Impact of donor CMV status on viral infection and reconstitution of multifunction CMV-specific T cells in CMV-positive transplant recipients

Prospective study subjects

Peripheral blood mononuclear cells (PBMC) from R⁺ HCT recipients were collected between day 40 and day 360 post-HCT (for details see Gallez-Hawkins et al³⁷). R⁺ HCT recipients (178), whose CMV status was determined using latex agglutination (CMV-SCAN; Becton Dickinson, San Jose, CA), were enrolled in an observational trial approved by the City of Hope Institutional Review Board (IRB) with informed consent obtained in accordance with the Declaration of Helsinki. CMV reactivation is defined as the initial demonstration of CMV infection either by blood culture (BC) or by 2 consecutive quantitative serum polymerase chain reaction (PCR) assays of CMV DNA during a monitoring period that began at day 21 post-HCT and continued thereafter twice weekly until day 100 as previously described.³⁷ After day 100, CMV in plasma was monitored in patients at high risk because of graft-versus-host disease (GVHD) or immunosuppressive medication. Subjects with PCR or BC-confirmed CMV reactivation were treated with GCV for 6 weeks.³⁸ A subset (n = 135) of the subjects limited only by sample availability were evaluated for CMV-specific CD8⁺ T cells producing IFN-γ, and a smaller subset (n = 62) limited by levels of CMV-specific T cells and complexity of the assay was further evaluated for T-cell multifunctional profile. Patient characteristics of all 3 groups of subjects, including CMV and reactivation status are summarized in Table 1.

Peptides

Pools of overlapping 15-mer CMV pp65 peptides (PepMixes) were purchased from JPT Peptide Technologies (Berlin, Germany).

Stimulation of PBMC for ex vivo analysis of CMV-pp65 T cells and functional markers

CMV pp65 PepMixes (1 μg/mL) or diluent were added to 1 million freshly thawed cryopreserved PBMC, and cells were processed for flow cytometry according to published methods. For assays evaluating MIP-1β, TNF-α, and IFN-γ secretion as well as CD107a/b degranulation/mobilization, costimulatory antibodies to CD28 and CD49d (all antibodies were from Pharmingen, San Diego, CA) were added to the cultures to 1 μg/mL each at the same time as CMV pp65 PepMix or diluent. Flow-based degranulation was measured by incubation of PBMC with fluorescein isothiocyanate (FITC)–conjugated antibodies to CD107a and CD107b, as well as 1 μL monensin (GolgiStop; Pharmingen) and 1 μL brefeldin (GolgiPlug; Pharmingen) were also added to the medium before incubation.³³

Flow cytometric analysis

For intracellular cytokine (ICC) assays evaluating only IFN-γ production, aliquots of cells were treated overnight (O/N) with pp65 PepMix, washed, and labeled for 20 minutes at 4°C with phycoerythrin (PE)–CD8 before permeabilization (Cytofix/Cytoperm; Pharmingen), followed by labeling with allophycocyanin (APC)–IFN-γ for 30 minutes. In the case of assays evaluating multiple cytokine secretion, cells were first stained with both anti–human PE-Cy5-CD8 and APC-Cy7-CD3 before permeabilization, followed by staining afterward with PE-Cy7–TNF-α, PE–MIP-1β, and APC–IFN-γ. Flow analysis was performed on a FACSCanto flow cytometer (Becton Dickinson). Flow gating strategies are described in the Results section. Events (100 000-200 000) were collected for each patient sample.

Statistical methods

Reactivation and antiviral use were analyzed by proportional hazards regression.³⁹ Models were fit using the coxph function within the R statistical programming language, treating reactivation as time-to-event data, and antiviral use as recurrent events. The product-limit estimator was used to estimate survival functions from censored time-to-event data, and the cumulative distribution of recurrent antiviral use was estimated as the complement of the estimator of Kalbfleisch and Prentice.⁴⁰ Both estimates were computed in R using the survfit function. These survival-time methods accommodate variation in time at risk and analysis of recurrent events. Wald tests within a multivariable model were used to evaluate the effect of donor CMV status while adjusting for baseline and time-dependent covariables. A propensity score⁴¹ was used to simultaneously adjust for baseline covariables, as discussed below. Simpler comparisons, such as Fisher exact and rank-sum tests, were also used where noted.

