Vertical Cytomegalovirus Transmission From HIV-Infected Women Randomized to Formula-Feed or Breastfeed Their Infants

Abstract

Background. Cytomegalovirus (CMV) is associated with morbidity and mortality in human immunodeficiency virus (HIV)–exposed infants. We assessed the effect of and relative contribution of breastfeeding to CMV acquisition among infants delivered by HIV-infected mothers.

Methods. Between 1993 and 1998 pregnant, HIV-infected women in Nairobi, Kenya, were randomly assigned to breastfeed or formula-feed their infants in an HIV transmission study. Women were allocated equally between treatment arms, and the study was not blinded. The primary endpoint of this nested study was time to infant CMV infection.

Results. CMV infection was assessed in 138 breastfed and 134 formula-fed infants. Baseline characteristics were similar between arms. Breastfed infants acquired CMV earlier than formula-fed infants (median age of acquisition, 4.26 vs 9.87 months; P < .001) and had a higher 1-year probability of CMV infection (0.89 vs 0.69; P < .001). Breastfeeding was associated with a 1.6-fold increased risk of infant CMV acquisition independent of infant HIV status (multivariable hazard ratio, 1.61; 95% confidence interval, 1.20–2.16; P = .002). Approximately one third of CMV infections occurred during the peripartum period, with 40% acquired through breastfeeding and the remainder acquired through modes other than breast milk.

Conclusions. Preventing CMV acquisition may be a priority for HIV-exposed infants, but there is a narrow window of opportunity for intervention. Approaches that reduce maternal cervical and breast milk CMV reactivation may help delay infant infection.

Cytomegalovirus (CMV) infects a large proportion of human immunodeficiency virus (HIV)–infected adults and children and plays a role in HIV pathogenesis that is not well understood. In addition to being an important opportunistic pathogen in severely immunosuppressed individuals, CMV is associated with a number of adverse outcomes in HIV-infected individuals. CMV viremia is independently associated with increased HIV disease progression and mortality [1–3]. In HIV-infected infants and HIV-exposed uninfected infants, CMV infection has been associated with early mortality, developmental delays, neurologic sequelae, and elevated immune activation [2–8].

Transmission rates are high between HIV-infected mothers and their children; in Kenya, >80% of HIV-exposed infants acquire CMV before 3 months of age [9]. A recent report from the Breastfeeding, Antiretrovirals, and Nutrition (BAN) study reported a CMV prevalence of 70% at 6 months of age in HIV-exposed infants from Malawi [10]. Infant CMV acquisition can occur in utero, during the intrapartum period, or via contact with infected breast milk, saliva, or urine [11–14]. Approximately one third of healthy pregnant women experience CMV reactivation in the cervix during the third trimester [15–17], and nearly all women have CMV DNA detectable in breast milk [18, 19]; virus replication in these compartments is associated with infant infection. The kinetics of breast milk CMV reactivation appear to differ according to HIV status. In HIV-uninfected women, Yasuda et al reported that, by 2 weeks after delivery, most breast milk specimens had detectable CMV DNA, with copy numbers peaking 4–6 weeks after delivery (range, 10⁴–10⁶ copies/mL) and declining rapidly in the following 2–4 weeks [20]. In Kenyan HIV-infected women, CMV loads peaked at 5.5 log₁₀ copies/mL 2 weeks after delivery but remained very high for 6 months [21]. Although each of these modes of transmission are understood to be important, their relative contribution to CMV infection in HIV-exposed children is unknown. A better understanding of the timing and modes of CMV transmission from HIV-infected women to their infants is important in designing interventions to prevent or delay CMV transmission and in the strategic deployment of a CMV vaccine for this population.

We used specimens and data from an historic randomized trial of mother-to-child HIV transmission, in which Kenyan HIV-infected women were randomly assigned to breastfeed or formula feed their infants [22]. In this study, we compared rates of CMV infection in HIV-exposed infants randomized to breast feeding or formula feeding and estimated the relative contribution of peripartum, breast milk, and other non-breast milk postnatal transmission.

