Research ArticleImmunology

CD4+ T cell–dependent and CD4+ T cell–independent cytokine-chemokine network changes in the immune responses of HIV-infected individuals

See allHide authors and affiliations

Science Signaling  20 Oct 2015:
Vol. 8, Issue 399, pp. ra104
DOI: 10.1126/scisignal.aab0808
  • Fig. 1 Dimensionality reduction by PLSDA of 16 cytokines measured at 6 and 72 hours after the stimulation of PBMCs from healthy donors.

    (A to D) Representative sample of four cytokines from the univariate analysis of 16 cytokines that were secreted by PBMCs from healthy donors 6 hours (white) and 72 hours (gray) after stimulation under the four indicated conditions. Data are presented as means ± SD of five donors. Statistically significant differences were determined with Friedman tests [nonparametric one-way analysis of variance (ANOVA)] followed by Dunn’s test to calculate multiplicity-adjusted P values. *P < 0.05, **P < 0.01. (See table S1 for pairwise statistical analysis.) (E and F) PLSDA of VIP-selected cytokines resulted in stimulus-specific classification across all five healthy donors (scores plot; E) with 95% calibration accuracy and 89% cross-validation accuracy. Unstimulated: “no stim,” black; anti-CD3/CD28–stimulated, “CD3/28,” blue; R848-stimulated, “R848,” orange; LPS-stimulated, “LPS,” green. Specific profile compositions can be visualized by colocalization of sample scores (scores plot; E) and cytokine loadings (loadings plot; F); 6-hour cytokine loadings are indicated in lowercase, whereas 72-hour cytokine loadings are indicated in uppercase. (F) Anti-CD3/CD28 stimulation (blue) on the scores plot colocalized with IL-2 (6 and 72 hours), IL-5 (6 and 72 hours), IL-9 (72 hours), IL-4 (6 and 72 hours), IL-17 (6 and 72 hours), and IFN-γ (72 hours) on the loadings plot. R848 stimulation on the scores plot (orange) colocalized with IL-15 (6 and 72 hours), IL-9 (6 hours), and IL-12p70 (6 hours) on the loadings plot. LPS stimulation (green) on the scores plot colocalized with IL-1β (6 and 72 hours) and IL-18 (6 and 72 hours). A model with three LVs captured 63% of the variance in the cytokine and chemokine data (X) and 75% of the variance between stimulus classes (Y). (G to J) Two-dimensional subplots of scores and loadings for visualization purposes.

  • Fig. 2 Comparison of the multivariate cytokine profiles of PBMCs from healthy donors, CD4+ T cell–depleted PBMCs, and PBMCs from HIV-infected individuals.

    (A) Flow cytometric analysis of the percentages of CD4+ T cells among PBMCs from healthy donors (healthy), healthy donor PBMCs from which CD4+ T cells were experimentally depleted (CD4 dep), and PBMCs from HIV-infected patients (HIV). Horizontal bars indicate median values, and each symbol represents a single donor. Data are from PBMCs from four healthy donors and three HIV+ donors. Statistically significant differences were determined with a Kruskal-Wallis test followed by Dunn’s test to calculate multiplicity-adjusted Pvalues. *P< 0.05. NS, not significant. (B and C) Analysis of the cytokine profiles of the indicated sources of PBMCs in response to stimulation with anti-CD3/CD28 beads. A model with two LVs captured 61% of variance in the cytokine and chemokine data (X) and 55% of the variance between cohorts (Y) with 95% calibration accuracy, and achieved 88% cross-validation accuracy for classifying HIV+ responses. (D to G) Analysis of the cytokine profiles of the indicated sources of PBMCs in response to stimulation with R848 (D and E) and LPS (F and G). The R848 model (D and E) captured 42% of the variance in the cytokine and chemokine data (X) and 51% of the variance between stimulus classes (Y), with 100% calibration accuracy and 100% cross-validation accuracy for classifying HIV+ responses. The LPS model (F and G) captured 50% of the variance in the cytokine and chemokine data (X) and 47% of the variance between stimulus classes (Y), with 100% calibration and 95% cross-validation accuracy for classifying HIV+ responses.

  • Fig. 3 DTA reveals the hierarchy of importance of cytokine secretion events in distinguishing the responses of healthy PBMCs, CD4+ T cell–depleted PBMCs, and HIV+ PBMCs.