IFN-γ–CD8⁺ T-cell assays were performed on blood samples obtained for a subset of 135 R⁺ recipients. The longitudinal patterns of IFN-γ–CD8⁺ T-cell percentages were explored graphically and analyzed using linear models on a logarithmic scale, fit by the generalized estimating equation (gee) method using an independent working correlation.⁴² The generalized estimating equation and geepack packages for the R programming language were used for model fitting. This method accounts for the stochastic dependence of repeated measures on the same subject. Wald tests were calculated by dividing parameter estimates by their respective standard errors (SE), conservatively using the so-called naive SE if the GEE robust SE was smaller. The effect of donor CMV status was modeled simultaneously with the propensity score to accommodate pretreatment covariates, with additional adjustment for days posttransplant (DPT), prior reactivation, and GVHD. Interactions were modeled by the standard device of including product terms in the design matrix. Simultaneous adjustment for individual pretreatment covariates was used as an alternative to the propensity score, to check robustness.

Multifunctional CD8⁺ T-cell assays were evaluated on samples from 62 patients. Analysis of this subset addresses the effect of multifunctional T cells as a percentage of IFN-γ–CD8⁺ T cells, conditional on adequate numbers of the latter. GEE models were fit to accommodate the stochastic dependence of the longitudinal measurements. A square-root transformation was used to reduce skewness. Linear and logarithmic scales were also analyzed to assure robustness of findings. The effect of donor CMV status was adjusted for time-dependent covariables, including DPT, maximum steroid dose in the preceding 14 days, CMV reactivation, and the covariates established at HCT. The latter group of covariates was combined into a propensity score by fitting a logistic regression for donor CMV status⁴¹ using the subset of 62 patients.

Covariate adjustments in both proportional hazards and GEE models involved the use of propensity scores to adjust for potentially confounding factors associated with the patient or HCT procedure. For a variable to be confounding, it must be associated with the risk factor (ie, donor CMV status) and the outcome of interest. Statistical adjustment using a propensity score reduces confounding by addressing associations between covariates and risk factor. The propensity score is computed by fitting a logistic regression of donor CMV status on all of the potentially confounding covariates. This yields a score based on all covariates, which is then used for statistical adjustment via standard models. GVHD was treated as a time-dependent covariate.

Sampling considerations

Separate propensity scores were computed for each of the 3 subsamples (Table 1), as it is the actual imbalances in each sample that may require adjustment. The effect of subsampling on the distribution of covariates was subjected to statistical testing, as a screen for unanticipated selection effects. In addition to testing each covariate for association with subgroup membership, its association with donor CMV status and the joint association of covariate, donor CMV status, and subgroup were tested as a screen for possible systematic changes in covariate balance. Pearson χ² was used for the categorical variables, and t tests were used for age. The full sample of 178 R⁺ patients, and the subsample of 135 R⁺ recipients with available blood samples, can both be regarded as unselected and unbiased cohorts of R⁺ HCT patients. The subset of 62 patients is a conditional sample, requiring minimal levels of IFN-γ–CD8⁺ T cells. Although all 135 recipients with blood samples might have been analyzed by assuming that low levels of IFN-γ–CD8⁺ T cells imply levels of multifunctional T cells below detection limits, such an approach would not distinguish effects on the numbers of IFN-γ–CD8⁺ T cells from effects on the fraction that are multifunctional. The subset of 62 patients addresses the latter question, which involves a distinct, conditional population.