MATERIALS AND METHODS

Study Design

The study was approved by the Institutional Review Board of the University of Washington and the Ethics and Research Committee of Kenyatta National Hospital. All participants provided written informed consent for participation, and human experimentation guidelines of the University of Washington were followed. This is a retrospective analysis of specimens collected during a randomized controlled trial of breastfeeding versus formula feeding and HIV transmission, conducted in Nairobi, Kenya, between 1993 and 1998 [22]. We assumed the rate of CMV acquisition in the breastfeeding arm would be similar to that in our previous observational study in Kenya, which was 90% by 6 months of age [9]. With equal allocation to study arms and estimating 300 infants would meet eligibility criteria, the study was powered to detect a 30% difference in transmission between arms, using Cox regression with β = 0.80 and α = 0.05, assuming a 2-sided test. Eligibility criteria for the randomized controlled trial included pregnancy of <32 weeks gestation, willingness to have feeding modality randomly assigned, access to clean municipal water, and intent to reside in Nairobi for >2 years. For the purpose of this substudy, we included all infants who had dried blood spot (DBS) specimens collected before 14 weeks of age.

Cohort and Randomization

This trial was conducted before antiretroviral therapy (ART) for prevention of mother-to-child HIV transmission (PMTCT) became available in Kenya; all women and infants were ART naive. At enrollment, caregivers provided sociodemographic characteristics and medical histories and underwent a clinical examination. At approximately week 32 of gestation, women were randomly assigned to the breastfeeding or formula-feeding arm at a 1:1 ratio, using computer-generated block randomization. Treatment allocation was not blinded to study staff; assignment to study arm was revealed to women and clinicians via presealed envelopes. In the formula group, safe formula preparation was demonstrated during home visits, and women were provided with free dried milk formula throughout the study. Women and infants were assessed in clinic monthly in the first year and quarterly during the second year of the infant's life and as needed for illness. Postnatal visits included assessment of both mother's and baby's clinical status, as well as infant growth and feeding. Infant blood specimens were collected at birth, at 6 weeks of age, at 14 weeks of age, and every 3 months thereafter until 24 months of age for HIV testing [22]. Women delivering at home or another facility were asked to bring their newborns to the study clinic for clinical examination and HIV testing as soon as possible following delivery. Infant HIV diagnosis was conducted in this study using polymerase chain reaction (PCR) to detect HIV gag DNA in peripheral blood mononuclear cells or DBS as previously described [22, 23]. Due to the high incidence of infant HIV infection in the breast fed arm, in April 1998 the data safety and monitoring board (DSMB) closed the study early, and recommended that breast feeding women be advised to stop breast feeding and be given formula [22].

CMV Diagnostic Assays

All CMV loads were measured in DBS as previously described [21]; plasma specimens from this study have been extensively used in vertical transmission studies over the past decade and were no longer were available for longitudinal CMV assessments. In our assay, when CMV loads are >300 copies/mL in plasma, the 2 assays are comparable for CMV detection, although they return lower CMV load measurements (Atkinson, unpublished data). CMV loads were measured in all specimens for each infant up to 12 months of age. The lower limit of detection was 100 copies/mL.

Infants with no detection of CMV DNA throughout follow-up underwent serologic testing of their last-collected plasma specimen closest to 12 months of age, to determine final infection status; only plasma specimens collected beyond 6 months of age were included for serologic analysis, to avoid confounding by maternal antibodies. Enzyme-linked immunosorbent assay (ELISA) was used to detect a panel of antibodies targeting CMV immunoglobulin G (CMV ELISA II test kit Wampole Laboratories, Princeton, New Jersey).

Statistical Analysis

All analyses were intent-to-treat and were performed using SPSS, version 20.0 (IBM). Baseline characteristics for infants included in the CMV analyses were compared between randomization arms using Mann–Whitney U tests (for continuous variables) and χ² tests (for categorical variables). HIV-exposed uninfected infants were defined as infants with their last CMV test at or after their last HIV-negative test and who never had HIV infection diagnosed. The proportion of infants infected with CMV at birth were compared between arms using Fisher's exact test; the proportion with CMV DNA detection and the proportion with CMV antibody detection were compared between arms using Pearson χ² tests. Time to CMV infection was compared between arms using Kaplan–Meier curves and the log-rank test. z tests were used to compare median time to CMV infection and the probability of CMV infection at 1 year. Cox proportional hazards models were used to assess the independent effects of time-varying HIV infection status and feeding modality, as well as any potential effect modification of these 2 variables on CMV acquisition.

