Maternal HIV, antiretroviral timing, and spontaneous preterm birth in an urban Zambian cohort: the role of local and systemic inflammation

Abstract

Objective:

To assess plasma and vaginal inflammation in 3 antenatal groups (HIV-uninfected women, HIV-infected women entering care on preconceptional ART, and HIV-infected women not on preconceptional ART) and whether these measures are associated with spontaneous preterm birth (sPTB).

Design:

Case-control study nested within a pregnancy cohort in Lusaka, Zambia.

Methods:

We analyzed 11 pro- and 2 anti-inflammatory markers in 207 women with paired plasma and vaginal specimens collected between 16–20 gestational weeks. Among 51 HIV-infected women, we repeated the assays in 24–34 week samples. We used confirmatory factor analysis to create inflammation scores and compared them among the 3 groups.

Results:

At baseline, HIV-infected women not on ART had higher vaginal pro-inflammatory scores than HIV-uninfected women (mean 0.37 [95%CI −0.06, 0.80] vs. −0.02 [−0.32, 0.27], p=0.02). In repeat testing, women not on preconceptional ART had an increase in vaginal inflammation between the baseline and 24–34 week visits compared to those continuing preconceptional ART (mean 0.62 [95%CI −0.80,4.20] vs. −0.07 [−2.78,2.11], p=0.04). In multivariate analyses, baseline vaginal inflammation predicted sPTB (aOR 1.5; 95%CI 1.0–2.3; p=0.02). Plasma inflammation did not differ by HIV or ART exposure and was not associated with sPTB.

Conclusions:

Women not receiving ART at entry into pregnancy care had more vaginal inflammation than women entering on treatment. They also experienced an increase in vaginal inflammation between the 2 sampling timepoints, possibly as a consequence of ART initiation. Vaginal (but not systemic) inflammation was associated with sPTB and offers a potential mechanistic insight into this important adverse birth outcome.

Introduction:

Preterm birth (PTB) is the world’s leading cause of neonatal death[1, 2]. The highest burden of PTB is borne by women living in Africa and south Asia, where more than 60% of cases occur and clinical infrastructure is often ill-equipped to care for premature newborns[3–6]. In sub-Saharan Africa, where HIV is endemic[7], PTB rates are among the highest in the world. Maternal HIV and exposure to antiretroviral therapy (ART) have been associated with elevations in prematurity risk[8, 9], with some studies suggesting that women initiating ART prior to conception have a higher risk of delivering prematurely than those initiating ART during pregnancy[10, 11].

PTB is a complex syndrome with several distinct phenotypes[12]. At the broadest level, PTB phenotypes can be divided into those arising spontaneously and those initiated by a provider. Intrauterine infection, which may be overt or subclinical, has long been associated with the spontaneous PTB phenotype (sPTB)[12–18]. Chronic inflammation and immune activation occurring in the absence of infection are also correlated with sPTB. For instance, rates of sPTB are higher in settings of obesity[19] and autoimmune disease[20–22]. The mechanisms by which intrauterine infection and inflammation lead to sPTB may be related to innate immune system activation[23], which, through a complex network of cytokines, chemokines, and other signaling molecules, promotes cervical ripening, uterine contractions, and amniotic membrane rupture[24–26]. The finding that women who become pregnant on ART (and whose HIV is better-controlled) have higher rates of PTB than women who enter antenatal care not on therapy is a counterintuitive observation that may be related to selection bias[27, 28].

Although the relationship between inflammatory cytokines and sPTB has been previously reported[29–31], few studies have been conducted in HIV-infected or African cohorts[32–34]. In this study, we sought to characterize local and systemic inflammation through the evaluation of vaginal and plasma cytokine profiles in a Zambian antenatal cohort, stratified by HIV serostatus. We hypothesized that the elevated risk of sPTB in HIV-infected women and those with preconceptional ART exposure may be related to measurable differences in the local or systemic immune milieu.