Pretransplant patient characteristics

HCT donors and recipients were enrolled in a longitudinal study investigating the role of adaptive immunity in limiting CMV sequelae.³⁷ Table 1 summarizes pretransplant characteristics (upper half) and clinical outcomes (lower half) by donor type, within each of the 3 nested sets of transplants that are distinguished in this study. Group 1 (n = 178) includes the full set of R⁺ recipients who were followed for clinical outcome measures. Group 2 (n = 135) is the subset composed of all recipients who had blood samples available for monitoring IFN-γ–CD8⁺ T cells. Group 3 (n = 62) is a subset based solely on additional immunologic studies that were conducted and discussed later in this report. The most prominent pretransplant difference between D⁺/R⁺ and D⁻/R⁺ recipients in groups 1 to 3 was the higher percentage of unrelated D⁻ donors, and the correspondingly higher number of sibling D⁺ donors. This difference is well beyond random variation (P < .001) and may reflect the similarity of CMV status among siblings. The relation of donor and recipient would confound the comparison of D⁺/R⁺ with D⁻/R⁺ transplants for any end point that is strongly associated with the relation, and we have observed that donor relation explains most of the difference between survival of D⁺/R⁺ and D⁻/R⁺ HCT recipients (data not shown). Donor relation is not, however, associated with any of the end points described in this report, and adjustment for donor relation did not alter any of our conclusions. None of the other pretransplant factors (Table 1) showed evidence of systematic differences between the D⁺/R⁺ and D⁻/R⁺ transplants in groups 1-3 (P > .2 generally, P = .09 for total body irradiation [TBI] in group 2 only). Neither did pretransplant factors show any statistical evidence of association with group membership, nor was subgroup definition associated with systematic variation in the balance of pretransplant factors across donor types. Although the validity of simple comparison of D⁺/R⁺ versus D⁻/R⁺ transplants in any of groups 1 to 3 is supported by the screening tests, and the lack of association of donor relation with the outcomes of interest, we also used statistical adjustment via propensity scores to adjust for the actual imbalances in all pretreatment covariables, regardless of statistical significance.

Clinical virology outcomes

The average length of follow-up was 10.4 months for D⁺/R⁺ recipients and 9.2 months for D⁻/R⁺ recipients, a small but statistically significant difference (P = .01, 2-sided rank-sum test). Consequently, statistical methods for censored event times were used, as these accommodate differences in time under study, as well as incorporating time-of-event information. However, most of the difference in follow-up developed late, after most of the events of interest had already occurred, so simple comparisons of proportions should have little bias. Component measures of CMV reactivation are shown in Table 1. All are more frequent in D⁻/R⁺ recipients, although most are not statistically significant when taken individually. Detection of viremia by PCR is one of the signals for treatment with GCV, and the hazard ratio for a first positive PCR was estimated between 0.85 and 1.9 (with 95% confidence, unadjusted). GCV was required by 56% of D⁻/R⁺ recipients, compared with 44% of D⁺R⁺ recipients (P = .05, by log-rank test). The estimated hazard ratio is 1.5, with 95% confidence limits of 0.99 and 2.5. Figure S1 (available on the Blood website; see the Supplemental Materials link at the top of the online article) shows the Kaplan-Meier estimates of the probability of remaining GCV-free during the 1-year observation period for both donor groups. However, the statistical significance of this finding is not robust to adjustment for confounding variables via the propensity score (P = .18).

Association of donor CMV status with antiviral usage

Treatment of HCT recipients in response to PCR detection of CMV reactivation often requires multiple rounds of antivirals. Prolonged or recurrent treatment reflects difficulty in resolving the infection and subjects a patient to an increased toxicity hazard. While the higher frequency of initiating GCV therapy in the D⁻ transplants was of equivocal statistical significance, the average duration of therapy, once begun, was 20 days longer for D⁻ than D⁺ recipients (Table 2). D⁻ recipients were also twice as likely to require antivirals for more than 6 weeks and more than twice as likely to require more than 7 weeks of antiviral therapy (Table 2). The tendency of D⁻ recipients toward more days on antivirals was statistically significant (P < .05, 2-sided rank-sum test). The total days on antiviral drugs are shown in Figure 1A, overlayed with a pair of boxplots. The D⁺ group includes more recipients who never received GCV (median at zero) and fewer outliers who required antivirals for extended periods.

