Cytomegalovirus Infection in Gambian Infants Leads to Profound CD8 T-Cell Differentiation^▿

Subjects.

Subjects were recruited at birth from the maternity ward of Sukuta Health Centre as part of an ongoing birth cohort. Informed consent was obtained from their mothers and documented by signature or thumb print. Recruitment was restricted to healthy singleton infants, defined by a birth weight of at least 1.95 kg and no congenital abnormalities. The catchment area of Sukuta Health Centre is periurban and characterized by low income and crowded living conditions. The human immunodeficiency virus statuses of the study subjects were unknown, but adult prevalence in the region was 1 to 4% at the time of the study (National AIDS Control Program, unpublished data) and so was not thought to be a significant confounder of this study.

Mononuclear cells, plasma, and buffy coats were separated from umbilical cord blood and stored. Limited molecular human leukocyte antigen (HLA) typing was carried out on the buffy coats (4) in order to identify subjects with HLA-A2, -B7, -B8, and -B35, for which known epitope peptides were available.

In total, 258 infants were followed up for 12 months to establish the age at CMV infection. A subset of 116 infants was selected for studies on the development of the CD8 T-cell response to CMV, which was biased toward HLA types for which peptides were available.

The prevalence of CMV in the adult population was established by serological detection of CMV-specific immunoglobulin G (IgG) in a subsample of 194 mothers, using ELICYTOK-G enzyme-linked immunosorbent assay (Diasorin). All samples were seropositive.

The study was approved by the Medical Research Council-Gambian Government Ethics Committee.

Diagnosis of CMV infection and blood-sampling schedule.

To determine the onset of CMV infection, urine samples were collected at birth, fortnightly for the first 3 months of life, and monthly until the age of 12 months and tested for the presence of CMV DNA by PCR. If it was detected, a second sample was taken immediately to confirm infection, and urine sampling was discontinued. Excretion in the urine is usually continuous for at least 6 months following infection (D. J. C. Miles and S. Kaye, unpublished data). Congenital infection was defined by CMV detection in a urine sample collected in the first 2 weeks of life.

Following CMV diagnosis, a blood sample was taken as soon as possible and follow-up samples were scheduled for 4 and 12 months after diagnosis. Samples were collected in heparin and transported to the laboratory within 4 h. If an infant reached 12 months of age without CMV DNA being detected, a blood sample was collected at that time.

Whole blood was used for flow cytometry, and peripheral blood mononuclear cells (PBMCs) were separated on lymphoprep (Axis-Shield PoC AS) columns for use in enzyme-linked immunospot (ELISpot) assays.

Flow cytometry.

All flow cytometry reagents were from Becton Dickinson. The CD8 population was identified using peridinin chlorophyll-conjugated anti-CD8 antibody. Preliminary experiments demonstrated that it was possible to identify a pure population of CD3⁺ CD8^high T cells and that cells binding tetramers expressed CD8 at the same high level. Further phenotyping for surface markers was carried out using allophycocyanate-conjugated antibodies to CD27, CD45R0, and CD62L and fluorescein isothiocyanate-conjugated antibodies to CD28, CD45RA, CD95, and the class II HLA-D molecules DR, DP, and DQ. Intracellular markers were identified using fluorescein isothiocyanate-conjugated antibodies to Bcl-2, Ki-67, perforin, and granzyme A. All antibodies were titrated prior to use to ensure that they were used at their optimal concentrations. Where possible, CMV-specific cells were identified with phycoerythrin-conjugated B7-TPRVGGGAM and A2-NLVPMVATV tetramers, synthesized as previously described (13, 27). New batches of tetramer were initially used in parallel with older batches to ensure consistent staining. Red blood cells were lysed using FACSlyse solution, and lymphocytes were permeabilized using FACSperm II.

Statistical analysis of flow cytometry data.

As the study involved the collection of over 1,800 flow cytometry samples over a prolonged period of time, and the study design involved both cross-sectional and longitudinal data, objectively defining subpopulations of interest was a major challenge. In order to eliminate the inconsistencies and possible bias inherent in a human operator repetitively analyzing large amounts of data, automatic algorithms were used to define clusters for various combinations of markers using MATLAB 6.1 (The Mathworks Inc). For each marker set, the same algorithm was applied to each subject to maintain objectivity and consistency of the flow cytometry analysis. As no objective criteria exist for the definition of the subpopulations described in the present study, we made the assumption that the algorithms were valid if they recognized the same subpopulations as an experienced flow cytometry operator.

