Multi-objective optimization reveals time- and dose-dependent inflammatory cytokine-mediated regulation of human stem cell derived T-cell development

A two-phase screening strategy for enhancers of ProT-cell differentiation

Using the DL4 + VCAM-1 ETN platform, we sought to identify soluble cytokines that positively regulate T-lineage differentiation and expansion from UCB-derived CD34⁺ HSPCs (Supplementary Fig. 1). A list of 15 candidate molecules was assembled from a survey of T-cell development literature in both mouse and human. These molecules were tested in combination at three concentrations each for total, CD7⁺ lymphocyte, and CD7⁺CD5⁺ proT-cell expansion. To separate the effects of cytokines on early HSPCs and emerging proT-cells, the experiment was performed in two separate stages (Fig. 1a). Test conditions were compared to a control condition that contained SCF, Flt3L, TPO, and IL-7 (4F) at 100 ng/ml each, as was used in our previous work²².

Of the 15 cytokines tested, SCF, IL-3, and TNFα elicited strong proliferative effects from days 0–7 on all three populations, while IL-7 had only a small effect on total and CD7⁺ cell numbers (Fig. 1b). From day 7–14, the effect of IL-7 was much greater, although cells still responded most strongly to SCF, IL-3, and TNFα. Other cytokines, such as IFNγ and IL-6, had a negative effect on the expansion of one or more of the cell populations and were excluded from future experiments. A similar experiment was performed including IL-3 and TNFα at a higher range of concentrations to confirm our observations (Supplementary Fig. 2). Based on this experiment, working concentrations for IL-3 and TNFα were chosen as 10 and 5 ng/ml, respectively.

We measured cell proliferation using carboxyfluorescein succinimidyl ester (CFSE) dye for 4F cytokines (control) or 4F plus one of IL-3 and TNFα. Cells treated with IL-3 proliferated more than the control, but this was primarily in the non-lymphoid (CD7^–) fraction (Fig. 1c, d). Cells treated with TNFα proliferated similarly to the control. All cells transitioned through proT1 and proT2 stages after 5 days, consistent with development on OP9-DL4⁶. However, a significantly (p < 0.05) higher proportion of CD7⁺ cells treated with TNFα had a proT2 phenotype (37.6 ± 4.9%) compared to the IL-3 (17.5 ± 2.1%) and control (15.2 ± 2.6%) (Fig. 1e).

Interactions between TNFα and the Notch pathway enhances T-lineage differentiation

The early increased proportion of a proT2 phenotype in cells treated with TNFα made us ask whether it was interacting with the Notch1 pathway to enhance T-lineage specification. TNFα signals through the NF-κB pathway, which, in other cell types, has been shown to interact extensively with Notch in a context-dependent manner (Fig. 2a)²³. We were interested to see if the effects of TNFα were specific to HSCs and multipotent progenitors (MPPs) or their downstream progeny. We therefore sorted CD34⁺ HSPCs into CD38^lo/- and CD38⁺ fractions to separate HSCs/MPPs from more differentiated progenitors. We seeded each fraction on DL4 + VCAM-1 with and without TNFα and measured the expression of Notch target genes after 5 days (Fig. 2b). The addition of TNFα increased the expression of GATA3 and TCF7 (encoding TCF-1)—genes that are important for T-lineage specification²⁴—in both the CD38^lo/− and CD38⁺ fractions (Fig. 2c). BCL11B, a gene associated with T-lineage specification and commitment²⁵, was also upregulated in CD38^lo/− cells treated with TNFα. No significant differences were observed in HES1, DTX1, E2A, or NOTCH1 mRNA levels, suggesting that this effect was not due to an increase in Notch1 receptor expression and overall Notch pathway activation. TNFα also induced a modest decrease in the myeloid gene SPI1 (encoding PU.1) in CD38^lo/− HSPCs, and a significant decrease in CEBPA in CD38⁺ HSPCs. The decrease in CEBPA mRNA levels was not due to an increase in HES1 expression, which antagonizes CEBPA²⁶. The upregulation of T-lineage genes and decrease in pro-myeloid-lineage genes by TNFα provides a mechanism by which it inhibits myeloid differentiation. The increased expression of BCL11B in only the CD38^lo/- fraction implies that they have a higher propensity for T-lineage differentiation, consistent with the previous reports⁶.