Association of donor CMV status with recurrent antiviral usage

We define recurrent antiviral usage as restarting antiviral drugs after a hiatus of more than 14 days. Subjects were regarded as at risk for a recurrent CMV event when they had gone 14 days without antivirals. The recurrent need for antivirals was almost entirely in recipients with D⁻ donors (8 of 50 vs 1 of 128, Table 2). The difference in frequencies is highly significant (P < .001, Fisher exact test). Considering the time at risk for initial or recurrent antiviral use in a proportional hazards model leads to a similar conclusion (P < .001, likelihood ratio test), while allowing for variation in observed time at risk³⁹ (Figure 1B). The plot of cumulative incidence shown in Figure 1B compares the D⁺ and D⁻ donor groups with regard to prescribed antiviral treatments, estimated from recurrent CMV reactivation events. These estimates use the method of Kalbfleisch and Prentice and reflect differences between groups in both the need for a first course of GCV and in the recurrent need for antivirals, as defined in the legend for Table 2.⁴⁰ Acute GVHD is associated with the hazard for requiring antivirals, but adjusting for GVHD has little effect (11%) on the estimated relative hazard associated with donor CMV status, which remains highly significant (Table 3). This indicates that the association of donor CMV status with antiviral usage by recipients is only minimally effected by incidence of GVHD. The excess hazard associated with a CMV-negative donor remained significant (P = .009) after adjusting for confounding factors via the propensity score (Table 3). These data are consistent with an effect of donor status on recurrent usage of antivirals, which is a critical risk factor that is known to effect survival outcomes in HCT.

IFN-γ–CD8⁺ frequency in recipients with CMV-positive donors

We measured immunity to CMV by flow detection of ex vivo CD8⁺ T cells producing IFN-γ (IFN-γ–CD8⁺) in response to stimulation with CMV-pp65 PepMix. Of 135 evaluated subjects, 98 received allografts from D⁺ and 37 from D⁻ donors (Table 1). We modeled the level of IFN-γ–CD8⁺ as a function of DPT, donor CMV status, and whether CMV reactivation occurred before the sampling date. The data are displayed in Figure 2. Models were fit using the GEE method to accommodate the stochastic dependence of repeated measurements of the same subjects⁴² (Table 4). The major finding is that the average (geometric mean) level of IFN-γ–CD8⁺ T cells in D⁻/R⁺ patients without CMV reactivation is only 13% of mean levels in D⁺/R⁺ patients. However, levels of IFN-γ–CD8⁺ in postreactivation D⁻/R⁺ PBMC samples substantially rose and were at similar levels to PBMC samples from D⁺R⁺ patients. There was no detectable time trend (DPT row in Table 4), supporting the findings in Figure 2. All 4 groups displayed in Figure 2 were simultaneously evaluated in a model that tested for the effect of donor status on IFN-γ–CD8⁺ levels. Within the subset of measurements taken before reactivation, the D⁻ donor group produced lower levels of IFN-γ–CD8⁺, on average, than the D⁺ donor group (Figure 2, P < .001, unadjusted for other factors). Of the pretransplant covariates (Table 1), only recipient age was associated with IFN-γ–CD8⁺ levels (P = .04) (data not shown). Furthermore, inclusion of acute GVHD in the model did not alter the significant difference of measurements made with prereactivation D⁻/R⁺ patients compared with all other groups. GVHD is not associated with IFN-γ–CD8⁺ levels and does not mediate the effect of CMV-negative donor status. In conclusion, taking into account pretransplant covariates and posttransplant outcomes, a substantially smaller level of IFN-γ–CD8⁺ is found in the D⁻/R⁺ prereactivation samples compared with all other groups (P < .001, adjusted for propensity score, DPT, GVHD in Table 4).