Lymphocyte populations were identified using a combination of contouring and morphological operators on the pixels of the front versus side scatter plot. After gating this population, similar algorithms were used to identify CD8 T-cell populations and tetramer⁺ cells. This part of the process was tested on a small number of samples stained for CD3 and CD8, and it was established that while there was a substantial overlap between the expression of CD8 on CD3⁻ and CD3⁺ cells, >90% of the CD8⁺ cells identified were CD3⁺ T cells, and this population included an average of 72% of all CD8⁺ CD3⁺ cells (data not shown).

Cells stained with CD27 and CD28, CD27 and perforin, and CD27 and granzyme A were divided into three subpopulations, and cells stained with CD95 and CD62L were divided into two subpopulations using a fuzzy c-means clustering algorithm. The extremities of the population region were defined by the furthest outliers.

Cells expressing Ki-67 and cells expressing either high or low levels of Bcl-2 were identified using morphological operators to locate a region of highest event density. Where this was not possible, a weighted average of the events was used to separate the two regions.

For cells stained with HLA-D and CD45R0, an area of highest event density was identified where possible using algorithms similar to those used for Ki-67 and Bcl2, and the populations were then classified using a cross-hair placed with horizontal and vertical positions defined by the extreme borders of the densest region.

For cells stained with CD45RA and CD45R0, a breakpoint regression through the smoothed trend of CD45RA expression was used to identify at which point only CD45RA was expressed, thus defining a CD45RA^bright CD45R0⁻ population. To further explore the phenotype of the cells as defined by the expression of CD45 isoforms, a line of regression was plotted through the population of cells expressing both CD45RA and CD45R0, and the midpoint was calculated. A “CD45 ratio” was calculated by dividing the distance from the midpoint of the regression line to the border with the CD45RA^bright CD45R0⁻ subpopulation by the distance from the midpoint to the event with the lowest expression of CD45RA as defined by fluorescence intensity. A relatively high value for the CD45 ratio indicated a high level of expression of CD45R0 and a low level of expression of CD45RA.

A conservative approach was used for clustering definition, and for each algorithm limits were set to define a successful and meaningful cluster/region definition. Of the 1,886 samples scanned, 130 (6.9%) were discarded because they could not be characterized after an iterative process in which scans were manually checked to identify obvious algorithm failure.

For each set of markers, appropriate region statistics were defined, and depending on their distribution, they were summarized using means and medians. The within-subject repeated measurements were accounted for using panel-averaged models (fitted using the generalized estimating equation function in Stata 8). Where necessary, the region statistics were appropriately transformed to improve the normality and constant variance assumptions. Due to the small number of time points, the most appropriate correlation structure was the exchangeable model, which acknowledges the correlation but assumes it is the same between all time points. The significance of differences over time was quantified by a Wald test, and if significant at the 5% level, appropriate contrasts were calculated and P values were adjusted using a step-down Bonferroni method.

Diagnosis of EBV infection.

A subset of plasma samples from 12-month-old infants was tested for Epstein-Barr virus (EBV) seropositivity. To allow for the persistence of maternal IgG, an infant was regarded as infected if he/she expressed IgM to the viral capsid antigen (VCA) at 12 months or a higher level of IgG to the VCA antigen than was present in umbilical cord plasma collected. The tests were carried out using the ETI-VCA-G and ETI-EBV-M enzyme-linked immunosorbent assays (Diasorin).

ELISpot.

A kit for the detection of IFN-γ (Mabtech) was used in ELISpot assays. The peptides used were the A2-restricted MLNIPSINV, NLVPMVATV, VLGPISGHV, YILEETSVM, and IAGNSAYEYV; the B7-restricted RPHERNGFTV, TPRVTGGGAM, and CRVLCCYVL; the B35-restricted IPSINVHHY and VFPTKDVAL; and the B8-restricted DANDIYRIF (11, 22, 33, 38, 42, 47, 49).