Materials and Methods:

Patient Population:

The Zambian Preterm Birth Prevention Study (ZAPPS) is an ongoing prospective obstetrical cohort at the Women and Newborn Hospital of the University Teaching Hospital (UTH) in Lusaka, Zambia[35, 36]. Written informed consent for collection of clinical data and biological specimens is obtained prior to study enrollment. Women are enrolled in early pregnancy, and antenatal care is provided at a study-run clinic. Ultrasounds are performed to determine gestational age, to measure cervical length at 20–24 weeks, and to monitor fetal growth in the third trimester. At delivery, birth phenotypes are assigned based on maternal and fetal factors as well as pathway to delivery. Using these phenotypes, parturition is classified as either spontaneous (e.g. cervical dilation, contractions) or provider-initiated (no signs of spontaneous labor prior to initiation of induction of labor or cesarean) [36, 37].

For the current nested case-control study, we defined a 3-level exposure variable that characterizes participants by HIV serostatus at enrollment and, among the HIV-infected, by the timing of ART exposure (preconceptional or not) (Supplemental File).

Specimen Collection and Storage:

Vaginal swabs and plasma were collected and stored according to standard operating procedures (Supplemental File). Study specimens are periodically shipped from Zambia to a sister biorepository in Chapel Hill, NC where we maintain a duplicate archive.

Sample Selection:

Eligibility for this study was limited to ZAPPS participants with paired plasma and vaginal specimens collected between 16–20 weeks of gestation available in the initial May 2017 biorepository shipment (Figure 1). We excluded women with an unknown delivery outcome (n=86), provider-initiated PTB (n=7), and unknown PTB phenotype (n=2). Additionally, HIV-infected women with unknown ART exposure status (n=2) were excluded. To maximize information from available specimens, we used a double sampling technique of eligible specimens that included all participants who were HIV-infected and all participants with sPTB. Among remaining HIV-uninfected participants delivering at term, we selected specimens at random, at a proportion dictated by available budget. We applied inverse probability sampling weights to our factor analyses as well as all subsequent univariate and multivariate analyses using the probability of selection among the baseline samples. Weighted regression models were estimated with robust standard errors computed by the linear variance estimator[41], using the probability svy function in STATA.[42] Additionally, for all HIV-infected women with matched repeat vaginal and plasma specimens obtained between 24–34 gestational weeks, we repeated assays. Further, as twin pregnancy and short midtrimester cervix (<2.5cm) are strong independent predictors of sPTB and rare in this study, women with these risk factors were excluded from PTB analyses (Figure 1).

Figure 1:

Study Sampling Technique

Cytokine Array Analysis:

Plasma and vaginal swab samples were analyzed by multiparameter bead array for interleukin-1β (IL-1β), IL-2, IL-6, IL-8, IL-10, IL-12p70, interferon-ɣ (IFN-ɣ), IFN-ɣ-induced protein 10 (IP-10), tumor necrosis factor-α (TNF-α), macrophage inflammatory protein-1α (MIP-1α), MIP-1β, and transforming growth factor-β (TGF-β) using a commercially available Milliplex-kit (Millipore-Sigma, Minneapolis, MN). In addition, we measured soluble CD14 (sCD14) by ELISA (R&D Systems, Darmstadt, Germany). Cytokine concentrations were calculated according to the manufacturer’s protocols. For values below the limit of detection, we imputed the value at half of the lower limit of detection. We did not apply newer methods for handling limits of detection[43, 44] because they cannot currently be applied to factor analysis. All concentrations were logarithmically transformed prior to analysis.

Data Analyses:

We analyzed baseline demographic data of the sub-study cohort, calculating median and interquartile range (IQR) for each continuous variable, and frequency and percentage for each categorical variable. We analyzed differences in continuous independent variables between exposure groups using Wilcoxon ranksum and Kruskal-Wallis test for comparisons between two or more than two groups, respectively.

