Food Insecurity and Undernutrition are Associated with Distinct Immunologic Profiles in People with Tuberculosis and Advanced HIV Starting Antiretroviral Therapy

Abstract

Background

Food insecurity and undernutrition are related but distinct concepts contributing to poor HIV and tuberculosis outcomes. Pathways linking them with immunologic profile, which may relate to clinical outcomes, remain understudied.

Methods

We analyzed data from a cohort study of 165 ART-naïve adults with advanced HIV and newly diagnosed tuberculosis in Botswana from 2009–2013. 29 plasma biomarkers were measured pre-ART and four weeks post-ART initiation. We used principal components analysis (PCA) and multivariable linear regression models to assess relationships between immunological profiles and food insecurity (based on the Household Food Insecurity Access Scale), undernutrition (Body Mass Index < 18.5 kg/m²), and clinical outcomes.

Results

PCA identified 5 principal components (PCs) with eigenvalues >1. After adjustment, food insecurity was associated with PC3 pre-ART (0.19 per increased category of severity, 95% CI 0.02 to 0.36) and post-ART (0.24, 95% CI 0.07 to 0.41). PC3 was driven by higher levels of IFN-α, IFN-γ, IL-12p40, VEGF, IL-1α, and IL-8, and decreased concentrations of IL-3. Undernutrition was associated with PC5 post-ART (0.49, 95% CI 0.16 to 0.82). PC5 was driven by higher levels of IL-8, MIP-1α, IL-6, and IL-10, and decreased concentrations in IP-10 and IFN-α. Post-ART PC3 (4.3 percentage point increased risk per increased score of 1, 95% CI 0.3 to 8.9) and post-ART PC5 (4.8, 95% CI 0.6 to 8.9) were associated with death in adjusted models.

Discussion

We identified two distinct immunologic profiles associated with food insecurity, undernutrition, and clinical outcomes in patients with advanced HIV and TB. Different pathophysiologic processes may link food insecurity and undernutrition with poor outcomes in this vulnerable patient population. Future studies should assess the impact of improving food access and intake on immune function and clinical outcomes.

Background

Food insecurity and undernutrition are distinct concepts (though related and often conflated)¹ that have each been implicated as contributors to poor outcomes in the context of HIV and tuberculosis (TB).^2–7 Food insecurity is a lack of stable access to food in adequate quantity or quality, is characterized by worry about food and (when more severe) the experience of hunger, and can exist across multiple dimensions — access, availability, utilization, and stability.⁸ In contrast, undernutrition is a physiologic condition characterized by energy or nutrient deficiencies, and in adults is generally defined by an abnormally low body mass index (BMI).⁹ Undernutrition can be a direct result of food insecurity (in the setting of decreased food intake resulting in a combination of micronutrient, macronutrient, and energy deficiencies), but can also be caused by anorexia induced by inflammatory cytokines, increased catabolism (common in advanced HIV and TB), diarrhea and malabsorption (due to enteropathogens or HIV enteropathy), altered utilization or nutrients (e.g., anabolic block), or excess alcohol consumption. Food insecurity, on the other hand, does not always lead to undernutrition (e.g., a household with irregular food access may experience food insecurity without developing nutritional deficiencies) and can affect health outcomes through other mechanisms like health behaviors and mental health.¹⁰

Food insecurity or undernutrition may be linked with poor HIV and TB outcomes due to immunologic changes, but these pathways remain understudied. Analyses have focused on the general population,^11–14 animal models,⁶ latent TB infection,^15–18 or HIV on long-term antiretroviral therapy (ART) in high-income countries.^19–21 These studies generally suggest impairments in immune responses are associated with undernutrition, and increased immune activation and inflammation in the context of food insecurity and HIV on ART.

However, the degree to which these relationships between food insecurity or undernutrition and immunologic dysfunction are distinct from each other remains an open question. Immunologic profiles of food insecurity and undernutrition among people co-infected with HIV and TB have also not been evaluated, including around the time of ART initiation, a period associated with high levels of morbidity and mortality and specific immunologic changes already known to be related to clinical outcomes, particularly in the context of advanced disease.^22,23 To address these gaps in the literature, we assessed the distinct relationships between food insecurity and undernutrition with immunologic profiles among people living with advanced HIV and pulmonary TB in a high-burden setting both before and soon after ART initiation.