Analysis and gating scheme for CD8⁺ T cells expressing multiple cytokines and CD107

We observed IFN-γ levels to be similar, on average, in D⁺/R⁺ and D⁻/R⁺ recipients who experienced CMV reactivation (Figure 2). However, the D⁺/R⁺ group had less frequent need for GCV (Figures 1A,B). We speculated that this difference based on donor status might be explained by functional characteristics other than IFN-γ levels of CMV-specific T cells. Therefore, we investigated functional properties of CD8⁺ T cells in a subset of 62 HCT recipients (group 3, Table 1) who had available lymphocyte samples at day 90, day 180, or day 360 post-HCT with measurable IFN-γ–CD8⁺ greater than 0.2% of CD8⁺ T cells. Eligibility criteria was solely conditional on a measurable quantity of IFN-γ–CD8⁺. The immunologic analysis consisted of an expanded set of T-cell functional parameters (IFN-γ, TNF-α, CD107a/b, and MIP-1β) measured simultaneously using 6-color flow cytometry.^34,43 Figure S2 shows the gating scheme. A primary gate was placed on lymphocytes by forward and side scatter, and a second gate on CD3⁺/CD8⁺ T cells (Figure S2A). These cells were then analyzed for coexpression of IFN-γ, TNF-α, CD107a/b, and MIP-1β by Boolean gating. Representative examples of D⁺/R⁺ (Figure S2B) and D⁻/R⁺ subjects (Figure S2C) are shown. Mock stimulated samples sometimes contained low levels of CD8⁺ T cells that were positive for only IFN-γ, MIP-1β, or CD107 (Figure S2B and C, bottom rows of plots). These single-positive populations were thus considered nonantigen-specific, and their levels were subtracted from the observed levels of pp65-stimulated T cells.

D⁺/R⁺ recipients maintain higher levels of CMV-pp65 CD8⁺ T-cell subsets with multiple functions than D⁻/R⁺

An initial evaluation of the expression of 4 markers that were followed in the subset of 62 patients (group 3, Table 1) was done as individual staining of each marker on PBMC and grouping all longitudinal measurements (Figure 3). In all 62 subjects tested, CD8⁺ T-cell subsets producing TNF-α, MIP-1β, or CD107a/b, were present at a lower frequency than CD8⁺ T-cell subsets producing IFN-γ (Figure 3). Among the 4 markers evaluated, only IFN-γ levels were significantly different based on donor CMV status, and only when all 3 observation times were combined (in contrast, see Figure 4B). Therefore, we examined another approach to comparing T-cell phenotypes by measuring subsets of IFN-γ–CD8⁺ composed of 8 different combinations of all 4 markers, while graphically comparing D⁻/R⁺ and D⁺/R⁺ patients for differences in frequencies of all subsets (Figure 4). We refer to the left-most points for each longitudinal comparison shown in Figure 4A as “4-functional” T cells. T cells positive for TNF-α were not common, with medians in the 1% to 5% range for each combination of markers that includes TNF-α (Figure 4A). D⁺/R⁺ subjects reconstituted significantly higher levels of 4-functional CD8⁺ T cells at day 180 (P < .001) and day 360 (P < .001), with weaker evidence (P = .07) at day 90 (Figure 4A). In contrast, higher levels of subsets lacking TNF-α were not correlated with CMV-positive donor status (Figure 4B). In summary, CMV-pp65 CD8⁺ T-cell populations from D⁺/R⁺ recipients contained higher levels of 4-functional and 3-functional subsets than D⁻/R⁺ recipients, whose CD8⁺ T cells expressed fewer functions, analogous to HIV progressors³⁴ (Figure 4C). Levels of IFN-γ⁺-CD8⁺ were comparable, albeit highly variable, in D⁻/R⁺ and D⁺/R⁺ subjects who reactivated CMV (Figure 2).