From a biostatistical perspective, the analysis of cytokines is challenging due to strong correlations between cytokines, particularly those with similar underlying biological processes. Previous work by Genser et al[45] found that most immunological studies use basic statistical methods that do not account for correlations or relationships among study variables, such as cytokines. Thus, they recommend that more advanced and sophisticated multivariate statistical techniques be used to assess complex immunological data. Factor analysis is one such multivariable statistical technique that reduces a large number of intercorrelated variables into a smaller set of independent clusters underlying physiological relationships. As the biologic actions of cytokines can be broadly classified as either pro- or anti-inflammatory, we performed a confirmatory factor analysis, which is a hypothesis-driven approach developed to assess whether associations among many variables confirm hypothesized underlying factors. [20, 46] This approach contrasts an exploratory factor analysis, which lacks a hypothesis and a priori assumptions (and was not used in this study). We assigned cytokines as either pro-inflammatory (MIP-1α, MIP-1β, IL-1β, IL-2, IL-6, IL-8, IL-12p70, IFN-ɣ, IP-10, TNF-α, and sCD14) or anti-inflammatory (IL-10 and TGF-β). Consequently, pro- and anti-inflammatory activity were the hypothesized underlying factors or latent variables. Cytokines with standardized loading or R2 values (coefficient of determination) with p-values >0.01 were excluded to reduce estimated parameters and improve model fit, as assessed by Root Mean Square Error of Approximation (RMSEA), Comparative Fit Index (CFI) and Tucker-Lewis Index (TLI). Additionally, as previous studies have strongly hypothesized that immunological processes and expression of cytokines evolve differently in blood versus vaginal fluid[29, 31], confirmatory factor analyses were conducted with separate models for blood and vaginal fluid.

Factor scores from the confirmatory factor analysis at baseline were estimated for each patient. A factor score is a relative value to a latent variable (i.e. the factor: pro- or anti-inflammatory activity) for a particular patient as compared with other women included in the analysis. By default, the estimated factor scores have a mean of zero. Thus, factor scores cannot be interpreted in absolute terms but instead are relative values (e.g. a factor score of 0 does not indicate an absence of inflammation). Based on this model, an individual with a positive vaginal pro-inflammatory factor score is assumed to have increased vaginal inflammation as compared with the mean. Associations between underlying pro- and anti-inflammatory activities in both blood and vaginal fluid were assessed by plotting factor scores and calculating Spearman correlations coefficients.

We modeled factor scores versus HIV and ART exposure groups using linear regression with inverse-probability weighting, controlling for maternal age, parity, BMI, and prior PTB. As the pro- and anti-inflammatory factor scores were highly correlated within the plasma and vaginal compartments, only the pro-inflammatory factors were included in subsequent modeling, owing to their representation of the interaction and diversity of more cytokines and their improved model fit statistics. Associations between baseline factor scores and sPTB were assessed using logistic regression to calculate unadjusted and adjusted case-control prevalence odds ratios.

In the subset of HIV-infected participants with repeat specimens (Table 1), we repeated the model developed from our baseline factor analysis. In this analysis, we included specimens from baseline (16–20 weeks) and repeat (24–34 weeks) visits to estimate factor scores for each participant at both time points. We then produced a delta pro-inflammatory score by subtracting the baseline factor score from the repeat factor score. Consequently, a positive delta pro-inflammatory score represents an increase in relative inflammation from baseline to repeat, and a negative delta pro-inflammatory score represents a decrease. Delta pro-inflammatory scores were compared by preconception ART exposure groups using Wilcoxon ranksum test.

Table 1:

