Neonatal erythropoiesis and subsequent anemia in HIV-positive and HIV-negative Zimbabwean babies during the first year of life: a longitudinal study

Subjects were drawn from participants of ZVITAMBO, which enrolled 14,110 mother-baby pairs within 96 hours of delivery from maternity clinics and hospitals in Harare between November 1997 and January 2000 to study the impact of maternal and neonatal vitamin A supplementation on several outcomes [18‐20]. Pairs were eligible if neither the mother nor the baby had an acutely life threatening condition and the infant was a singleton with birthweight >1500 g. Written informed consent was obtained from the mother. At the time the study was carried out, neither vitamin A supplementation nor HIV antiretroviral therapy was part of routine perinatal or postnatal care. Among infants enrolled beginning in October 1998, a 34% sub-sample (seven of every eight born to HIV-positive mothers and one of every ten born to HIV-negative mothers) was randomly selected for the anemia substudy, to yield an adequate sample size among HIV-positive and HIV-negative infants to detect a 5 g/L difference in hemoglobin concentration. Babies who had received a blood transfusion during the neonatal period were excluded from the anemia substudy. Only babies assigned to the placebo control group (i.e., both mother and baby received placebo) were included in the present analysis for a total of 136 and 575 babies born to HIV-negative and HIV-positive mothers, respectively. The ZVITAMBO trial protocol was approved by the Medical Research Council of Zimbabwe (MRC-Z), the Committee on Human Research of The Johns Hopkins Bloomberg School of Public Health (CHR-JHBSPH), the Medicines Control Authority of Zimbabwe, and the Montreal General Hospital Research Ethics Committee. The MRC-Z and the CHR-JHBSPH provided further approval to conduct additional laboratory analyses for this anemia substudy.

At recruitment, demographic and obstetric information was collected by maternal history or transcription from the medical record. Gestational age was estimated using the method of Capurro [21]. Birthweight was measured within 96 hours of delivery to the nearest 5 g using an electronic scale (Seca model 727, Hanover, MD). Mothers and babies were followed at 6 weeks, 3 months and thereafter 3 monthly up to 24 months of age in a study clinic or at home. Detailed infant feeding information, including breast-feeding status and whether or not any of 22 nonmilk liquids, nonhuman milks (animal milks and commercial formula), medicines (traditional fluids, oral rehydration salts, other prescribed), or solid foods had ever been given to the infant were collected at enrolment, 6 weeks, 3 months, and 6 months after delivery.

Medical care, including treatment of acute illness, and counseling for pre- and post- HIV test, psycho-social support, and infant feeding, were offered throughout the trial; travel costs to the follow-up clinic were reimbursed.

At baseline and each follow-up visit, study midwives collected maternal and infant blood by venipuncture or heelstick (capillary) into clot and EDTA tubes. Maternal and infant serum and plasma were prepared and stored at -70°C within two hours of phlebotomy. Aliquots of infant whole blood were refrigerated and cell pellets prepared (Amplicor whole blood PCR sample preparation method; Roche Diagnostics Systems, Alameda, CA, USA) generally within a few days of collection and stored at -70°C. Aliquots of baseline maternal serum were refrigerated.

Refrigerated maternal serum was tested for HIV status by two ELISA assays for detecting antibody to both HIV 1 and 2 [GeneScreen (Sanofi Diagnostics Pasteur PRx, Johannesburg, South Africa) and Murex (Murex Diagnostics, Johannesburg, South Africa)] run in parallel. Discordant ELISA results were resolved by Western Blot. CD4 cells from maternal baseline samples that tested HIV positive were enumerated using a Facscount (Becton Dickinson International, Erembodegem, Belgium) within 48 hours of phlebotomy.

After all patient contact was completed, the last available infant specimen was tested for HIV (plasma by ELISA for samples collected at >18 months; pellets by Amplicor HIV-1 DNA test version 1.5 (Roche Diagnostic Systems) for samples collected at younger ages). If this sample was HIV-negative, the baby was classified as negative and no further testing was carried out. If the last sample was HIV-positive, samples collected at intervening time points were tested to determine the timing of infection.

Hemoglobin was measured on whole blood samples on the day of collection for all mothers at baseline and for all infants in the anemia substudy at baseline and visits up to 12 months using the HemoCue hemoglobinometer (HemoCue, Mission Viejo, CA). Daily quality control of the instrument was performed. Hemoglobin data were unavailable if the baby missed the study visit, or if the blood sample was insufficient as a result of the mother's refusal for her baby to give blood, inadequate sample volume, sample clotting, or failure by the laboratory to perform the test before the end of the day in which the specimen had been drawn.