Given that TNFα enhances the expression of T-lineage specification genes, we next tested whether it could decrease dependence on Notch signaling during T-cell development. We placed CD34⁺ HSPCs on DL4 + VCAM-1 for 14 days and used the γ-secretase inhibitor (GSI) DAPT to inhibit Notch activation. We calculated the 50% effective dose (ED₅₀) of GSI for CD7⁺CD5⁺ cells with and without TNFα using linear interpolation. Without TNFα, the ED₅₀ for the frequency of CD7⁺CD5⁺ cells was 0.12 ± 0.02 µM (mean ± standard error); with TNFα, the ED₅₀ was 0.52 ± 0.10 µM (Fig. 2d, e). Similarly, the ED₅₀ for the fold-change in CD7⁺CD5⁺ cell numbers without TNFα was 0.12 ± 0.03 µM and with TNFα was 0.36 ± 0.06 µM. TNFα was able to maintain CD7⁺CD5⁺ cell generation with a significantly higher concentration of GSI than without. Thus, although it cannot replace Notch activity completely, TNFα can partially compensate for Notch pathway inhibition through the regulation of Notch target genes.

TNFα synergizes with IL-3 to enhance ProT-cell expansion

Cells treated with only IL-3 divided more than those not treated with IL-3. However, increased proliferation was observed predominantly in the CD7^– non-lymphoid population. This finding contrasted with our screening experiment that found that IL-3 stimulated the proliferation of CD7⁺ lymphoid cells. Because our screen tested all factors in combination, we hypothesize that IL-3 may be interacting synergistically with TNFα. To test this, we seeded CD34⁺ HSPCs on DL4 + VCAM-1 with IL-3+TNFα in combination. The use of both cytokines led to significantly higher expansion than with any single cytokine alone, and cells were confluent by day 7 and required passaging.

By day 7, the IL-3+TNFα group had expanded 75.0 (55.6–86.8)-fold (median, 5–95 percentile) compared to 20.7 (12.2–42.7)-fold in IL-3, 29.5 (9.74–55.9)-fold in TNFα, and 5.1 (2.4–15.9)-fold in the control groups (Fig. 3a, b). They also had a higher frequency of CD7⁺CD5⁺ cells than the IL-3 and control groups. CD7⁺CD56⁺ NK frequencies were less than 4% in all groups, and the frequencies of CD14/33⁺ myeloid cells were variable and not significantly different. By day 14, fold expansion was an order of magnitude greater in the IL-3+TNFα group at 753.2 (532.4–1026.9)-fold, compared to 69.0 (45.8–190.8)-fold in the IL-3, 90.7 (27.5–159.1)-fold in the TNFα, and 8.9 (4.3–21.5)-fold in the control groups (Fig. 3c, d). The frequency of CD7⁺CD5⁺ cells was also significantly greater with IL-3+TNFα than all other conditions. As on day 7, CD7⁺CD56⁺ NK frequencies were low, and both TNFα and IL-3+TNFα groups had a significantly lower frequency of CD14/33⁺ myeloid than IL-3 or control groups. Thus, IL-3+TNFα synergize to elicit significant proliferation and preferentially enrich cultures for CD7⁺CD5⁺ cells.

TNFα regulates expression of the IL-3 receptor

To elucidate a mechanism for the synergistic effect of IL-3 and TNFα, we examined the expression of the IL-3 receptor (CD123) after stimulation of CD34⁺ HSPCs with TNFα (Fig. 3e). TNFα has been reported to upregulate CD123 in human bone marrow (BM)-derived CD34⁺ HSPCs²⁷. Consistent with this, we found that TNFα increased the frequency of CD123⁺ cells after 24 h (Fig. 3f). TNFα stimulation also increased the median fluorescent intensity (MFI) of CD123 expression, indicating an increase in the number of CD123 molecules on the surface of cells. We observed no change in the frequency and MFI of the SCF (CD117) and IL-7 (CD127) receptors in our UCB-derived HSPCs. The synergy between IL-3 and TNFα is therefore due, at least in part, to increased responsiveness of cells to IL-3 through an increase in the number of cells expressing the receptor as well as an increase in the level of expression.