Baseline characteristics of participants with specimens analyzed

Characteristic	Baseline				Repeat
	Overall	HIV-uninfected	HIV-infected preconceptional ART	HIV-infected no preconceptional ART	Overall	HIV-infected preconceptional ART	HIV-infected no preconceptional ART
	N (%) or median (IQR)	N (%) or median (IQR)	N (%) or median (IQR)	N (%) or median (IQR)	N (%) or median (IQR)	N (%) or median (IQR)	N (%) or median (IQR)
Total number of women	207	130	46	31	51	30	21
EGA at collection, weeks	18 (17, 19)	18 (17, 19)	18 (16, 18)	18 (17, 19)	18 (16, 19)	18 (16, 18)	18 (17, 19)
Age, years	26 (22, 32)	25 (22, 31)	30 (27, 33)	25 (21, 29)	28 (25, 32)	30 (27, 33)	25 (21, 29)
<20	16 (8%)	13 (10%)	0 (0%)	0 (0%)	2 (4%)	0 (0%)	2 (10%)
20–34	164 (80%)	97 (76%)	39 (84%)	28 (90%)	43 (86%)	24 (83%)	19 (90%)
≥35	24 (12%)	17 (13%)	7 (15%)	3 (10%)	5 (10%)	5 (17%)	0 (0%)
missing	3	3	7	3	1	1	0
Married or cohabitating	173 (84%)	110 (85%)	38 (84%)	25 (83%)	42 (82%)	26 (87%)	16 (76%)
missing	2	0	1	1	1	0	1
BMI, kg/m2	23 (21, 27)	24 (21, 28)	24 (22, 26)	22 (20, 25)	23 (21, 25)	23 (22, 26)	23 (21, 25)
<18.5	7 (4%)	4 (3%)	1 (2%)	2 (7%)	1 (2%)	0 (0%)	1 (5%)
18.5–30	161 (80%)	96 (77%)	38 (84%)	27 (90%)	45 (90%)	27 (90%)	18 (90%)
>30	32 (16%)	25 (20%)	6 (13%)	1 (3%)	4 (8%)	3 (10%)	1 (5%)
missing	7	5	1	1	1	0	1
Parity	1 (0,2)	1 (0,2)	2 (1,3)	1 (1,2)	2 (1,3)	2 (1,3)	1 (1,2)
Prior PTB
0 (nulliparous)	54 (26%)	41 (32%)	6 (13%)	7 (23%)	7 (14%)	3 (10%)	4 (19%)
0 (parous)	93 (45%)	55 (42%)	27 (59%)	11 (35%)	25 (49%)	18 (60%)	33 (7%)
1 or more	60 (29%)	34 (26%)	13 (28%)	13 (42%)	19 (37%)	9 (30%)	10 (48%)
Twins	8 (4%)	5 (4%)	1 (2%)	2 (6%)	1 (2%)	1 (3%)	0 (0%)
Short Cervix < 2.5 cm	5 (2%)	2 (2%)	2 (5%)	1 (3%)	2 (4%)	1 (4%)	1 (5%)
missing	18	13	4	1	2	1	1

IQR, interquartile range; BMI, body mass index; PTB, preterm birth; EGA, estimated gestational age

Confirmatory factor analysis and generation of factor scores were performed with Mplus software (version 8.0; Muthen & Muthen, Los Angeles, CA). We used STATA® (version 15.0; StataCorp, College Station, TX) for data management, descriptive statistics, and analyses. ZAPPS is registered at ClinicalTrials.gov (NCT02738892), and has ongoing approval by the University of Zambia Biomedical Research Ethics Committee and the University of North Carolina Institutional Review Board.

Results:

Study Sample:

We performed cytokine assays on 207 (58.3%) of the study-eligible 355 participants (45.8% of available 452 participants in initial biorepository shipment) with paired vaginal and plasma baseline specimens, including all HIV-infected women with known ART exposure (77/77), all HIV-uninfected women with sPTB (19/19), and a random sample of HIV-uninfected women who delivered at term (111/259, 42.9%; Figure 1). Of the paired specimens available from HIV-infected participants with known ART exposure, 46 (59.7%) were on preconceptional ART, 31 (40.3%) were not (Table 1). Of women with preconceptional ART exposure, 6 (13.0%) had detectable viral loads at enrollment.

The median age of participants at enrollment was 26 (IQR: 22, 32). HIV-infected participants on preconceptional ART were older (median: 30; IQR: 27, 33) than uninfected women (median: 26; IQR: 22, 32) and HIV-infected women without preconceptional ART (median: 25; IQR: 22,31; p=0.01). HIV-infected participants without preconceptional ART exposure had lower BMIs (median: 22.4; IQR: 20.3, 25.5) than both uninfected participants (median: 24.2; IQR: 21.9, 28.8) and infected participants on preconceptional ART (median: 24.4; IQR: 22.1, 26.4; p<0.01). Parity also differed by HIV serostatus and ART exposure: HIV-infected participants with preconceptional ART exposure had higher parity (median: 2; IQR: 1, 3) compared to those without preconceptional ART exposure (median: 1; IQR: 1, 2) and uninfected participants (median: 1; IQR: 0, 2; p=0.03).