Methods

Study Objectives

The primary objective of this analysis was to identify and characterize immunologic profiles (primary outcomes) associated with food insecurity or undernutrition (primary exposures) among people with advanced HIV and recently diagnosed pulmonary tuberculosis, both before and after ART initiation. We developed a conceptual model illustrating known and hypothesized pathways between food insecurity, undernutrition, and HIV/TB outcomes, and highlighting the central research question of this analysis (Figure 1). Secondary objectives included assessing whether immunologic profiles associated with food insecurity or undernutrition were also associated with clinical outcomes (death or TB-immune reconstitution inflammatory syndrome [TB-IRIS]), evaluating whether relationships between food insecurity or undernutrition and immunological profiles were influenced by the inclusion of an interaction term with the other variable attenuated after additionally controlling for each other (i.e., controlling food insecurity estimates for undernutrition, and vice versa) or for change in HIV viral load during the first four weeks of ART, or after exclusion of overweight/obese participants.

Figure 1.

Conceptual model showing pathways between food insecurity, undernutrition, and HIV/TB outcomes, and highlighting the central research question of this analysis related to immunologic profiles associated with food insecurity and undernutrition. Dotted lines represent potential consequences of HIV/TB that may feed into vicious cycles with food insecurity or undernutrition.

Participants

We analyzed data from a prospective cohort study that enrolled participants consecutively between 2009 and 2013 at 22 public clinics and Princess Marina Hospital in Gaborone, Botswana.^22,23 Botswana has a high burden of HIV (prevalence 19%),²⁴ TB (annual incidence 238 per 100,000, 48% with HIV co-infection),²⁵ food insecurity (prevalence 53%),²⁶ and undernourishment (prevalence 22%).²⁷ Eligibility criteria of the original cohort study included: (1) age ≥21 years, (2) living with HIV, (3) ART-naïve (except for prior prevention of mother-to-child transmission), (4) CD4 T-cell count ≤ 125 cells/mm³, and (5) new diagnosis of pulmonary TB. TB was diagnosed through one of the following: (1) ≥1 positive acid-fast bacilli sputum smear, (2) meeting the World Health Organization criteria for smear-negative pulmonary TB, or (3) Xpert MTB/RIF assay (Cepheid Inc) positive for Mycobacterium tuberculosis. Exclusion criteria included: (1) drug-resistant TB, (2) prior TB treatment within the last 12 months, and (3) receipt of corticosteroids or other immunomodulatory therapy since diagnosis of TB. The study was approved by the institutional review boards of the University of Pennsylvania, the Botswana Ministry of Health, and the Princess Marina Hospital. All participants provided written informed consent.

Data

Participants were assessed at baseline and four weeks after initiating ART, with details previously described.^22,23 Demographic, clinical, and laboratory data included in this analysis were age, sex, alcohol use during the last year (assessed at baseline through the Alcohol Use Disorders Identification Test [AUDIT]-C),²⁸ presence of non-TB opportunistic infections (baseline), presence of extrapulmonary TB (baseline), ART regimen, height (baseline), weight (baseline), CD4 T-cell count (baseline and four weeks post-ART), and HIV viral load (baseline and four weeks post-ART).

Baseline food security was assessed using the Household Food Insecurity Access Scale (HFIAS).²⁹ The HFIAS is an experiential scale that covers a recall period of 30 days and includes 18 items — nine occurrence questions and nine frequency questions. Participants were categorized into one of four categories: food secure, mildly food insecure, moderately food insecure, or severely food insecure.²⁹ Baseline nutrition status was defined based on body mass index (BMI) in the following categories: severe undernutrition (<16 kg/m²), moderate undernutrition (16–16.99 kg/m²), mild undernutrition (17–18.49 kg/m²), normal (18.5–24.99 kg/m²), and overweight or obese (≥25 kg/m²).^5,30

We used a magnetic bead Luminex panel (EMD Millipore, Billerica, Massachusetts) to assess major immunologic biomarkers by measuring plasma levels of the following cytokines, chemokines, and growth factors at enrollment (i.e., just prior to initiating ART) and four weeks after ART initiation (post-ART): epidermal growth factor (EGF), vascular endothelial growth factor (VEGF), granulocyte-colony stimulating factor (G-CSF), granulocyte macrophage colony-stimulating factor (GM-CSF), interferon (IFN)-α, IFN-γ, interleukin(IL)-1RA, IL-1α, IL-1β, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL17a, IFN-γ-induced protein(IP)-10, monocyte chemoattractant protein (MCP)-1, macrophage inflammatory protein (MIP)-1α, MIP-1β, eotaxin, and tumor necrosis factor (TNF)-α, and TNF-β. IL-4, IL-13, and TNF-β were below the limit of detection and excluded. Undiluted plasma was stored at −80°C and subsequently tested in duplicate on the Bio-Plex2000 Luminex platform (Bio-Rad, Hercules, California), with no samples exposed to thawing and re-freezing prior to assays being run. Data were analyzed using a 5-point log-log standard curve with Bio-Plex Manager software (Bio-Rad, Hercules, California).