TNFα switches from an enhancer to an inhibitor during T-cell development

Next, we sought to identify cytokine signaling requirements for T-lineage maturation on DL4 + VCAM-1. We used response surface methodology (RSM) to model the dose-response of cells to SCF, Flt3L, IL-3, IL-7, TNFα, and CXCL12 (Fig. 4a). Multivariate regression was used to fit polynomial models to experimental data (Supplementary Fig. 3). We excluded TPO after we found that removing it reduced CD14/33⁺ myeloid generation without detrimentally affecting CD7⁺ lymphoid expansion (Supplementary Fig. 4). CXCL12 was included because we observed a modest but positive effect in screening experiments and because of its reported positive effect on cell survival during β-selection^28,29. Following our Notch inhibition experiment and reports that αβT-cell development requires reduced Notch pathway activation around the β-selection checkpoint³⁰, we titrated DL4 and reduced the concentration 7.5-fold while maintaining similar proportions of proT-cells (Supplementary Fig. 5). Experiments were conducted over 7-day intervals, and the number of cells in each population was measured using flow cytometry. Because the RSM has higher predictive power than the definitive screening design, we included the first 2 weeks of differentiation to estimate cytokine dose responses more accurately during T-cell specification. We measured proT-cells, CD4ISPs, and early DPs (CD3⁻) during the first 14 days. From day 14 onwards, we included CD4ISPs, early DPs, late DPs (CD3⁺), and CD8SPs (CD4⁻CD8α⁺CD3⁺).

On day 7, large populations of proT and CD4ISP cells were present in cultures. These cells responded strongly to increasing concentrations of SCF, IL-3, and TNFα (Fig. 4b). Positive two-factor interactive effects were observed between SCF and TNFα and IL-3 and TNFα (Supplementary Fig. 6). A small number of early DP cells was also present by day 7. Between days 7–14, all populations were responsive to increasing concentrations of SCF and IL-7 and unresponsive to IL-3. Cells responded positively to TNFα and CXCL12 with a negative interaction, but the magnitude of these effects was small compared to SCF and IL-7 (Supplementary Fig. 6).

Cultures were primarily CD3⁻ on day 21, but the proportions of CD3⁺ cells steadily increased from day 28 onwards (Fig. 4c). Again, TNFα had a small positive effect on cell expansion at higher concentrations between day 14–21, but this became negative from day 21 onwards. Differentiation was dominated by SCF and IL-7, and the cytokines had some interactive effects from day 21 through day 35 (Supplementary Fig. 6). The response to IL-3 was small between day 21–28; cells were completely unresponsive to IL-3 from day 28 onwards. Flt3L had little or no measurable effect on cell expansion throughout the entire assay.

An optimized three-stage process for T-cell generation from blood stem cells

In order to define a set of preferred conditions for each step in the differentiation, we optimized the RSMs to find the cytokine concentrations that maximized the number of cells in different populations for each 7-day interval. We assumed an ancestor-progeny developmental relationship such that increasing the number of early T-lineage progenitors would have a positive impact on the number of later-stage cells. A desirability function for each population was used to calculate overall desirability, which was maximized using the basin-hopping algorithm (Supplementary Fig. 7)³¹. The optimization objectives were changed throughout the differentiation to reflect the populations present, first maximizing cells in the CD3^– populations and shifting to CD3⁺ cells (Fig. 4d).

The top five solutions from the optimization converged to the same overall desirability score indicating that they are global maxima (Fig. 4e and Supplementary Fig. 7). Next, we constructed a three-stage protocol that approximated the 7-day interval optima as closely as possible. We split the assay into the intervals [0, t₁], [t₁, t₂,], and [t₂, 42] days, where t₁, t₂ are multiples of 7 in [7, 42) and t₁ < t₂. We then averaged the predicted optimal cytokine concentrations within the intervals for every t₁, t₂ and found the pair that yielded the highest average overall desirability over the entire 42-day assay. Using this approach, we found t₁ = 7 and t₂ = 21 maintained desirability scores that were close to the 7-day interval scores (Supplementary Fig. 7). The three-stage optimum cytokine concentrations are provided in Supplementary Table 38.