In addition to the intent-to-treat analyses described above, all analyses were performed categorizing infants as breastfed or formula fed, based on the mother's self-reported feeding history. For these analyses, infants whose mothers ever reported breastfeeding were classified as breastfed, and infants whose mothers never reported breastfeeding were classified as formula fed.

Estimates of the proportion of CMV infections attributable to different modes of transmission in the first year of life were obtained by comparing the proportions of infants with CMV DNA detection in the breastfed and formula-fed infants at defined time points. Because of interval sampling at delivery and 6 weeks of age, we were unable to discriminate in utero from intrapartum infections; we thus combined infections detected at ≤6 weeks of age as “peripartum” transmissions, which are presumed to primarily include in utero and intrapartum and potentially very early breast milk or saliva transmissions. Infections due to saliva and/or urine exposure were defined as those infections occurring after 6 weeks of age in the formula-fed infants, and infections due to breast milk exposure were defined as the excess infections between the breastfed and formula-fed infants at 1 year of age. Because 30% of the women randomly assigned to the formula-feeding arm elected to breastfeed [22], this analysis was conducted as as-treated.

RESULTS

Cohort Characteristics

The trial CONSORT (Consolidated Standards of Reporting Trials) diagram is provided in Figure 1. A total of 138 breastfed and 134 formula-fed infants underwent CMV testing. Mothers in both arms were similar in terms of age, parity, and household size (Table 1). Median duration of breastfeeding was 14 months. At baseline, median prenatal CD4⁺ T-cell count was similar between arms (P = .2). Infants had a median of 11 study visits (interquartile range [IQR], 6–17 study visits) in the breastfeeding arm and 13 study visits (IQR, 8–18 study visits) in the formula feeding arm (P = .3), with a median of 4 study visits (IQR, 3–5 study visits) and 5 study visits (IQR, 3–5 study visits) assessed for CMV DNA, respectively (P = .2). As previously reported [22], breastfed infants were significantly more likely to acquire HIV than formula-fed infants (34.1% vs 14.9%; P < .001), and mothers in the formula-feeding arm were significantly more likely to cross over to the other treatment arm, compared with those in the breastfeeding arm (26.1% vs 4.35%; P < .001).

Figure 1.

CONSORT (Consolidated Standards of Reporting Trials) diagram, describing enrollment, allocation, follow-up, and analysis for the main randomized, controlled trial. ^aInfants with a specimen collected by 14 weeks of age were eligible for inclusion in the cytomegalovirus (CMV) substudy.

Table 1.

Characteristics of the Study Population

Characteristic	Breastfeeding Arm	Formula-Feeding Arm	P Valuea
Mothers, at enrollment
No.	138	134
Age, y	23 (20–27)	23 (21–27)	.9
Parity	1 (0–2)	1 (0–2)	.4
People in household, no.	3 (2–4)	3 (2–4)	.3
CD4+ T-cell count, cells/mm3	389 (261–530)	441 (282–549)	.2
HIV RNA load, log10 copies/mLb	4.68 (4.15–5.27)	4.59 (3.89–5.12)	.3
Infants
No.	138	134
Mother crossed over treatment	4.354 (6)	26.1 (35)	<.001
Breastfeeding duration, mo.	14 (7–18)	0 (0–0.4)	<.001
Acquired HIV	34.1 (47)	14.9 (20)	<.001
Study visits, no.	11 (6–17)	13 (8–18)	.3
Visits assessed for CMV, no.	4 (3–5)	5 (3–5)	.2
Died	20.3 (28)	14.9 (20)	.3

Data are median value (interquartile range) or proportion (%) of subjects, unless otherwise indicated.

Abbreviations: CMV, cytomegalovirus; HIV, human immunodeficiency virus.