A detectable response was defined as a mean treatment spot count that was more than 10 spots 10⁵ cells⁻¹ greater than twice the mean negative control. These criteria have previously been found to provide high specificity while allowing reasonable sensitivity in the detection of peptide-specific responses (18). Responses were measured as specific spot-forming units, calculated as the difference between the mean treatment spot count and the mean negative control spot count. Data were analyzed using Stata 8 and MINITAB 14.

Most infants are CMV infected by 12 months of age.

Congenital infection, defined as excretion of CMV in the urine within the first 2 weeks of life, was detected in 4.1% of the infants. Prevalence increased most rapidly between 6 and 12 weeks of age, by which time 62% were infected. Thereafter, prevalence continued to rise steadily until sampling concluded at 12 months, by which time 85% were infected (unpublished data).

Sampling for CD8 T-cell studies.

Longitudinal studies on the CD8 T-cell response to CMV infection were carried out on 98 infants. From these subjects, 64 samples were taken immediately after diagnosis, 92 4 months later, and 79 12 months later, so that 235 of the 294 scheduled samples were collected successfully. A further 18 samples were collected from infants who had reached 12 months of age without being infected with CMV, so the total cohort numbered 116 infants.

Frequencies of peptide-specific CD8 T cells were very variable.

Of 67 subjects who were identified as expressing HLA-A2, at least one sample containing CD8 T cells that bound the A2-NLVP tetramer was collected from 20 (30%). Of the 37 subjects who expressed HLA-B7, at least one sample that bound the B7-TPRV tetramer was collected from 16 (43%). The mean frequency of CD8 T cells that bound either tetramer was 3.4%, but there was considerable variability, with frequencies ranging from 0.05% to as high as 32%.

The age of CMV infection had no influence on the frequency of tetramer⁺ cells, and there were no significant differences in frequency between the samples collected immediately after diagnosis and those collected 4 or 12 months later (data not shown).

The phenotypes of tetramer⁺ CD8 T cells followed the same trends as the overall population.

The statistical interaction between the phenotypes of tetramer⁺ cells and the overall population was assessed to establish whether the time elapsed since diagnosis had different effects on the phenotypes of tetramer⁺ cells and the overall population. The only significant interaction found was in the case of the proportion of cells expressing Ki-67 (P = 0.015), which is discussed below. For all other parameters measured, changes in the phenotype of the tetramer⁺ cells were reflected by changes in the phenotype of the overall CD8 T-cell population, although the actual proportions of cells in different subpopulations at any given sampling time were often very different.

Age at time of diagnosis had no effect on the magnitudes of either the tetramer⁺ subpopulations or any of the subpopulations defined by phenotype (data not shown).

Activation was highest in the early stages of infection in tetramer⁺ cells and the overall CD8 T-cell population.

Cells were stained for CD45RA and CD45R0, and a breakpoint regression was used to divide them into a CD45RA^bright CD45R0⁻ subpopulation and a subpopulation expressing approximately reciprocal proportions of both isoforms. Nearly all tetramer⁺ cells expressed CD45R0 (P < 0.0001), which is typically expressed by functional memory or effector cells (37, 43), considerably more than the overall CD8 T-cell population (Table 1 and Fig. 1A and B). However, the CD45 ratio among the subpopulation that expressed both isoforms was not different between tetramer⁺ cells and the overall CD8 T-cell population (Table 1 and Fig. 1C), indicating that while a higher proportion of the tetramer⁺ subpopulation expressed CD45R0 than the overall CD8 T-cell population, the level of CD45R0 expression did not differ between these populations.

The proportion of cells expressing both isoforms was at its highest immediately postdiagnosis and was substantially lower in the samples taken 4 and 12 months later (P = 0.012) (Table 1 and Fig. 1B). The CD45 ratio showed the same pattern (P = 0.008), indicating that not only the proportion of cells that expressed CD45R0 but also the level of CD45R0 expression was higher immediately after diagnosis than in the following months.

Cells were also stained with CD45R0 and HLA-D, and the cells that expressed either or both were identified. As with CD45R0, the proportion of cells that expressed HLA-D was greater among tetramer⁺ cells than among the whole CD8 T-cell population (P < 0.0001), and expression was higher immediately after diagnosis than 4 or 12 months later (P = 0.004) (Table 1 and Fig. 1D and E). The proportion of cells expressing both CD45R0 and HLA-D molecules followed the same pattern (Table 1 and Fig. 1F).