This analysis was restricted to participants who started ART and had available biomarker and food security data. The outcomes of death and TB-IRIS were defined and assessed as previously described.^22,23

Statistical analysis

We summarized participant characteristics by food security status and undernutrition, assessing for differences in categorical variables using the Chi-squared test and continuous variables using the Wilcoxon rank-sum test (undernutrition) and the Kruskal-Wallis test (food insecurity).

We identified immunological profiles of pre-ART biomarkers among participants using principal component analysis (PCA).^31,32 PCA is a method for reducing the dimensionality of large, correlated datasets by creating Principal Components (PCs), which are linear combinations of the variables contained in the analysis. Each principal component accounts for as much of the overall variance as possible under the condition that it is not correlated with any prior component. For our PCA, we included all pre-ART immunologic biomarkers after log transformation and standardization. We retained all PCs with eigenvalues greater than 1.0, meaning that they should have greater predictive power than any of the original measured variables. We used PC loadings to generate PC scores for all participants at baseline (pre-ART) and after four weeks of ART (post-ART).

We used multivariable linear regression models to assess the relationships between pre- and post-ART PC scores and (1) food insecurity and (2) undernutrition. Because we hypothesized that immunologic changes associated with food insecurity would exist on a gradient, we analyzed food insecurity as a continuous variable from 0 (no food insecurity) to 3 (severe food insecurity), though we also conducted a secondary analysis with food insecurity as an ordinal variable. We analyzed undernutrition as a binary variable because of sample size limitations. We built linear regression models with pre- and post-ART PCs as the outcome variables and food insecurity or undernutrition as the independent variables. We then generated adjusted estimates after adding known or hypothesized confounders: age, sex, extrapulmonary TB, non-TB opportunistic infection, baseline CD4 count, baseline HIV log viral load, and number of days between initiating anti-TB therapy and ART (post-ART estimates only). Variance inflation factors (VIF) for all covariates were <1.5, indicating little to no multicollinearity.

For PCs that were significantly associated with food insecurity or undernutrition (either pre- or post-ART), we then assessed the likely contribution of individual biomarkers to these associations. To do this, we examined effect measures from similarly constructed multivariable linear regression models as above, except with log-transformed individual biomarkers substantially contributing to the relevant PCs (as indicated by the greatest positive or negative loadings) as the outcome variables. Because the biomarker variables were log-transformed, we exponentiated the regression coefficients and interpreted them as relative differences in the biomarkers compared to the reference group (food secure or no undernutrition).

We next turned to the two clinical outcomes measured in the study: death and TB-IRIS. We used multivariable linear regression models to evaluate the relationship between the PC scores associated with food insecurity or undernutrition and each clinical outcome, controlling for other variables potentially associated with death or TB-IRIS: age, sex, extrapulmonary TB, non-TB opportunistic infection, number of days between initiating anti-TB therapy and ART, baseline CD4 count, baseline HIV log viral load. Effect estimates for these binary outcomes were interpreted as percentage point changes in risk.³³

We conducted several other secondary analyses. We assessed for changes in effect sizes after the inclusion of undernutrition in the food insecurity models (because undernutrition may mediate the association), and vice versa. Similarly, for immunologic profiles associated with food insecurity or undernutrition, we added an interaction term between food insecurity and undernutrition. For the post-ART PC models, we also assessed for changes in effect sizes after inclusion of the change in HIV log viral load to assess for the contribution of ART-mediated viral response. Because obesity has been linked with inflammatory changes,³⁴ we conducted a sensitivity analysis assessing relationships between undernutrition and PC scores after excluding participants with BMI ≥25. Our analysis was underpowered to detect moderate or small associations between food insecurity or undernutrition and clinical outcomes — e.g., we would have only 28% power to detect a 50% reduction in mortality associated with undernutrition when assuming N=165 (N=67 with undernutrition), α=0.05, and 12% mortality among participants with undernutrition. Therefore, we calculated adjusted effect estimates and 95% confidence intervals for the associations between food insecurity and undernutrition with death and TB-IRIS using all available data in an exploratory fashion to generate a range of potential associations.