^a By the Mann–Whitney U test (for continuous variables) or the χ² test (for categorical variables).

^b Data are from the measurement closest to delivery and are for 120 subjects in the breastfed arm and 119 in the formula-fed arm.

Feeding Allocation and CMV Acquisition

A total of 1103 infant samples were tested for CMV DNA; 547 (49.6%) were from the 138 breastfed infants and 556 (50.4%) were from the formula-fed infants. Overall, 20.1% of these samples had detectable CMV DNA, with 24.0% positive in the breastfeeding arm and 16.4% positive in the formula-feeding arm.

The rate of CMV DNA detection at birth was 3.73% in the formula-fed infants and 3.62% in the breastfed infants (P = 1.0; Table 2). A greater proportion of infants in the breastfeeding arm had detectable CMV DNA by 1 year of age (62.3% in breastfeeding arm vs 42.5% in the formula-feeding arm; P = .001). The overall proportion of infants who tested positive in the first year (by either CMV DNA detection or serology) was also greater in the breastfed infants (80.4%), compared with the formula-fed infants (60.5%; P < .001).

Table 2.

Cytomegalovirus (CMV) Acquisition in the 2 Study Arms

Variable	Breastfeeding Arm	Formula-Feeding Arm	P Value
All infants
No.	138	134
CMV infected at birth	3.62 (5)	3.73 (5)	1.0a
CMV DNA positive	62.3 (86)	42.5 (57)	.001
CMV DNA negative and serology positive	18.1 (25)	17.9 (24)	1.0
CMV DNA or serology positive	80.4 (111)	60.5 (81)	<.001
Probability of infection at 1 y	0.89 ± 0.03	0.69 ± 0.05	<.001b
Time to CMV infection, mo, median ± SE	4.26 ± 0.97	9.87 ± 1.27	<.001b
HIV-exposed uninfected infantsc
No.	90	111
CMV infected at birth	5.56 (5)	4.50 (5)	.8a
CMV DNA positive	63.3 (57)	38.7 (43)	.001
CMV DNA negative and serology positive	23.3 (21)	18.9 (21)	.4
CMV DNA or serology positive	86.7 (78)	57.7 (64)	<.001
Probability of infection at 1 y	0.89 ± 0.04	0.66 ± 0.05	<.001b
Time to CMV infection, mo, median ± SE	5.80 ± 1.04	10.6 ± 0.71	<.001b
HIV-infected infants
No.	47	20
CMV infected at birth	0	0
CMV DNA positive	59.6 (28)	65.0 (13)	.7
CMV DNA negative and serology positive	8.51 (4)	10.0 (2)	1.0a
CMV DNA or serology positive	68.1 (32)	75.0 (15)	.6
Probability of infection at 1 y	0.88 ± 0.06	0.79 ± 0.10	.5b
Time to CMV infection, mo, median ± SE	3.74 ± 0.14	3.90 ± 1.60	.9b

Data are percentage (no.) of subjects or mean value ± SE, unless otherwise indicated. All estimates are presented as intent-to-treat.

Abbreviations: HIV, human immunodeficiency virus; SE, standard error.

^a By the Fisher exact test.

^b By the z test.

^c Data are for children for whom the last CMV test was performed at the time of or after the last HIV-negative test and who never tested HIV positive.

The median time to CMV infection was 4.26 months in the breastfeeding arm and 9.87 months in the formula-feeding arm (P < .001; Figure 2), and the probability of CMV infection at 1 year was 0.89 in the breastfeeding arm and 0.69 in the formula feeding arm (P < .001; Table 2). In a Cox proportional hazards model, breastfeeding was associated with a 1.67-fold increased risk of CMV acquisition (hazard ratio [HR], 1.67; 95% confidence interval [CI], 1.24–2.23; P = .001), and this was independent of time-varying HIV infection status (multivariable HR, 1.61; 95% CI, 1.20–2.16; P = .002).

Figure 2.

Infant cytomegalovirus (CMV) acquisition. Kaplan–Meier survival functions show the probability of infant CMV infection during the first year of life in breastfed (solid line) and formula-fed (dashed line) infants, with right censoring at death or study exit. Abbreviation: HIV, human immunodeficiency virus.