We measured TfR, ferritin and EPO concentrations in plasma (or serum when plasma was unavailable). TfR concentrations were measured by enzyme immunoassay (Ramco Laboratories, Houston, TX) (intra-assay coefficient of variation, 3.9% at 6.7 mg/L and 2.7% at 13.2 mg/L; inter-assay coefficient of variation, 8.8% and 6.6%, respectively); ferritin by enzyme-linked immunoassay (Ramco Laboratories) (intraassay coefficient of variation, 4.0% at 70 μg/L and 4.7% at 300 μg/L; inter-assay coefficient of variation, 9.6% and 9.2%, respectively); and EPO using the Quantikine In Vitro Diagnostic EPO ELISA (R&D Systems, Minneapolis, MN) (intra-assay coefficient of variation, 2.3% at 42 IU/L and 2.4 at 13 IU/L; inter-assay coefficient of variation, 5.2% and 3.4%, respectively).

The criteria we used to diagnose anemia were modified from those of the World Health Organization [22]. Several studies have suggested that during infancy, the WHO cut-off value of 110 g/L may overestimate anemia [23‐25]; therefore, we used a more stringent cut-off value of 105 g/L [26]. We did not define anemia in infants younger than 3 mo of age, because hemoglobin does not reflect iron status in early infancy but reflects a normal physiologic decline in concentration. Low plasma ferritin concentration is considered to be an accurate indicator of iron stores in the liver [27, 28]. We used the WHO standard of 12 μg/L to indicate absent or depleted iron stores in infants aged up to 6 months.

Plasma EPO, TfR, and ferritin concentrations were normalized by log transformation; geometric means (95% CI) are reported. Differences across HIV status groups in baseline characteristics and indicators of anemia and iron status were examined by ANOVA for continuous variables and the chi-squared test of homogeneity for categorical variables. We used multivariate linear and logistic regression analysis to compare hemoglobin values and the proportion of anemic children across HIV status groups and adjusted for birthweight in these models. The association between indicators of anemia and iron status was examined by linear regression and partial correlation coefficients. The relevant interaction terms between the independent variable and HIV status groups were included in regression models to examine differences in the association between indicators by HIV status group. Adjusting for birthweight in these models did not change the conclusions, and, therefore, was not included. We performed a likelihood ratio test to determine the overall statistical significance of the interaction between HIV status and the independent variable. For all analyses, we set the significance levels at α = 0.05. A Bonferroni correction was used to determine which associations were unlikely to be the result of chance alone.

We created three categories of babies: the 136 babies born to HIV-negative mothers (Nn group) and two groups of babies born to HIV-positive mothers: (1) babies who tested PCR negative at 6 months or later (Pn group, n = 324); and (2) babies who tested PCR-positive at baseline or 6 weeks (Pp group, n = 99). Babies who did not fall neatly into these three groups were excluded from the analysis. These included those who tested PCR-positive for the first time at 3 or 6 months (n = 37), and babies with an indeterminate HIV status (n = 115). Babies were classified as HIV-indeterminate if maternal baseline HIV status was unknown (n = 2) or maternal baseline status was positive and no results were available for the baby (n = 2), the baby's last available PCR result was prior to 6 months of age and was negative (n = 108), or the baby's PCR results were inconsistent such that a negative result followed a positive result (n = 3). Only 4 infants in the Nn group and 3 in the Pn group died by 12 months of age, but among Pp infants, 33 (33.3%) had died by 6 months of age and 51 (51.5%) by 12 months.

We carried out the analysis only on the available data, and did not impute values for missing observations. Hemoglobin values >160 g/L at 6 weeks (n = 1), 3 months (n = 1) and 6 months (n = 4) were excluded from the analysis. All analyses were performed with the use of Stata (verson 8.0, Stata Corporation, College Station, TX).

The only difference in indicators of anemia and iron status between Nn and Pn babies was a significantly higher TfR concentration at 6 weeks of age in Pn babies (P = 0.001; P = 0.003, Bonferroni adjusted) (Table 2). In both HIV-infected and uninfected babies, mean hemoglobin concentration decreased dramatically from birth to 6 weeks, in a pattern consistent with the postnatal physiologic decline, but did not rise again (Figure 2). HIV-infected babies had the lowest 6 week concentration even after adjusting for birthweight (mean ± SE; 103 ± 3.0 g/L), significantly lower than that of babies born to HIV-negative mothers (114 ± 3.1 g/L, P = 0.008; P = 0.024, Bonferroni adjusted). From this nadir, hemoglobin slightly increased, but the improvement was not sustained beyond 3 months of age, after which hemoglobin progressively declined to significantly lower values than either Nn or Pn babies at 9 and 12 months of age. By 6 months, the odds of anemia (hemoglobin < 105 g/L) was more than two-fold greater among HIV-positive babies compared to HIV-negative babies (59.8% vs. 36.6%, adjusted for birthweight, P = 0.021). At 12 months, 36.6% of HIV-negative babies were anemic and 75.9% of HIV-positive babies (adjusted odds ratio = 5.45 (95% CI: 2.22, 13.4; P < 0.001) were anemic.