We used a more statistical approach of highlighting the differences in phenotypic subsets of IFN-γ–CD8⁺ by testing association of donor status with all 8 marker combinations in a GEE model on a transformed scale (Figure 5A). This test was statistically significant (P < .001, on 8 degrees of freedom), justifying further analysis. Examination of the 8 immune marker combinations, averaged over time, showed that D⁻ recipients had, on average, a substantially higher fraction of IFN-γ–CD8⁺ that were negative for all other markers (40.6% in D⁻ vs 32% in D⁺) and relatively fewer IFN-γ–CD8⁺ that were positive for at least 2 of the other markers (Figure 5A). The difference in T cells positive for all markers was striking (mean 1.8% in D⁻ vs 4.7% in D⁺, rank-sum test P < .001). Individual comparisons shown in Figure 5A also showed striking differences exemplified by comparing triple-positive CD8 (IFN-γ, TNF-α, and CD107a/b) T cells in recipients with D⁻ and D⁺ donors (P = .014) as well as 4-functional T cells (P = .001). Because CMV reactivation influenced IFN-γ–CD8⁺ levels differently depending on donor status, we analyzed 4-functional results similarly to Figure 2 as shown in Figure 5B. In addition, the effects of steroids were added to the model as described in the next section. The unadjusted P values shown in Figure 5B show a clear difference based on CMV status that affects levels of 4-functional T cells, either prior or postreactivation.

Multivariate analysis of hazards that influence immunologic reconstitution

Steroids are given in response to a diagnosis of acute or chronic GVHD, and their use at levels greater than 1.0 mg/kg patient weight is known to suppress immune responses.⁴⁴ We calculated the dose level of administered steroids by dividing the amount taken by the weight of the patient, and recording the maximum, in mg/kg, in a 14-day window before each blood sampling time point (Table 1). Most steroid doses in our sample were below 1.0 mg/kg. Nonetheless, we fit a model to test for the effect of donor CMV status on the levels of 4-function T cells while adjusting for steroid dose, DPT, reactivation status, and pretransplant covariates (ie, propensity score). To accommodate the stochastic dependence of repeated measurements of the same subjects, we fit the model using the GEE approach.⁴² Our initial analysis used a square root transformation to remove skewness, but the findings with regard to donor CMV status were robust to the scale used. The fitted model estimates and test statistics are shown in Table 5, using the original scale for ease of interpretation. Donor CMV status was significantly associated with 4-function CD8⁺ T-cell levels (P < .003 on all scales, GEE Wald test), adjusting for all pretransplant covariates via a propensity score,⁴¹ and simultaneously adjusting for prior CMV reactivation, DPT, and steroid dose. The model-adjusted T-cell deficit associated with D⁻ donors is 2.08, which corresponds to 62% lower levels of 4-function T cells in the D⁻/R⁺ than D⁺/R⁺ recipients. Omitting the adjustment for steroid dose, which was not significantly associated with T-cell levels (P > .34, all scales, Wald test), allowed the recovery of 11 observations that had missing steroid data, which gave stronger evidence of association of donor status and T-cell levels (P < .001, square-root scale, Wald test). Substituting GVHD (grades 2-4) for steroid usage in the model did not statistically alter the results or conclusions reached (data not shown). The association of donor status with 4-function CD8 T-cell levels was robust to the method of covariate adjustment, remaining statistically significant when the adjustment for the pretransplant covariates was also made via simultaneous inclusion of covariates in a large multiple regression model (P < .02, all scales, Wald test). Most of the variables included in the large model had no evidence of any relationship to multifunctional T-cell levels. There was some evidence that reactivation and donor sex may be weakly associated with multifunctional T-cell levels, but these findings were not robust to adjustments for other variables. Only donor CMV status was consistently associated with multifunction T-cell levels.