Considering only the HIV-exposed uninfected infants, breastfed infants were significantly more likely than formula-fed infants to acquire CMV during the first year of life (86.7% vs 57.7%; P < .001). In addition, as seen in the cohort overall, time to CMV infection was significantly shorter for HIV-exposed uninfected infants in the breastfeeding arm, compared with those in the formula-feeding arm (P = .001, by the log-rank test), with median times to CMV infection of 5.80 months versus 10.6 months (P < .001) and probabilities of infection at 1 year of 0.89 vs 0.66, respectively (P < .001; Figure 2).

Sixty-seven infants who were tested for CMV became HIV infected during the study; 47 were in the breastfeeding arm, and 20 were in the formula-feeding arm. The proportion of these HIV-infected infants that were CMV infected (as determined by DNA detection or serology) was similar between randomization arms (Table 2). HIV-infected infants in both the breastfeeding arm and the formula-feeding arm acquired CMV infection rapidly after birth, with median times to CMV infection of 3.74 months and 3.90 months, respectively (P = .9). Both arms had similarly high rates of CMV infection at 1 year (0.88 in the breastfeeding arm vs 0.79 in the formula-feeding arm; P = .5; Figure 2).

Because of the high rate of treatment crossover in the formula arm of the study, all analyses were also performed classifying infants into feeding categories as-treated. This resulted in 165 infants being classified as breastfed and 102 being classified as formula fed. Results of these analyses were similar to those of the intent-to-treat analyses (data not shown).

Modes of CMV Transmission

The Kaplan–Meier models demonstrate nearly equivalent rates of CMV acquisition in infants from the 2 randomization arms during the first 6 weeks of life, followed by a plateau with few new infections diagnosed between 6–14 weeks and a clear divergence between the 2 feeding arms at 14 weeks (Figure 2). Among HIV-exposed uninfected breastfed infants assessed during the first year of life, we estimated that 33.3% of CMV infections occurred peripartum, based on detection of CMV by 6 weeks of life. Another 37.7% was estimated to have occurred through breastfeeding, based on differences between breastfeeding and formula-feeding arms. We estimated that another 29.0% occurred through non–breast milk postpartum modes, such as saliva and urine, based on new infections after 6 weeks in infants who never breastfed (Table 3). Among HIV-infected breastfed infants, we estimated that 36.7% of CMV infections occurred peripartum, 43.3% occurred through breastfeeding, and 20.0% occurred through non–breast milk modes.

Table 3.

Relative Modes of Cytomegalovirus (CMV) Acquisition

Mode	Breastfeeding Arm, %	Formula-Feeding Arm, %
HIV-exposed uninfected infants
Peripartum	33.3	33.3
Breast milk	37.7	NA
Non–breast milk	29.0	66.7
HIV-infected infantsa
Peripartum	36.7	45.5
Breast milk	43.3	NA
Non–breast milk	20.0	54.5

Data are from as-treated analysis.

Abbreviations: HIV, human immunodeficiency virus; NA, not applicable.

^a Data are for infants in whom HIV infection was diagnosed.

DISCUSSION

In a unique trial in which feeding modality was randomly assigned to HIV-infected women, we estimated the relative contribution of breast milk to CMV transmission. We found that breastfeeding was associated with substantially earlier acquisition of CMV and a higher rate of infant CMV infection at 1 year of age. In utero and intrapartum transmission likely accounted for a large proportion of the peripartum CMV infections observed. These data highlight the potential challenges in preventing CMV transmission to HIV-exposed infants.