Of 207 women included in this nested case-control study, 29 (14.0%) had sPTB. The majority of these sPTB outcomes were in HIV-uninfected participants (19/29, 66.5%). Ten (34.4%) sPTB occurred in HIV-infected women: 6 were on preconceptional ART and 4 were not. Of the 29 women with sPTB, 4 (14%) delivered before 28 weeks and 10 (34%) delivered before 32 weeks.

Confirmatory Factor Analysis:

A confirmatory factor analysis with cytokines assigned to pro- or anti-inflammatory biological action was conducted for vaginal and plasma samples at baseline. Factor loadings and R2 values were calculated (Supplemental Table 1). In vaginal fluid, MIP-1α and IFN-ɣ had standardized loadings with p-values >0.01 in the pro-inflammatory latent variable and were excluded from the final model. In plasma, MIP-1α, IL-6, IL-8, IL-12p70, IP-10, and sCD14 were excluded from the pro-inflammatory latent variable for the same reason. Using the model developed in the baseline confirmatory factor analysis, a second model was run including only matched specimens (women with both baseline and repeat specimens) to generate comparable baseline and repeat factor scores.

Using estimated factor scores from the confirmatory factor analysis for each patient, the correlations between underlying pro- and anti-inflammatory activities in both plasma and vaginal fluid samples were quantified with Spearman correlation coefficients (Supplemental Figure 1). Within each compartment, strong correlations were observed between factor scores for pro- versus anti-inflammatory activity (vaginal fluid: r=0.83; plasma: r=0.99). Pro-inflammatory or anti-inflammatory factor scores between the two distinct anatomic compartments – vaginal fluid versus plasma – were not or were only weakly correlated with each other.

Inflammation & HIV:

When controlling for maternal age, parity, BMI, and prior PTB, vaginal baseline specimens of HIV-infected women without preconceptional ART exposure had higher pro-inflammatory scores (mean 0.37 [95%CI −0.06, 0.80] vs. −0.02 [−0.32,0.27]; p=0.01) and anti-inflammatory scores (mean 0.33 [95%CI 0.11,0.55] vs. −0.03 [−0.18,0.12]; p<0.01) compared to HIV-uninfected women (Figure 2a,b). Pro-inflammatory scores (mean 0.21; 95%CI −0.21,0.62) and anti-inflammatory scores (mean 0.05; 95%CI −0.16,0.27) of HIV-infected women with preconceptional ART exposure did not differ from those of either their HIV-infected counterparts without preconceptional ART exposure or from HIV-uninfected women. In contrast to vaginal scores, no differences between HIV and ART sub-groups were observed for pro- or anti-inflammatory plasma scores (Figures 2c,d).

Figure 2:

Pro- and anti-inflammatory scores in vaginal fluid and plasma between 16 and 20 gestational weeks by HIV serostatus and preconceptional ART exposure

In analyses of matched baseline and repeat specimens of HIV-infected participants, women without preconceptional ART (mean 0.62; 95%CI −0.80,4.20) had a larger increase in vaginal pro-inflammatory scores than women with preconceptional ART (mean −0.07; 95%CI −2.78,2.11; p=0.04; Figure 3). No differences were observed with respect to changes in plasma pro-inflammatory scores by ART timing sub-groups.

Figure 3:

Delta pro-inflammatory scores in (a.) vaginal fluid and (b.) plasma from baseline to repeat by preconconceptional ART exposure in HIV-infected women.

Inflammation & sPTB:

In our study population of 207 women, 8 had twin gestation and 5 had short cervix, both known predictors of sPTB. In fact, 88% (7/8) of women with twins and 60% (3/5) with short cervix had sPTB. Given the small populations with these strong sPTB predictors, these 13 women were excluded from analyses of sPTB (Figure 1).