Plasma ferritin concentration decreased from birth levels as expected with the physiologic mobilization of iron from liver stores in all HIV status groups. However, at all time points, including birth, the ferritin distribution was shifted to higher values among HIV-infected babies. For example, at 3 months, geometric mean [95% CI] ferritin concentration was significantly greater among HIV-positive infants (140.7 [103.8, 190.7], n = 52) compared to HIV-negative infants (59.8 [54.5,65.6], n = 279; P < 0.0001). At 6 months, serum ferritin was not associated with hemoglobin in any HIV status group. Geometric mean plasma EPO increased with age and TfR increased to 3 months then leveled off (see Table 2). Compared to HIV-negative babies, mean EPO concentration was significantly higher among HIV-positive infants at 6 weeks, 3 months, and 6 months, and mean TfR was higher at 6 weeks and 3 months.

Among Nn and Pn infants, the slope of the fitted regression line between log₁₀ plasma EPO (U/L) and hemoglobin (i.e., the EPO response to anemia) was significantly negative at 6 weeks; this association remained at 3 months and 6 months but was only significant among the Pn infants (see Additional file 1). Among Pp infants, the EPO response to anemia was attenuated at 6 weeks and 3 months, but significant at 6 months. Among Nn babies, the EPO response to anemia became weaker with age, while among Pn and Pp babies, the response was stronger at 6 months than at 6 weeks or 3 months.

TfR and EPO were positively associated at all ages and for all HIV status groups. A significant interaction with HIV status was observed at 3 months of age. The increase in log_e plasma TfR (mg/L) for one log₁₀ decrease in plasma EPO (U/L) was significantly greater for Pp babies compared to Pn babies (P = 0.023), and there was a trend for a greater increase when compared to Nn babies (P = 0.11).

There was no consistent or significant relation between ferritin and TfR at 6 weeks or 3 months. However, there was a significant interaction between HIV status and TfR on ferritin concentration at 6 months (P = 0.01). Ferritin was inversely associated with TfR in HIV-negative infants born to HIV-positive and HIV-negative mothers but not in HIV-positive infants.

To compare the relative strength of association between TfR and ferritin vs. EPO, we examined the partial correlation coefficients between these variables. TfR was more strongly correlated with EPO than ferritin in all HIV status groups at every age, after adjusting for hemoglobin. There was a strong inverse association between TfR and ferritin only at 6 months in Pn infants, but, even then, more of the variability in TfR was explained by EPO concentration than ferritin.

One-third of HIV-negative infants had serum ferritin concentration <12 μg/L at 6 months suggesting that their anemia was largely caused by iron deficiency. Worldwide, the primary cause of anemia among infants is iron deficiency [29]. Infants have a relatively high demand for iron because they are rapidly growing, and human milk is low in iron, even after accounting for good bioavailability. In most developing countries, complementary food diets are largely plant-based and high in substances that inhibit iron absorption and are by themselves insufficient to meet dietary needs for iron [30‐32]. Foods with naturally high iron content and high bioavailability, such as meat, and iron-fortified infant formula are rarely consumed. We too observed these infant-feeding patterns in the present study. Furthermore, ~one-third of the mothers of HIV-uninfected infants were anemic (hemoglobin <110 g/L) at delivery, which is also a risk factor for infant anemia [33, 34].

In the absence of concurrent disease, serum ferritin parallels total body iron stores and can reflect deficient, normal, and excess iron stores [27]. However, ferritin is also a positive acute phase protein and can rise in the presence of infection or inflammation [35]. Among the HIV-positive infants in this study, we conclude that serum ferritin more strongly reflected concurrent infection than iron status because: a) it was consistently higher among HIV-infected infants compared to HIV-negative infants; and, b) it was inversely associated with TfR among uninfected infants but not among HIV positive infants. A measure of immune activation would have provided further clarification and should be incorporated into future studies.

Our findings that the EPO response to anemia was blunted in HIV-infected babies at 6 weeks and slightly attenuated at 3 months of age are consistent with several reports of studies carried out in HIV-infected adults [6, 9, 36]. This attenuation is thought to be through inhibition by inflammatory cytokines, such as interleukin-1 and tumor necrosis factor, or EPO gene transcription [10, 37, 38]. However, by 6 months of age, the EPO response to anemia was not different than that observed among uninfected infants. In fact, mean log₁₀ EPO concentration was highest in Pp infants at all ages, even after adjusting for hemoglobin concentration (data not shown). To our knowledge, the only other study, in addition to the present, to compare EPO in HIV-infected and uninfected infants was carried out in Malawi, and they too did not show an inadequate EPO response among HIV-infected infants [11]. It is noteworthy that the infants in the Malawi study were 12 months-old. They, like the babies in the present study at 6 months of age, represent a different stage in physiological development than 6 week old infants. It is not surprising that the age of the infant plays a crucial role in the regulation of erythropoiesis and perturbation to the system. These ages span highly significant changes in normal hematological development characterized by a switch in the site of EPO production from the liver to the kidney, greater oxygen consumption with increasing activity and growth, and a switch from fetal to adult hemoglobin with a decrease in oxygen affinity favoring oxygen delivery.