Breastfeeding conferred a 61% increase in CMV infection risk, after controlling for infant HIV status, and breastfed infants acquired CMV nearly 6 months earlier than formula-fed infants. Because of our interval sampling, we allowed an upper boundary of 6 weeks to define peripartum transmissions but were unable to discriminate in utero and intrapartum infections; the equivalence of our Kaplan–Meier transmission models until the week 6 time point suggests that there were few breast milk–associated transmissions detected during this time frame, after which CMV infections were observed at a high rate in the breastfed arm. Kumar et al similarly found that most CMV transmissions in a cohort of nonbreastfed infants occurred before 14 weeks [14]. The waning of maternal antibodies may explain the large increase in infections detected at 14 weeks; this pattern could thus be explained by an increase in susceptibility or unmasking of suppressed virus. We are unable to evaluate this with our current cohort sample archive, but this question has significance for understanding the immune mechanism of protection. Among HIV-infected infants, we did not observe a difference in CMV transmission by feeding modality, and maternal and/or infant mechanisms may explain this observation. Immunosuppressed mothers are more likely to transmit HIV and to be shedding CMV from multiple mucosal sites [21, 24], resulting in proportionally more transmissions via other routes. Infant immunosuppression may leave HIV-infected infants highly susceptible to all modes of CMV acquisition. Additionally, some infants acquired HIV before CMV and vice versa, which could impact our ability to discern a difference between arms in this stratum.

The large contribution of peripartum transmissions may reflect the high frequency of CMV reactivation in the cervical mucosa of HIV-infected women during their third trimester [21], which is associated with intrapartum transmission [12]. There was a high rate of in utero CMV acquisition in HIV-exposed uninfected infants in this cohort, consistent with earlier studies in HIV-infected cohorts from Kenya and elsewhere (3%–11%) [9, 25, 26]. We did not detect any in utero CMV transmissions in the HIV-infected infants, which does not suggest an absence of in utero transmission in this group, but is more likely due to the small number of infants sampled at delivery. We thus cannot rule out a large contribution of in utero CMV transmissions in those detected at ≤6 weeks of age, particularly among HIV-infected infants, who are at higher risk for in utero CMV infection [6, 9].

The rate of CMV transmission was also high in the infants randomized to formula, reaching 70% at 1 year. These infections were nearly equally distributed between peripartum and late transmissions. Peripartum CMV transmission rates were similar between breastfed and formula-fed infants, and overall transmission rates were similar in both our intent-to-treat and as-treated analyses, suggesting that the high rate of CMV transmission in the formula-feeding arm was not due to misclassification of feeding modality, but rather to the acquisition of CMV via non–breast milk modes of transmission. Infections can originate from other contacts from both within and outside the household [27], and because we did not sequence CMV in this study, we are unable to definitively link transmissions between mothers and infants.

Although the rate of CMV infection was high among breastfed infants, it was lower than we reported previously in other cohorts (95% in HIV-exposed uninfected infants and 88%–100% in HIV-infected infants) [9]. We used DBS for CMV diagnosis in this study, which has a lower sensitivity than plasma at low viral loads. Use of serology to screen CMV DNA–negative infants enabled us to identify most infants with false-negative test results, but we were unable to verify final infection status in 18 formula-fed and 19 breastfed infants because they lacked plasma specimens after 6 months of age. Although we underestimate the true burden of CMV infection in the cohort, the very high rate of transmission enabled us to detect a difference between study arms despite some misclassification of outcome.

In summary, breastfeeding conferred earlier CMV transmission and a higher risk of transmission between HIV-infected women and their infants. The significance of early CMV acquisition for outcomes in HIV-exposed uninfected infants is not yet clear, although there is increasing evidence to suggest that HIV-exposed infants would benefit from a CMV vaccine. The high rate of early transmissions underscores the challenges for vaccine deployment in this population. An alternative approach to preventing or delaying CMV infection in HIV-exposed infants may be to reduce CMV replication in the maternal breast milk and/or genital tract. In HIV-infected women, maternal immunosuppression is associated with a higher risk of infant CMV infection [21], suggesting maternal ART could reduce CMV transmission, but to date, only a small number of studies have examined this, with variable findings [26, 28–30], and significant impacts on breast milk or cervical reactivation have not yet been demonstrated. As World Health Organization Option B+ PMTCT of HIV rolls out in sub-Saharan Africa and more women become pregnant while already receiving ART, it will be important to determine whether longer-term maternal HIV treatment changes this dynamic or whether additional measures to directly reduce CMV shedding are warranted.