    (A) After anti-CD3/CD28 stimulation (ovals), the responses of HIV+ PBMCs (n = 7 donors; orange) were differentiated from those of CD4+ T cell–depleted PBMCs (n = 5 donors; green) by secreting >112 pg/ml of IL-2 at 6 hours (bold). In response to innate stimuli (squares), the responses of HIV+ PBMCs were distinguished from those of healthy PBMCs (n = 5 donors; blue) and CD4+ T cell–depleted PBMCs by secreting <580 pg/ml of IFN-γ at 6 hours (bold). Overall, the pruned tree performed with 94% (48 of 51) classification accuracy for differentiating healthy/CD4+ dep and HIV+ innate (R848 and LPS grouped together) responses and healthy, CD4+ dep, and HIV+ adaptive responses (anti-CD3/CD28). Bottom: Table listing the class membership for each leaf node, with correctly classified members in bold. (B to E) Flow cytometric analysis indicated that the maintenance of IL-2 (B and C) and IFN-γ (D and E) production by HIV+ PBMCs originated from CD4+ and CD8+ T cells, respectively. No statistically significant differences or noticeable trends were observed for B cells, NK cells, and monocytes (fig. S5). (F) Flow cytometric analysis indicated that the impaired ability of HIV+ PBMCs to secrete IFN-γ in response to innate stimuli was caused by NK cells. No statistically significant differences or noticeable trends were observed for CD4+ T cells, CD8+ T cells, B cells, or monocytes (fig. S5). Bars and whiskers are medians ± range for healthy PBMCs and CD4+ T cell–depleted PBMCs (n = 4 donors) and HIV+ PBMCs (n = 3 donors). Multiplicity-adjusted P values were calculated from Kruskal-Wallis tests followed by Dunn’s test. *P < 0.05; **P < 0.01.

  • Fig. 4 The magnitude of IFN-γ secretion at 6 hours after stimulation is associated with the divergence of cytokine profiles at 72 hours.

    (A) Cytokine and chemokine profiles 72 hours after the stimulation of the indicated populations of PBMCs with R848. This analysis distinguished the responses of HIV+ PBMCs (n = 7 donors; orange) from those of CD4+ T cell–depleted PBMCs (n = 5 donors; green) and healthy PBMCs (n = 5 donors; blue) with 88% cross-validation accuracy and 94% calibration accuracy. A model with two LVs captured 60% of the variance in the cytokine and chemokine data (X) and 40% of the variance between cohorts (Y). (B) LV1 illustrates the linear combination of cytokine parameters at 72 hours that best differentiate the responses of HIV+ PBMCs from those of other sources of PBMCs in response to R848. The magnitude of IFN-γ secretion at 6 hours after stimulation was statistically significantly correlated with late cytokine response (or score on LV1). Spearman r = −0.54; P < 0.05. (C) Cytokine and chemokine profiles 72 hours after the stimulation of the indicated populations of PBMCs with LPS. This analysis classified the responses of HIV+ PBMCs with 95% calibration and 88% cross-validation accuracy. A model with two LVs captured 50% of the variance in the cytokine data (X) and 46% of variance between cohorts (Y). (D) LV1 illustrates the linear combination of cytokine parameters at 72 hours that best differentiate the responses of HIV+ PBMCs to LPS. The secretion of IFN-γ at 6 hours after stimulation was statistically significantly correlated with late cytokine response (or score on LV1). Spearman r = −0.55; P < 0.05. The secretion of other non-VIP cytokines at 6 hours did not statistically significantly correlate with the cytokine profiles at 72 hours (table S3).

  • Fig. 5 Prior knowledge predicts that IFN-γ is a critical upstream regulator of identified 72-hour profiles.

    (A) Given an input of profile loadings in the 72-hour models (Fig. 4), IPA models predicted the shared master regulators of the cytokine profiles of cells at 72 hours after stimulation with R848 or LPS as determined by sorted activation z-scores. (B and C) Expansion of prior knowledge networks and in silico activity modeling linked early IFN-γ responses to innate stimuli and late cytokine profiles for R848 (B) and LPS (C). Connections were pruned to include 0 and +1 edges (thus constraining model predictions to the simplest, most likely network connections), autoregulatory loops and IFN-γ were excluded, nonrelated TLR pathways were pruned, and the knowledge base was constrained to observations in human immune cells or cell lines. The networks are overlaid with molecule activity predictions upon increased IFN-γ (see Materials and Methods, activation z-score calculations). Positive associations are in red; negative associations are in blue; unpredicted effects are in white. Color intensity represents the confidence of the prediction.

  • Fig. 6 Blockade of IFN-γR in healthy PBMCs validates the influence of magnitude of IFN-γ secretion at 6 hours on the predicted cytokine and chemokine profiles at 72 hours.