The apparent transient impact of HIV on the EPO response to anemia might suggest that the attenuating effect on EPO is more pronounced closer to the time of initial infection and weakens with progression of HIV disease. Acute HIV infection is characterized by a peak in viremia and rapid decline in CD4+ cells, but an immunologic response ensues resulting in a lowering of the plasma level of HIV RNA to a steady state "set point" that reflects a balance between production and destruction of virions [39]. It is possible that we have observed this steady state in HIV infected infants at 6 months of age. Alternatively, the higher mean EPO levels in the Pp group may be explained by dyserythropoiesis (due to a direct effect of HIV and/or through cytokine activation) or reduced number of stem cells in HIV infected infants. However, we observed a strong relationship between EPO and TfR suggesting that bone marrow cells are responding to stimulation by EPO. A measure of reticulocyte count would be helpful to clarify this mechanism and should be considered in future studies of infant hematopoiesis. Finally, we speculate that the stronger EPO response observed in Pp infants at 6 months may in part be attributed to a survival bias. We compared the EPO-Hb slope among Pp infants who died by 12 months of age to that of Pp infants who survived to 12 months. Only 2 infants who died in the first year had both EPO and Hb measurements at 6 months of age making it impossible to estimate the slope in this group at 6 months. A comparison of the slopes at 3 months of age showed a trend for a flatter slope among those who died (+0.0072) than among those who survived (-0.0017) suggesting greater attenuation of the EPO-Hb association among those who died and, perhaps, unique resilience among those who survived. On the other hand, the slope at 6 weeks was steeper among infants who died (-0.0042) than among those who survived (-0.0025) and more like that observed among Pn infants (-0.0044). This post hoc analysis was limited by sample size, and differences among HIV-infected infants who survive and among those who die should be further explored.

It is important to note that the inverse association between EPO and hemoglobin was not statistically significant in Nn babies at 3 and 6 months of age. The failure to reach statistical significance may in part be attributed to a lack of sample size since slopes of similar magnitude in the Pn group were statistically significant. However, the data also suggest a relatively weak inverse association between these indicators. In fact, the observed slopes of the regression line between log₁₀ EPO and hemoglobin (g/L), even the steepest observed value (-0.0109) in Pp infants at 6 months, were less steep than values previously reported in the literature. The values of slopes for control groups comprised of HIV-uninfected infants [11], children with iron deficiency anemia [40] and sickle cell disease [41], and children with hematological disease [42] have ranged from -0.011 to -0.0553. Young children in developing countries are often sick, with the prevalence of diarrhea ranging from 10 to 19 percent [43‐45], and the prevalence of acute respiratory illness among young children in developing countries even higher, 25 to 60 percent [46]. We speculate that the inflammatory response to common childhood infections may have caused an attenuation of the EPO response to anemia. Additionally, vitamin A has been shown to stimulate EPO production in cell culture and animal models [47‐50]. Thus, it is possible that vitamin A deficiency might have reduced the EPO response to low hemoglobin concentrations.

A noteworthy effect of HIV infection on erythropoiesis was the modification of infant infection on the association between ferritin and TfR at 6 months of age. The absence of a negative association between TfR and serum ferritin in HIV-infected babies may be explained by confounding due to infection, yet it is consistent with the sequestration of iron in liver stores during infection. The concurrent lack of association at 6 months between ferritin and hemoglobin supports this explanation since an increase in ferritin did not translate into increased hemoglobin production. However, these results should be interpreted with caution since serum ferritin is an acute phase protein and elevated during infection and inflammation [35] at which times the indicator has limited ability to accurately reflect iron stores. The lack of association between ferritin and hemoglobin that we observed at 6 months in all HIV status groups supports the notion that ferritin levels do not specifically reflect iron stores but are also affected by infection status.

The observation that at all ages and in all HIV groups, EPO concentration explained more variability in TfR levels than did serum ferritin concentration suggests that TfR is not primarily an indicator of iron deficiency in this age group. We speculate that during periods of high erythropoietic activity (e.g. early infancy), an increase in TfR more strongly reflects an increase in erythroblasts (i.e. erythropoiesis) rather than an increase in the density of receptors per red blood cell consistent with tissue iron deficiency. While the strong and consistent association between TfR and EPO is striking, particularly when compared to the lack of association between TfR and ferritin, it was not surprising that ferritin did not explain much of the variability in TfR given its association with infection and inflammation.