitute a population that still merits more detailed investigation in microbiome research.In this cohort, greater frequency of service in combat, increased military service connection, PTSD symptoms, and a history of military sexual assault were all to some degree associated with β-diversity among veterans residing in a food desert.The body of literature on gut microbiota in military veterans remains too small to draw firm conclusions regarding factors that may be driving these correlations.Nevertheless, our findings suggest that while veterans have much in common with the general population in terms of microbial associations, general demographics, and common mental health disorders, their service-related exposures may exert a unique influence on their subsequent socioeconomic and environmental characteristics, which may in turn shape microbial composition.
Therefore, we propose that future microbiome studies expand on our findings not only by examining more veteran populations but also by evaluating populations that have had exposures comparable to those of veterans, namely, regimented-style living combined with profound physical and psychological stressors.For example, studies of people in communal living settings, such as barrack-style emergency housing, residential rehabilitation programs for youth, and cloistered religious communities, and who have experienced stressful events, such as political unrest or violence, crime, war, or natural disasters, may provide insights into how such exposures shape subsequent mental health-related and socioeconomic correlations with the microbiome.
It is also important to note that only a small percentage of the United States population serves in the military, with socioeconomic factors playing an important role in terms of individuals enlisting.To increase ecological validity, microbiome studies investigating the influence of factors over time may benefit from the use of an approach that Dowd and Renson (79) termed "social epidemiology of the microbiome." Dowd and Renson's approach not only encompasses all of the environmental and behavioral exposures that can shape microbiota at the time of data sampling but also evaluates the exposures and stressors across a person's life course, all in a holistic socioeconomic, psychological, and cultural context.In terms of veterans, this would allow for increased understanding regarding whether or not pre (e.g., adverse childhood experiences), military (e.g., combat), and post-military (e.g., homelessness) exposures may influence microbial composition over time.

Limitations
This study has a number of limitations.The binary categorization of food desert status may have significant limitations, with recent studies moving toward more nuanced descriptions of areas of food landscapes, including machine learning-based identification of food desert status (80).Of note, food swamps have been shown to be a better predictor of adverse health outcomes (e.g., obesity) than food deserts (81), but current USDA map data, which were our primary source, do not provide more than a binary differentiation.Moreover, whether one resides in a food desert or not does not necessarily indicate one's actual food choices, and many residents may not be aware that they reside in a food desert.Next, 26 participants reported being unhoused (no permanent address) at the time of data collection, and while they provided the addresses of the locations at which they were residing on the given day, this may have been an inaccurate reflection of their long-term food desert status.Although we collected data on service connections, we were not able to collect income data on the participants in this cohort, limiting our ability to parse out specific financial influences on food desert associations.Additional measures, such as income, income sources (e.g., employment, Social Security), transportation access, subjective food security, and purchasing locations for food (i.e., near home versus near work), are all likely influential factors whose absence in this study may limit the validity of our findings.It is also worth noting that food frequency questionnaires and the HEI are limited in their ability to accurately assess dietary patterns, and more accurate measures, such as multiple 24-h recalls (82), should be used in future studies, though dietary patterns among our participants were overwhelmingly homogeneous.We were not able to assess specieslevel associations, and the genus-and phylum-level correlations did not allow us to identify if food desert status was associated with a truly "beneficial" taxon.Our sampling method, like other large-scale fecal studies, directed individuals to ship the samples after freezing, a request that was not always adhered to in practice.Lastly, as data from only one time point were analyzed, if and how associations between food desert status, related covariates, and the gut microbiome may fluctuate over time could not be evaluated.

Conclusion
Overall, our findings suggest that food desert status had a modest association with gut microbial composition, though other variables, such as race, education, and mood disorders, were more robustly associated with this outcome of interest.In addition, veterans residing in a food desert differed from their non-food desert counterparts in demographic characteristics, combat history while in the military, and mental health diagnoses, but not in dietary quality.Moreover, while food desert status and related covariates accounted for some variance in β-diversity, it is likely that our use of a binary food desert metric was insufficiently robust to account for the complex environmental, psychosocial, and economic forces that shape veterans' health and functioning.We propose, therefore, that future research should use a social epidemiological approach to characterizing the life course of factors that impact the gut microbiome in this unique population.

FIG 1 FIG 2
FIG 1 CCA R 2 estimates for β-diversity of the gut microbiome analyzed using (A) weighted and (B) unweighted UniFrac distance metrics.

TABLE 1
Characteristics of the US-VMP cohort, overall and by food desert status (Continued on next page)

TABLE 1
Characteristics of the US-VMP cohort, overall and by food desert status (Continued)

TABLE 2
Total HEI and component indices, overall and by food desert status

TABLE 3
Prevalence of mental health disorders, overall and by food desert status

DSM-5 disorders identified by SCID-5 a Overall Not living in food desert Living in food desert P-value N = 342 N = 234 N = 108 Statistical test
a Six participants were missing information about mood disorders.Two participants were missing information about nonaffective psychotic disorders.Four participants were missing information about alcohol use.Five participants were missing information about substance use.Five participants were missing information about anxiety disorders.Six participants were missing information about PTSD.Five participants were missing information about other stress disorders.Eight participants were missing information about other DSM-5 disorders.b Other lifetime disorders include the following: obsessive-compulsive disorder, hoarding disorder, body dysmorphic disorder, anorexia nervosa, bulimia nervosa, binge eating disorder, avoidant/restrictive food intake disorder, other eating disorders, and other DSM-5 disorders.c Other current disorders include the following: adult attention-deficit/hyperactivity disorder, intermittent explosive disorder, other specified trauma-and stressor-related disorders, anorexia nervosa, bulimia nervosa, binge eating disorder, avoidant/restrictive food intake disorder, other eating disorders, acute stress disorder, hoarding disorder, gambling disorder, insomnia disorder, hypersomnolence disorder, other sleep disorders, body dysmorphic disorder, somatic symptom disorder, illness anxiety disorder, and other DSM-5 disorders.