We defined significance as two-tailed p<0.05. We performed statistical analysis using SAS version 9.4 (SAS Institute, Cary, NC, USA) and R version 4.2.3 (R Foundation for Statistical Computing).

Results

Participant characteristics

There were 170 participants enrolled from November 2009 to July 2013 who started ART, with baseline characteristics of the overall cohort previously reported²³ — 165 (97%) of these had available biomarker and food insecurity data and were therefore included in this analysis. Of note, participants reported very little alcohol use over the prior year, with only two participants (1%) having a positive AUDIT-C screen. Seventeen (10%) of the included participants died during follow-up, with circumstances surrounding the deaths previously reported,²³ and 32 (19%) developed TB-IRIS.

There was a broad distribution of food insecurity severity at baseline, with 46 (28%) participants being food secure, 62 (38%) experiencing mild food insecurity, 35 (21%) with moderate food insecurity, and 22 (13%) with severe food insecurity. There were no significant differences in participant characteristics by food security status (Table 1). Most participants had either a normal (N=78, 47%) or overweight/obese (N=20, 12%) nutritional status, with the remainder having mild (N=20, 12%), moderate (N=14, 8%), or severe (N=33, 20%) undernutrition. There were no significant differences in the characteristics of participants with undernutrition compared to participants without undernutrition (Table 2). There were 49 (30%) participants with both undernutrition and food insecurity. Distributions of non-TB opportunistic infections were similar by food security and nutrition status (Supplementary Table 1)

Table 1. Participant Characteristics at Baseline (Pre-ART) by Food Security Status.

Data presented as N (%) unless otherwise specified.

	Food Secure	Mild Food Insecurity	Moderate Food Insecurity	Severe Food Insecurity	p-value
	N=46	N=62	N=35	N=22
Female	24 (52)	23 (37)	18 (51)	8 (36)	0.29
Age, median (IQR)	34 (29–40)	35 (30–39)	36 (31–40)	42 (33–44)	0.18
Time between anti-TB therapy and ART initiation, days, median (IQR)	26 (17–42)	27 (18–48)	29 (21–45)	29 (22–48)	0.60
Extrapulmonary TB	5 (11)	3 (5)	1 (3)	0 (0)	0.22
non-TB Opportunistic Infection	4 (9)	8 (13)	6 (17)	2 (9)	0.67
CD4 T-cell count, cells/mm3, median (IQR)	56 (27–84)	67 (38–100)	48 (26–95)	61 (17–88)	0.38
HIV viral load, log10 copies/mL, median (IQR)	5.4 (4.9–6.0)	5.6 (5.0–5.9)	5.6 (5.2–6.0)	5.5 (5.1–6.1)	0.70
BMI, median (IQR)	18.8 (17.5–22.9)	19.6 (16.4–22.1)	19.1 (16.4–22.0)	19.2 (15.9–21.2)	0.63
Nutrition					0.51
Severe undernutrition	6 (13)	12 (19)	8 (23)	7 (32)
Moderate undernutrition	4 (9)	5 (8)	3 (9)	2 (9)
Mild undernutrition	8 (17)	8 (13)	3 (9)	1 (5)
Normal	18 (39)	32 (52)	18 (51)	10 (45)
Obese	10 (22)	5 (8)	3 (9)	2 (9)
Antiretroviral Therapy Initiated					0.62
EFV/FTC/TDF	38 (83)	53 (85)	30 (86)	19 (86)
NVP/FTC/TDF	7 (15)	8 (13)	4 (11)	3 (14)
NVP/FTC/ZDV	1 (2)	1 (2)	0 (0)	0 (0)
EFV/3TC/ABC	0 (0)	0 (0)	1 (3)	0 (0)

Table 2. Participant Characteristics at Baseline (Pre-ART) by Undernutrition Status.

Data presented as N (%) unless otherwise specified.