The proportion of CD8 T cells in the mitotic cycle was highest immediately after infection.

The age at the time of diagnosis did not affect the proportion of cells expressing Ki-67 (i.e., in the mitotic cycle) (12) at any of the sampling times (data not shown).

Mitosis was the only variable for which the development of the tetramer⁺ population followed a different pattern than the overall CD8 T-cell population. Immediately after diagnosis, the tetramer⁺ cells exhibited significantly more mitosis than in either of the two subsequent samples (P = 0.0006) (Table 1 and Fig. 2A and B), and more tetramer⁺ cells were undergoing mitosis than in the overall CD8 T-cell population (P < 0.0001). However, the proportion of tetramer⁺ cells in mitosis halved from 31% immediately after diagnosis to 15% 4 months later and ceased to be different from the overall CD8 T-cell population.

Among the overall CD8 T-cell population, the highest proportion of CD8 T cells in mitosis never exceeded 12%, and there were no significant differences between the proportion of cells expressing Ki-67 at different sampling times (Table 1 and Fig. 2B).

Expression of CD62L, CD95, and Bcl-2 remained constant in the year following infection with CMV.

The CD8 T-cell population was divided into an undifferentiated CD62L⁺ subpopulation (20, 43), expressing various levels of CD95 (Fas), and a differentiated CD62L⁻ subpopulation, in which nearly all cells expressed CD95 (Fig. 3A and B). Nearly all of the tetramer⁺ cells were in the latter subpopulation (P < 0.0001). Differentiation status could also be defined by high and low levels of Bcl-2 expression, and nearly all tetramer⁺ cells were in the more differentiated Bcl-2^low subpopulation (P < 0.0001) (23) (Table 1 and Fig. 3C and D).

In contrast to CD45 isoforms, HLA-D, and Ki-67, the relative sizes of the subpopulations defined by CD62L and Bcl-2 remained constant for 12 months following diagnosis in both the tetramer⁺ cells and the overall CD8 T-cell population.

The proportion of differentiated CD8 T cells is constant in the year following CMV infection, but CD27 is lost progressively.

Cells stained for the costimulatory molecules CD27 and CD28 were divided into naïve (CD27⁺ CD28⁺), intermediate (CD27⁺ CD28⁻), and fully differentiated (CD27⁻ CD28⁻) subpopulations (1, 16, 43). Few tetramer⁺ cells expressed any CD28, so they nearly all fell into the last two subpopulations (Table 1 and Fig. 4A). Among the CD8 T-cell population as a whole, cells in the intermediate and fully differentiated populations nearly all expressed perforin (Table 1 and Fig. 4B) and granzyme A (data not shown).

Age at diagnosis did not affect the relative distribution of cells between the three populations or their subsequent changes in phenotype. Similarly, the proportion of cells expressing the naïve phenotype remained constant in the 12 months following diagnosis. However, there was a progressive, albeit nonsignificant, reduction of the intermediate population and a more pronounced increase in the fully differentiated population among both the tetramer⁺ cells and the overall CD8 T-cell population (P = 0.0003), indicating loss of CD27 expression from the differentiated CD8 T-cell population (Table 1 and Fig. 4A).

The changes that were apparent when the populations were defined by CD27 and CD28 expression remained when the stain for CD28 was replaced with stains for perforin (Table 1 and Fig. 4B) or granzyme A (data not shown).

Production of IFN-γ by peptide-specific cells is impaired in younger infants.

The IFN-γ production of the PBMCs of samples that responded to tetramers was assessed by ELISpot assay, using the peptide incorporated into the tetramer as the stimulant. The proportion of subjects in which a response was detected was higher in the samples collected 12 months after diagnosis (16/19) than in those collected immediately (6/11) or 4 months (2/11) after diagnosis, though the only significant difference was between the 4- and 12-month postdiagnosis responses (Fisher's exact probability test; P = 0.0012). As frequencies of CMV-specific cells remained constant with time since diagnosis, the increased number of IFN-γ-producing PBMCs indicated an increase in the proportion of functional CD8 T cells. Among the samples in which responses were detected, there were no effects of either age at time of sampling or time since diagnosis on the magnitude of the response (data not shown).