Women with sPTB had higher vaginal pro-inflammatory scores (mean 0.57 [95%CI 0.14, 0.99] vs. 0.02 [−0.26,0.23], p=0.02) and anti-inflammatory scores (mean 0.22 [95%CI 0.02,0.41] vs. −0.01 [−0.14,0.11]; p=0.05) in the midtrimester than their counterparts with term birth. When including vaginal pro-inflammatory score as a continuous variable in prevalence analyses, it was associated with increased sPTB in both unadjusted (OR 1.4; 95%CI 1.0,1.9; p=0.05) and adjusted (AOR 1.5; 95%CI 1.0,2.3; p=0.02; Table 2) analyses. Vaginal anti-inflammatory score also demonstrated an association with sPTB in adjusted analyses (AOR 1.5, 95%CI 1.0–2.3, p=0.03). Neither plasma pro- nor anti-inflammatory factor scores were associated with sPTB in unadjusted or adjusted analyses.

Table 2:

Prevalence odds ratios of spontaneous PTB in participants with baseline specimens analyzed

Exposure	Spontaneous PTB <37 weeks
	OR (95% CI)	p-value	AOR (95% CI)	p-value
Vaginal Pro-Inflammatory Factor Score	1.4 (1.0–1.9)	0.05	1.6 (1.1–2.3)	0.02
Vaginal Anti-Inflammatory Factor Score	1.3 (0.9–1.8)	0.07	1.5 (1.0–2.3)	0.03
Plasma Pro-Inflammatory Factor Score	1.4 (0.6–3.0)	0.39	1.3 (0.6–3.1)	0.53
Plasma Anti-Inflammatory Factor Score	1.2 (0.6–2.8)	0.60	1.1 (0.5–2.6)	0.87

OR: odds ratio; AOR: adjusted odds ratio

Multivariable model estimates of adjusted odds ratios are adjusted for HIV serostatus, ART exposure, prior PTB, maternal age, and BMI

Models exclude women with twin pregnancies (n=8) or short cervix (n=6)

Discussion:

Using confirmatory factor analysis, we identified pro- and anti-inflammatory cytokines that reflect the underlying local (vaginal fluid) and systemic (maternal plasma) immune milieu during the midtrimester in a Zambian cohort with high HIV and PTB prevalence. Our results demonstrate a relationship between a pro-inflammatory vaginal environment and the risk of subsequent sPTB. Further, HIV-infected women without preconceptional ART exposure were found to have a greater increase in vaginal inflammation between baseline and repeat sampling as compared to their HIV-infected counterparts with preconceptional ART exposure. These data support lower genital tract inflammation as a mechanism of sPTB, and further imply a potential link between timing of ART initiation and the elevated PTB risk among HIV-infected women.

Infection and inflammation are well-defined risk factors for sPTB[12]; thus, many cytokines in serum, amniotic fluid, or cervical fluid have been implicated as mediators of sPTB[29, 30, 47–55]. As cytokines function as part of an elaborate and complex network of molecules that regulate the immune response, we characterized the systemic and local immune milieu through confirmatory factor analysis, categorizing cytokines as pro- or anti-inflammatory. This approach is well-suited to assess the association and correlation between pro- or anti-inflammatory activities expressed biologically by cytokines and has been used in other diseases for this purpose[56–59]. Previously, Simhan and colleagues identified an association between elevated cervical cytokine inflammation at <16 weeks gestation and increased risk of sPTB using exploratory factor analyses[31]. Our study supports this finding by demonstrating a relationship between increased lower genital tract inflammation in the midtrimester and subsequent sPTB.

Consistent with current literature evaluating local and systemic biomarkers in pregnancy[47, 60], inflammatory scores between vaginal and venous compartments were poorly correlated. We also found no association between plasma inflammatory scores and sPTB. Although some prior studies have identified an association between specific plasma cytokines and sPTB[29, 52, 60, 61], our data are supported by multiple studies[62, 63] including a systematic review[47] disputing this relationship. We do highlight that while our vaginal pro-inflammatory latent variable included factor loadings from 9 cytokines, our plasma pro-inflammatory latent variable included factor loadings from only 5 cytokines, following our exclusion criteria to optimize fit. This finding may indicate that pro-inflammatory cytokines in serum are less strongly correlated, their relationships are more complex than in vaginal fluid, or they are impacted by a greater variety of stimuli than the cytokines in vaginal fluid.