	No Undernutrition	Undernutrition	p-value
	N=98	N=67
Female	44 (45)	29 (43)	0.84
Age, median (IQR)	36 (31–42)	35 (29–40)	0.34
Time between anti-TB therapy and ART initiation, days, median (IQR)	29 (19–50)	26 (20–35)	0.13
Extrapulmonary TB	5 (5)	4 (6)	0.81
non-TB Opportunistic Infection	13 (13)	7 (10)	0.59
CD4 T-cell count, cells/mm3, median (IQR)	60 (29–95)	62 (32–86)	0.73
HIV viral load, log10 copies/mL, median (IQR)	5.6 (4.9–6.0)	5.5 (5.2–5.9)	0.96
BMI, median (IQR)	21.4 (19.8–23.8)	16 (14.5–17.7)	<.0001
Food insecurity			0.97
Severe	12 (12)	10 (15)
Moderate	21 (21)	14 (21)
Mild	37 (38)	25 (37)
None	28 (29)	18 (27)
Antiretroviral Therapy Initiated			0.46
EFV/FTC/TDF	80 (82)	60 (90)
NVP/FTC/TDF	17 (17)	5 (7)
NVP/FTC/ZDV	1 (1)	1 (1)
EFV/3TC/ABC	0 (0)	1 (1)

Identification of immunologic profiles through principal components analysis

We used PCA to identify co-correlated biomarkers that accounted for variance across the dataset of pre-ART biomarkers. Pre-ART and post-ART biomarker PC scores were available for 165 and 156 participants, respectively (7 died prior to four weeks of ART and 2 did not have post-ART biomarker measurements). There were 5 PCs with eigenvalues greater than one which collectively accounted for 71% of the variance — PC1 (46%), PC2 (10%), PC3 (5%), PC4 (5%), and PC5 (5%) (Figure 2A–C). The component loadings for each PC show the biomarkers contributing to each PC (Figure 2D).

Figure 2. Principal Component Analysis Scores and Weightings.

(A) Scree plot showing eigenvalues of the principal components, with a horizontal line at an eingenvalue of 1 (B) Cumulative proportion of variance explained by the principal components (C) Variance of each of the first five principal components amongst all participants (D) Heat map showing the loadings of each baseline biomarker for the first five principal components.

Associations between food insecurity, undernutrition, and immunologic profiles

We next proceeded to our primary analyses — assessing the relationships between food insecurity and undernutrition with immunologic profiles (Table 3). In both unadjusted and adjusted analyses, food insecurity was significantly associated with PC3 both pre-ART (adjusted difference 0.19 per increased category of food insecurity severity, 95% CI 0.02 to 0.36) and post-ART (adjusted difference 0.24 per increased category of food insecurity severity, 95% CI 0.07 to 0.41). PC loadings and effect estimates suggested that the association between food insecurity and PC3 may have been driven by a combination of higher levels of the pro-inflammatory cytokine IFN-α (and to a more modest extent IL-1α), higher concentrations of the Th1 cytokines IFN-γ and IL-12p40, more modestly by higher levels of the chemokine IL-8 and the growth factor VEGF, and post-ART by decreases in IL-3 (Figure 3, Supplementary Figure 1). Among baseline characteristics, HIV viral load was positively associated with PC3, both pre-ART (0.51 difference per log₁₀[copies/mL]; 95% CI 0.26 to 0.76) and post-ART (0.49 difference per log₁₀[copies/mL]; 95% CI 0.25 to 0.74).

Table 3. Regression analyses assessing associations between principal components and (1) food insecurity and (2) undernutrition, pre- and post-ART initiation.

Adjusted estimates control for age, sex, extrapulmonary TB, non-TB opportunistic infection, number of days between initiating anti-TB therapy and ART (post-ART only), baseline CD4 count, and baseline HIV log viral load.

	Unadjusted		Adjusted
	Difference	95% confidence interval	Difference	95% confidence interval
	Food Insecurity, Pre-ART
PC1	0.21	−0.31 to 0.73	0.25	−0.29 to 0.80
PC2	−0.07	−0.32 to 0.19	−0.06	−0.34 to 0.22
PC3	0.20	0.02 to 0.37	0.19	0.02 to 0.36
PC4	−0.10	−0.27 to 0.07	−0.06	−0.24 to 0.12
PC5	−0.08	−0.25 to 0.08	−0.12	−0.30 to 0.06
	Food Insecurity, Post-ART
PC1	0.27	−0.27 to 0.80	0.35	−0.19 to 0.89
PC2	0.15	−0.08 to 0.39	0.13	−0.07 to 0.43
PC3	0.22	0.05 to 0.39	0.24	0.07 to 0.41
PC4	0.00	−0.18 to 0.18	−0.02	−0.21 to 0.17
PC5	0.00	−0.16 to 0.16	−0.04	−0.21 to 0.13
	Undernutrition, Pre-ART
PC1	−0.08	−1.14 to 0.98	−0.16	−1.25 to 0.93
PC2	0.00	−0.51 to 0.50	0.10	−0.45 to 0.64
PC3	−0.02	−0.38 to 0.34	−0.07	−0.41 to 0.27
PC4	0.02	−0.32 to 0.36	0.08	−0.27 to 0.43
PC5	0.36	0.03 to 0.68	0.33	−0.02 to 0.68
	Undernutrition, Post-ART
PC1	0.15	−0.94 to 1.23	−0.11	−1.19 to 0.96
PC2	0.00	−0.48 to 0.48	−0.01	−0.51 to 0.49
PC3	0.01	−0.34 to 0.36	−0.02	−0.36 to 0.33
PC4	0.13	−0.24 to 0.49	0.18	−0.21 to 0.56
PC5	0.55	0.23 to 0.86	0.49	0.16 to 0.82