To establish whether the same trend was apparent in a larger data set, the ELISpot response to peptides with the appropriate HLA restriction was evaluated using PBMCs from 18 of the 50 infants assessed immediately postdiagnosis who had a detectable response to at least one peptide, while 17 of 34 (50%) responded at 4 months postdiagnosis and 29 of 65 (44%) responded at 12 months postdiagnosis, indicating no increase in the number of detectable responses over time since diagnosis.

However, when the magnitudes of the response in those cases where the response was detectable at all were considered, the age at diagnosis correlated with the IFN-γ response at 12 months postdiagnosis (Spearman's r = 0.46; n = 29; P = 0.012), although not in the samples collected immediately and 4 months postdiagnosis (Fig. 5A, B, and C).

As the age at which the 12-month-postdiagnosis sample was taken depended on the age at diagnosis, it was necessary to establish whether the correlation was a direct consequence of the age of infection or simply due to IFN-γ production increasing as the infants aged. To eliminate the effect of age at sampling on the magnitude of the IFN-γ response, a subset of infants who were all the same age of 40 to 60 weeks was identified, but it included infants that had been infected either 4 or 12 months previously. The infants who were sampled 4 months postdiagnosis had been infected at 17 to 39 weeks of age, and those sampled 12 months postdiagnosis had been infected at 4 to 10 weeks of age. There was no difference in the magnitude of response by Mann-Whitney U test (Fig. 5D), indicating that age at diagnosis was not a determinant of the response in itself.

In order to establish whether there was an effect of age at sampling, the magnitudes of response at 4 and 12 months postdiagnosis were compared longitudinally among the subset described above. Nearly all showed a higher magnitude of response at 12 months postdiagnosis than at 4 months postdiagnosis (Wilcoxon statistic = 59; n = 11; P = 0.023) (Fig. 6E). As the increase in the magnitude of the response increased with age at sampling but was not related to the age at diagnosis, the results indicate that the function of the CD8 T cells was low in early infancy and increased with age.

There was no effect of time since diagnosis on the magnitudes.

The range of peptides recognized does not alter with age.

The number of subjects with the appropriate HLA type that responded to any given peptide at any given time point varied from 0 to 59%, with an average of 18.5%. The PBMC samples collected immediately after diagnosis responded on average to 16% of the peptides with the appropriate restriction, but samples collected 4 months later responded to only 12%. A broader response was apparent 12 months later, as PBMCs responded to an average of 28% of peptides tested (data not shown). These differences were not significant by the Kruskal-Wallis test.

Infection with CMV drives differentiation of the CD8 T-cell subset over the first year of life.

It was not possible to ascertain whether the changes in the CD8 T-cell subpopulations were a consequence of CMV infection or of the ontogeny of the CD8 T-cell subset by considering only changes within the CD8 T-cell populations of infected infants. Consequently, the overall CD8 T-cell phenotypes of 51- to 60-week-old uninfected infants were compared with those of infants of a similar age who had been diagnosed at least 43 weeks previously.

Over 90% of the CD8 T cells of 10 of 16 (63%) uninfected infants from whom data on CD27 and CD28 were collected fell into the naïve (CD27⁺ CD28⁺) population, which was a higher proportion of naïve cells than was observed in any CMV-infected infant (P = 0.003) (Table 1 and Fig. 6A). Similar results were found when the stain for CD28 was replaced with a stain for perforin or granzyme A (Fig. 6B and C).

The more differentiated phenotype of the CD8 T cells of infants infected with CMV was also indicated by the greater expression of CD95 and loss of expression of CD62L (P = 0.024) (Table 1 and Fig. 6D). No differences in the relative expression of CD45 isoforms, Bcl-2, HLA-D, or Ki-67 were detected.

The possibility remained that the differences between infants infected or uninfected with CMV were not a direct result of CMV infection but were due to exposure to a wider range of immune challenges than those that remained uninfected. If CMV status is indeed a correlate of exposure, the same is probably true of EBV. The EBV status of the subset of infants that were included in the analyses of the effect of CMV on the CD8 T-cell phenotype was determined, and 29 out of 46 (63%) were found to be infected at the time of sampling. All of the phenotypes that were different between CMV-infected and uninfected infants were tested to establish whether there were similar differences according to EBV status, but none were found (data not shown).