Our results support the important role of lower genital tract inflammation as a risk factor for HIV- and ART-associated sPTB. Previously, Kyongo and colleagues[32] demonstrated elevated cervicovaginal pro-inflammatory cytokines in HIV-infected women on preconceptional ART in the first trimester of pregnancy, as compared to HIV-uninfected women. Our comparison of lower genital tract inflammation in HIV-infected women stratified by preconceptional ART exposure is novel. This is particularly important for pregnant women in sub-Saharan Africa as 91% of the 1.4 million HIV-infected pregnant women not currently on ART reside in this region[7, 64, 65]. We found that women without preconceptional ART have elevated vaginal inflammation at their initial prenatal encounter that was further increased later in pregnancy. While most HIV-infected women in Zambia who are not already taking ART do initiate during pregnancy, this study did not explicitly confirm initiation of ART. Despite this limitation, these data suggest that immune reconstitution after ART initiation may result in increased local inflammation, a finding that conflicts with epidemiologic reports that those initiating ART prior to pregnancy have a higher PTB risk than untreated women[10, 11]. These data further support recent work proposing selection bias in observational studies to be responsible for the posited association between preconceptional ART and elevated PTB risk[27]. Specifically, women who deliver preterm before initiating ART are systematically excluded from analyses, resulting in a falsely lowered risk of preterm birth among women remaining and inflating the preterm birth risk when compared to women on preconceptional ART. Further, multiple studies have identified a link between the use of protease inhibitors and elevated PTB risk[66–69]. However, first-line ART does not include protease inhibitors in Zambia. As the majority of HIV-infected women in Lusaka are prescribed a fixed-dose combination of tenofovir, emtricitabine, and efavirenz, analyses by drug regimen were not conducted. However, future studies should consider the effect measure modification of therapeutic regimens on inflammation.

Strengths of this study include early ultrasound gestational age dating and careful phenotyping of each PTB outcome. We are confident in our categorization of gestational age at delivery and exclusion of non-spontaneous phenotypes (e.g., provider-initiated delivery for preeclampsia), short cervix, and twin gestation. We chose a sampling approach that deliberately enriched for HIV and sPTB.

In the ZAPPS cohort, HIV-infected women represent 24% of all participants[35]; whereas they accounted for 37% in the current study. Although we adjusted all analyses to account for our sample selection, characterizing the inflammatory profiles of all ZAPPS participants after all women complete their pregnancies would provide more generalizable population-level data. Additionally, it is generally recommended that factor analysis models obtain fit statistics of RMSEA <0.06 as well as CFI and TFI >0.95[70]. We recognize that our factor scores do not meet all of these thresholds. Logarithmic transformation and a limited sample size may have influenced model fit. Additionally, the categorization of cytokines as pro- or anti-inflammatory does not fully reflect their dynamic functions and complex interactions[71]. Despite this limitation, the fits of our models are comparable or better than previously published factor analysis models of cytokines[56, 72–74]. Lastly, we would expect the association between lower genital inflammatory milieu and sPTB to be strongest with parturition resulting from intrauterine infection or inflammation. Placental histopathology, culture, or molecular pathogen detection methods would be necessary to confirm these processes. Unfortunately, these data are not yet available for the ZAPPS cohort and were a limitation of this analysis.

In conclusion, our study demonstrates a relationship between increased genital tract inflammation and subsequent sPTB, offering a potential mechanistic insight into this important adverse birth outcome. Women not receiving ART at initiation of pregnancy care had more vaginal inflammation than women entering on treatment. Further, they experienced an increase in vaginal inflammation between the two sampling timepoints, possibly as a consequence of ART initiation. As many African settings offer near universal coverage of HIV testing in pregnant women, antenatal care provides a unique opportunity to connect HIV-infected women with treatment. Initiation of ART in pregnancy in previously untreated HIV-infected women is an important component of prevention of mother to child transmission and optimization of maternal health. However, further research is needed to characterize the nuanced relationship between ART initiation and sPTB as well as inflammation and sPTB.

Supplementary Material

Supplemental Figure 1

Supplemental Figure 1: Correlations between factor scores from the confirmatory factor analysis