Figure 3. Post-ART associations between food insecurity and undernutrition with individual log-transformed biomarkers substantially contributing to principal component 3 (for food insecurity) and principal component 5 (for undernutrition).

Reference groups are the food secure participants, and participants without undernutrition.All estimates use adjusted for age, sex, extrapulmonary TB, non-TB opportunistic infection, number of days between initiating anti-TB therapy and ART, baseline CD4 count, and baseline HIV log viral load.

Undernutrition was significantly associated with PC5 post-ART (adjusted difference 0.48, 95% CI 0.15 to 0.81), but only with pre-ART PC5 prior to adjustment (unadjusted difference 0.36, 95% CI 0.03 to 0.68; adjusted difference 0.33, 95% CI −0.02 to 0.68) (Table 3). In contrast to food insecurity, analyses of individual biomarkers suggested that the post-ART association between undernutrition and post-ART PC5 may have been driven principally by increases in the chemokines IL-8 and MIP-1α, the pro-inflammatory cytokine IL-6, and to a lesser extent decreases in the chemokine IP-10, the pro-inflammatory cytokine IFN-α, and increases in the anti-inflammatory cytokine IL-10 (Figure 3). Among baseline characteristics, CD4 count was inversely associated with post-ART PC5 (−0.005 difference per cell/mm³; 95% CI −0.01 to −0.0005).

Associations between immunological profiles and clinical outcomes

We next evaluated associations between immunologic profiles associated with food insecurity or undernutrition and clinical outcomes (Table 4). After adjustment, both post-ART PC3 (adjusted 4.1 percentage point increase in risk per score increase of 1, 95% CI 0.001 to 8.1) and post-ART PC5 (adjusted 4.8 percentage point increase in risk per score increase of 1, 95% CI 0.6 to 8.9) were associated with death.

Table 4. Regression analyses assessing associations with death (N=18 outcomes) and TB-IRIS (N=32 outcomes).

Estimates are generated using linear regression models and adjusted estimates control for age, sex, extrapulmonary TB, non-TB opportunistic infection, number of days between initiating anti-TB therapy and ART, baseline CD4 count, baseline HIV log viral load.

	Unadjusted		Adjusted
	Percentage Point Change in Risk	95% confidence interval	Percentage Point Change in Risk	95% confidence interval
Death (N=17 for pre-ART PCs, N=10 outcomes for post-ART PCs)
PC3 (pre-ART)	1.1	−2.9 to 5.2	3.2	−1.2 to 7.7
PC3 (post-ART)	2.7	−0.9 to 6.2	4.1	0.001 to 8.1
PC5 (post-ART)	3.4	−0.4 to 7.2	4.8	0.6 to 8.9
TB-IRIS (N=32 outcomes)
PC3 (pre-ART)	2.9	−2.2 to 8.0	3.3	−2.7 to 9.4
PC3 (post-ART)	4.6	−1.1 to 10.2	6.0	−0.01 to 12.4
PC5 (post-ART)	−3.7	−9.8 to 2.3	−4.9	−11.5 to 1.7

Supplementary Figure 2 illustrates the distribution of post-ART PC3 and PC5 scores, showing the overrepresentation of positive PC3 and PC5 scores among subsequent deaths, the overrepresentation of food insecurity among positive PC3 scores, and the overrepresentation of undernutrition among positive PC5 scores.