We then tested the three-stage system for its ability to support T-lineage development. As a control, we used the first stage cytokine concentrations—similar to our unoptimized concentrations used previously—over the entire 42 days (Fig. 5a). Both the optimum and control conditions yielded comparable numbers of cells during the first 14 days, after which the number of lymphocytes in the control steadily decreased while the optimum plateaued but did not decrease (Fig. 5b). On day 42, the median absolute count of CD3⁺TCRαβ⁺ cells in the optimum was ~600-fold greater than the control: 1.02 M (95% CI: 0.13M–2.91 M) versus 1680 (340–4171) cells from 2000 HPSCs seeded per well at the start of the experiment (Fig. 5c). In both groups, CD3⁺TCRαβ⁺ cells were predominantly DP with a small number of CD4SP (CD4⁺CD8α⁻) and CD8SP (CD4⁻CD8α⁺) T-cells (Fig. 5d, e). We further characterized cells generated with the optimized cytokines. Of the CD3⁺TCRαβ⁺ population, 82.2 ± 3.6% (mean ± standard error) expressed the CD8αβ heterodimer and were transitioning from a CD28⁻CD27⁻ (33.9 ± 6.0%) through CD28⁺CD27⁻ (56.0 ± 6.6%) to CD28⁺CD27⁺ (9.2 ± 1.0%) phenotype (Fig. 5f). CD45RA was not expressed by any CD3⁺TCRαβ⁺ cells. Collectively, this positions the majority of cells as progressing through selection but not yet functionally mature³². Bulk VDJ sequencing of the TRB locus showed similar patterns of Vβ and Jβ gene usage to peripheral T-cells and postnatal thymocytes (Fig. 5g, h). Likewise, complementarity-determining region 3 (CDR3) lengths were similar between all T-cell sources (Fig. 5i, j). Neither TCR Vα24-Jα18 expressed by invariant NKT-cells nor TCR Vα7.2 expressed by mucosal-associated invariant T (MAIT)-cells were detected by flow cytometry (Supplementary Fig. 8). Thus, the three-stage optimum cytokines enhance survival and/or proliferation to provide a substantial increase in CD3⁺TCRαβ⁺ cells expressing a diverse TCR repertoire. These results validate the optimization methodology used for predicting dynamic cytokine signaling requirements during development.

Generated conventional T-cells secrete cytokines upon TCR stimulation

To measure T-cell function, cells generated using the three-stage ETN system were stimulated for 2 days with an anti-CD3 monoclonal antibody and IL-2 followed by 5 days with IL-2 alone. Of the resultant CD3⁺TCRαβ⁺ cells, 71.6 ± 8.5% were CD8SP T-cells that expressed variable levels of the CD8αβ heterodimer and CD8αα homodimer, consistent with conventional T-cells after TCR stimulation (Fig. 6a)³³. Compared to the human postnatal thymus, the frequency of CD27 expressing CD3⁺TCRαβ⁺CD8αβ⁺ cells was similar, while fewer ETN-generated cells expressed CCR7. However, the proportion of cells expressing CD45RA was markedly higher than thymocytes. A small population (2.5 ± 2.0%) of CD4SP T-cells was present with fewer expressing CD27 than ETN-generated CD8SP or CD4SP thymocytes (Fig. 6b). Subsequent stimulation with phorbol 12-myristate 13-acetate (PMA) and ionomycin induced IFNγ secretion in 50.2 ± 1.2% and IL-2 secretion in 3.8 ± 0.4% of CD3⁺ cells (Fig. 6c, d). We compared the cytokine secretion of ETN-generated T-cells to postnatal thymocytes and pan-CD3⁺ peripheral T-cells. The thymocytes and peripheral T-cells were first primed using anti-CD3 or anti-CD3/28 with IL-2 and then stimulated using PMA and ionomycin. Few postnatal thymocytes secreted IFNγ (6.5 ± 0.6%) or IL-2 (6.6 ± 0.6%). The frequency of IFNγ secreting peripheral T-cells (46.6 ± 3.1%) was comparable to generated T-cells, although peripheral T-cells secreted much more IL-2 (22.3 ± 2.1%). Thus, cells generated using this technology can mature into phenotypically elaborated CD8SP T-cells that are capable of secreting cytokines upon nonspecific TCR stimulation.