    (A and B) PBMCs from normal healthy donors incubated with blocking antibody against the IFN-γR or isotype control antibody were stimulated with R848 (A) or LPS (B) for 72 hours, and the amounts of the indicated cytokines that were secreted were determined. The cytokine amounts secreted by the blocked cultures were normalized to those of the control cultures and are presented as fold differences. A Wilcoxon signed rank test was used to test whether each cytokine differed from a theoretical mean of 1.0. Data are means ± SEM from five donors; each symbol represents a single donor. **P < 0.01. (C to F) Blockade of the IFN-γR for 6 hours in cultures of healthy PBMCs (yellow) resulted in cytokine profiles at 72 hours in response to R848 (C and D) and LPS (E and F) that shifted along LV1 to become more similar to the 72-hour cytokine profiles of HIV+ PBMCs at 72 hours after stimulation with either R848 (C and D) or LPS (E and F). The model was generated with original 72-hour cytokine data from healthy PBMCs (n = 5 donors; blue), CD4+ T cell–depleted PBMCs (n = 5 donors; green), and HIV+ PBMCs (n = 7 donors; orange). Data from IFN-γR blockade experiments with healthy PBMCs (n = 5 donors; blue triangles) were used as training data. The generated model was then used to test the responses of PBMCs from five donors subjected to IFN-γR signaling blockade (n = 5 donors; yellow circles). The first two LVs accounted for 65% of the variance in the R848 model and 63% of the variance in the LPS model.

Supplementary Materials

  • www.sciencesignaling.org/cgi/content/full/8/399/ra104/DC1

    Fig. S1. Analysis of cytokine production by PBMCs from healthy donors.

    Fig. S2. Comparison of PLSDA classification with classification based on individual cytokines.

    Fig. S3. Experimental validation of PLSDA results with PBMCs from healthy donors.

    Fig. S4. Analysis of the cellular composition of PBMCs from various donors.

    Fig. S5. Full, unpruned decision tree.

    Fig. S6. Analysis of the percentages of IL-2–secreting cells in PBMCs stimulated with anti-CD3/CD28 beads.

    Fig. S7. Analysis of the percentages of IFN-γ–secreting cells in PBMCs stimulated with R848 or LPS.

    Fig. S8. Consistent, enhanced secretion of TNF-α by monocytes of all cohorts.

    Fig. S9. Decision tree sensitivity analysis.

    Fig. S10. Minimal standard curve variance between batches and patient cohorts.

    Table S1. Matrix of statistical significance tests for all pairwise univariate comparisons.

    Table S2. HIV-infected donor information.

    Table S3. Correlations between cytokine secretion at 6 hours and the divergence in cytokine profiles 72 hours after innate immune stimulation.

  • Supplementary Materials for:

    CD4+ T cell–dependent and CD4+ T cell–independent cytokine-chemokine network changes in the immune responses of HIV-infected individuals

    Kelly B. Arnold, Gregory L. Szeto, Galit Alter, Darrell J. Irvine,* Douglas A. Lauffenburger*

    *Corresponding author. E-mail: djirvine{at}mit.edu (D.J.I.);lauffen{at}mit.edu (D.A.L.)

    This PDF file includes:

    • Fig. S1. Analysis of cytokine production by PBMCs from healthy donors.
    • Fig. S2. Comparison of PLSDA classification with classification based on individual cytokines.
    • Fig. S3. Experimental validation of PLSDA results with PBMCs from healthy donors.
    • Fig. S4. Analysis of the cellular composition of PBMCs from various donors.
    • Fig. S5. Full, unpruned decision tree.
    • Fig. S6. Analysis of the percentages of IL-2–secreting cells in PBMCs stimulated with anti-CD3/CD28 beads.
    • Fig. S7. Analysis of the percentages of IFN-γ–secreting cells in PBMCs stimulated with R848 or LPS.
    • Fig. S8. Consistent, enhanced secretion of TNF-α by monocytes of all cohorts.
    • Fig. S9. Decision tree sensitivity analysis.
    • Fig. S10. Minimal standard curve variance between batches and patient cohorts.
    • Table S1. Matrix of statistical significance tests for all pairwise univariate comparisons.
    • Table S2. HIV-infected donor information.
    • Table S3. Correlations between cytokine secretion at 6 hours and the divergence in cytokine profiles 72 hours after innate immune stimulation.

    [Download PDF]

    Technical Details

    Format: Adobe Acrobat PDF

    Size: 1.04 MB


    Citation: K. B. Arnold, G. L. Szeto, G. Alter, D. J. Irvine, D. A. Lauffenburger, CD4+ T cell– dependent and CD4+ T cell–independent cytokine-chemokine network changes in the immune responses of HIV-infected individuals. Sci. Signal. 8, ra104 (2015).

    © 2015 American Association for the Advancement of Science

Navigate This Article