Secondary Analyses

Our primary findings were similar when considering food insecurity as a categorical variable (Supplementary Figure 3), after adjusting food insecurity estimates for undernutrition (Supplementary Table 2), after adjusting undernutrition estimates for food insecurity (Supplementary Table 3), after inclusion of interaction terms between food insecurity and undernutrition, (Supplementary Table 4), after adjusting for the change in HIV viral load during the first four weeks of ART (Supplementary Table 5), and after excluding participants with BMI ≥25 (Supplementary Table 6). In adjusted analyses, neither food insecurity (adjusted 2.0 percentage point increase in risk per increased category of food insecurity severity, 95% CI −3.0 to 7.0) nor undernutrition (adjusted 2.9 percentage point increase in risk, 95% CI −7.3 to 13.0) were significantly associated with death. Undernutrition (adjusted 14 percentage point decrease in risk, 95% CI −27.1 to −1.4) was associated with a large reduction in the risk of TB-IRIS, and food insecurity was not significantly associated with TB-IRIS (3.5 percentage point increase in risk, 95% CI −6.5 to 13.4).

Discussion

In this longitudinal analysis of 165 ART-naïve people with advanced HIV and pulmonary TB initiating ART, nearly three-quarters of whom were food insecure and over one-third of whom had undernutrition, we identified distinct immunologic profiles associated with food insecurity and undernutrition. The immunologic profile associated with food insecurity was driven by decreased concentrations of the growth factor IL-3 (and to a more modest extent increases in VEGF), higher levels of the pro-inflammatory cytokine IFN-α (and to a more modest extent IL-1α), higher concentrations of the Th1 cytokines IFN-γ and IL-12p40, and more modestly by increased levels of the chemokine IL-8. On the other hand, the immunologic profile of undernutrition was related to higher levels of the chemokines IL-8 and MIP-1α and the pro-inflammatory cytokine IL-6, and to a lesser extent decreases in the chemokine IP-10, the pro-inflammatory cytokine IFN-α, and increases in the anti-inflammatory cytokine IL-10. The immunologic profiles associated with food insecurity and undernutrition were both in turn associated with death. Undernutrition was also associated with a lower risk of TB-IRIS.

Undernutrition is thought to lead to impairments in innate and adaptive immune responses and is considered the most common cause of secondary immunodeficiency in the world.^6,14 IL-8, the primary biomarker elevated in the immunological profile associated with undernutrition in our study, is secreted by many cell types but recruits neutrophils to inflammatory sites, and both neutrophilia and IL-8 plasma concentrations have been associated with poor HIV and TB outcomes.^35–39 Among people with latent TB infection, undernutrition has been associated with increased neutrophil and T-cell activation, proinflammatory IL-1 and IL-6 signaling (as also seen in our analysis in the context of active TB), and a diminished Th1 response.^15,16,18 In addition, lower baseline and TB-antigen-stimulated levels of several chemokines have been identified, including IP-10 and MIP-1β.¹⁷ These perturbations are hypothesized to contribute to the high risk of incident TB disease with undernutrition, with a population attributable fraction >25%.⁴⁰ Some of these changes may be driven by helminth co-infection and a shift from a Th1 to a Th2 response, with preliminary evidence suggesting some reversals after anti-helminthic therapy.^41,42 These abnormalities, along with those identified in our analysis, may play a role in the poor outcomes seen among people with undernutrition and active TB and HIV (either separately or with co-infection).^4,5 In addition, these changes may also explain the lower risk of TB-IRIS seen with undernutrition in our study.

Food insecurity has been associated in the general population with increased levels of C-reactive protein.^11,12 Among women with HIV receiving ART in the US, food insecurity was associated with increased levels of IL-6 and tumor necrosis factor receptor 1 (TNFR1).¹⁹ In a different study of the same cohort, food insecurity was associated with increased activation of CD4 and CD8 T-cells, and increased senescence of CD8 T-cells.²¹ In another cohort of people with HIV receiving ART in the US, food insecurity was associated with greater immune activation, as evidenced by higher levels of soluble CD14 and soluble CD27.²⁰ In addition, angiogenesis, driven by pathways related to biomarkers like IL-8 (increased in profiles associated with both undernutrition and food insecurity in our study) or VEGF (increased in the profile associated with food insecurity in our study), has been implicated as a potential mechanism of TB dissemination.^43,44

There are several plausible mechanisms by which food insecurity may drive inflammatory changes independent of undernutrition. In the general population, food insecurity has been associated with diets with greater inflammatory potential,¹³ for example those containing higher intakes of fat, sugar, or processed meat.^45–48 High-fat diets, in particular, have been associated with food insecurity among people living with HIV, and may lead to changes in the gut microbiome, disturbance of the intestinal mucosal barrier, and increased risk of microbial translocation.^19,20,49 Importantly, however, these studies have been conducted in high-income settings where the relationship between food insecurity and dietary patterns may differ from Botswana and similar contexts. Food insecurity’s well-documented role as a psychological stressor and contributor to allostatic load are also potential mechanisms driving inflammatory changes.^50–53 The stress response is known to be a potent stimulator of inflammation in both acute and chronic settings.^54,55 People experiencing food insecurity may also have worse medication adherence and higher rates of substance use, which may influence inflammatory phenotype.²

Taken together with the existing literature, our findings suggest that different pathophysiologic processes may link food insecurity and undernutrition with poor outcomes in this precarious clinical setting. Consequently, some interventions designed to improve these outcomes may be more optimal for a person experiencing either food insecurity or undernutrition, and a simplistic interpretation and conflation of food insecurity and undernutrition may not lead to effective solutions. Conceptually, individuals with food insecurity, but not undernutrition, might benefit more from interventions that are more socioeconomic or structural in nature, such as unconditional cash transfers, general food provisions, livelihood interventions, housing or agricultural support, interventions designed to increase resiliency against climate change, or nutritional counseling (see the conceptual model in Figure 1).⁵⁶ Individuals experiencing primarily undernutrition, but not food insecurity, may benefit more from targeted food supplementation (either micronutrient or macronutrient), nutritional counseling, assessment and treatment for reversible causes of undernutrition such as through anti-helminthic therapy, and close follow-up of weight changes during the early months of treatment for HIV and TB.⁵ These kinds of directed interventions should be tested in future research individually or in combination.

An important limitation of our study is that undernutrition was assessed at treatment initiation, and we were unable to estimate pre-morbid BMI based on the duration of symptoms or self-reported estimates of weight loss. As a result, undernutrition as measured in our analysis represents a combination of longstanding deficiencies and more recent consequences of disease severity (e.g., catabolic states induced by HIV and TB, and diarrhea associated with HIV). However, our study only included people with advanced illness, so distributions of nutritional status in our population may better reflect pre-morbid distributions. In addition, both premorbid undernutrition and undernutrition at treatment initiation have been associated with poor outcomes,⁵ so the association between undernutrition at treatment initiation and immunologic phenotype remains relevant. Our study evaluated a very specific population — people co-infected with HIV and TB with advanced immunosuppression early during treatment — and future research should address whether the identified distinctions between undernutrition and food insecurity are also present in other contexts. Another limitation of our study is that there was very little reported alcohol use, which is known to be related to both food insecurity and undernutrition. Therefore, our results may not apply to populations with higher rates of alcohol use. Additionally, our food insecurity measure focuses primarily on food access, rather than food quality, which may also have important nutritional and immunologic implications and should be assessed in future research. Finally, since undernutrition and food insecurity are both associated with poverty, there may be other factors co-correlated with poverty (e.g., indoor air pollution, unclean water) that were not specifically measured in this study which account in part for the associations we identified. Neither food insecurity nor undernutrition were significantly associated with death in our analysis, though we would caution against drawing conclusions based on this given a lack of statistical power, and note that these have been associated with poor outcomes in other settings.

Conclusion

Our analysis identified two independent immunologic pathways associated with food insecurity and undernutrition in the context of HIV/TB coinfection with advanced immunosuppression before and early after ART initiation, each of which was associated with subsequent death. Different pathophysiologic processes may link food insecurity and undernutrition with poor outcomes. Individuals with food insecurity may require distinct interventions compared to those with undernutrition to improve immunological function and treatment outcomes.

Abbreviations:

ART

antiretroviral therapy

AUDIT

Alcohol Use Disorders Identification Test

BMI

body mass index

EGF

epidermal growth factor

G-CSF

granulocyte-colony stimulating factor

GM-CSF

granulocyte macrophage colony-stimulating factor

HFIAS

Household Food Insecurity Access Scale

HIV

human immunodeficiency virus

IFN

interferon

IP

interferon-γ-induced protein

MCP

monocyte chemoattractant protein

MIP

macrophage inflammatory protein

PC

principal component

PCA

principal components analysis

TB

tuberculosis

TNF

tumor necrosis factor

VEGF

vascular endothelial growth factor