Fig 1: Multiple genome-wide CRISPR screens for T cell resistance.a, Dropout of essential genes19 across screen conditions. X-axis is scaled and binned log2 fold change of CFSE low over CFSE high cells in each screen, Y-axis is the number of essential genes in each bin. As expected, essential genes tend to have a negative LFC in these CRISPR KO screens of primary human T cells (n = 4 human donors for Stim and Tregs screens, n = 2 for Adenosine, Cyclosporine and Tacrolimus screens, and n = 1 for the TGFß screen. Normalized Enrichment Score (NES) and adjusted p-value by GSEA and two-sided permutation test). b, Screen hits are expressed in human T cells. X-axis is scaled and binned log2 fold change in each screen, Y-axis shows the expression in activated human T cells20. Both negative and positive hits tended to be highly expressed in human T cells, suggesting these pooled KO screens point to relevant T cell biology (n = 4 human donors for Stim and Tregs screens, n = 2 for Adenosine, Cyclosporine and Tacrolimus screens, and n = 1 for the TGFß screen, dots are mean +/- SEM). c, Shared hits (y-axis) (z-score > 1.5, methods) across the screen conditions (x-axis) including hits unique to each individual screen. Shared hits for each subset are detailed in Supplementary Table 1. d, Heatmap of the pairwise Pearson’s correlation coefficient for gene-level z-scores for all screen conditions. e, Volcano plots showing p-value (MAGeCK RRA one-sided test and methods) on the y-axis and gene-level z-scores on the x-axis, comparing highly dividing cells in each suppressive condition to highly dividing in the vehicle condition. Highlighted are genes found to be specific to adenosine and TGFß screens, selected for further validation. f, Gene targets from screens were selected as either general (PAN) or more specific to certain suppressive contexts and were knocked out individually in T cells. CFSE stained, Cas9 RNP-electroporated edited T cells were stimulated and cultured in the different suppressive conditions. Percent of cells proliferating for each gene KO compared to control cells are displayed for each suppressive condition (n = 2 donors, 2 sgRNAs per gene target in triplicates. We highlighted gene KOs found to confer significant resistance in predicted conditions (adenosine, TGFB, and calcium/calcineurin inhibitors – Tacrolimus, and Cyclosporine), using a cut-off of FDR adjusted p-value < 0.05. For clarity, displayed are the significant p-values for gene targets according to their suppressive screen condition of origin, but results for all genes across conditions are detailed in Supplementary Table 3). As expected, ADORA2A, TGFBR1 and TGFBR2, FKBP1A, and PPIA KOs conferred resistance in the adenosine, TGFB, Tacrolimus, and Cyclosporine conditions, respectively. PDE4C and NKX2-6 KOs were found to confer relatively selective resistance in the adenosine condition, and NFKB2 KO was found to increase resistance in the calcineurin inhibitor (tacrolimus and cyclosporine) conditions. TMEM222, while scoring very highly in the screens, did not increase proliferative advantage in this arrayed validation (dots are individuals replicates, black vertical lines are the mean, *p < 0.05, **p < 0.01, ***p < 0.001 and ***p < 0.0001 for two-sided unpaired Student’s t-test). g, Log fold change (LFC) of guides targeting RasGAP genes or the RasGEF RASGRP1, across the different suppressive screen conditions shown here. h, Expression levels (scaled to minimum of 0 and a maximum of 1) of the RasGAP family members available in the BioGPS dataset27, including RASA2, across healthy human tissues revealed RASA2 as selectively expressed in CD8+/4+ human T cells. Data also shown for RASGRP1, a RasGEF with defined roles in TCR signaling and an expression pattern strikingly similar to that of RASA2. Source data
Fig 2: DDX3 mediates NEU3 upregulation by TGF-β1 and LAP. Human lung fibroblasts were pre-incubated with either 10 µM of RK-33 or an equal volume dilution of DMSO in protein free medium for 10 min. Protein free medium with or without (A) 10 ng/mL TGF-β1 or (B) 0.3 ng/mL LAP was added to wells for 5 min. Cells were then washed, fixed, and stained for NEU3 using immunofluorescence. RK-33 treated cells are indicated with red bars. Values are normalized to 0 min controls. Bars are mean ± SEM, n = 3–4 (1–2 male and 2 female cells lines). ** p < 0.01, *** p < 0.001 (One-way ANOVA, Dunnett’s test compared to 0 min control). Human lung fibroblasts were treated with (C–E) 10 ng/mL of recombinant human active TGF-β1 or (F–H) 0.3 ng/mL of recombinant human LAP for the indicated times. m indicates minutes, h hour. The 0 min samples were treated with protein free media only. Phosphorylated proteins were isolated using TALON PMAC magnetic beads. Phosphorylated proteins and supernatants of total cell lysates were stained for DDX3 using Western blots. Blots are representative of five donors (four male and one female). (D,E,G,H) Quantification of Western blots. Band densities of phosphorylated and total DDX3 were normalized to the integrated density of silver- or Coomassie-stained gels respectively of total protein for each sample (Supplementary Figure S7), then values were normalized to the 0 min control. Blue points indicate the averages for male cell lines and red points indicate the averages for female cells, circles represent healthy cell donors and triangles represent cells from IPF patients. Bars are mean ± SEM, n = 5 (4 male and 1 female). * p < 0.05, ** p < 0.01, *** p < 0.001 (One-way ANOVA, Dunnett’s test compared to 0 min control).
Fig 3: Proposed model of rapid NEU3 upregulation. Active TGF-β1 and LAP increase DDX3 interaction with NEU3 mRNA either by increasing binding to the common motif in NEU3 mRNA, found in 179 other proteins that are translationally regulated by TGF-β1 (orange), or to G-quadruplex structures (green; the image labels 4 of the 32 predicted G-quadruplex structures). DDX3 stimulates rapid NEU3 translation. Once translated and released from cells, NEU3 activates latent TGF-β1 (L-TGF-β1), releasing active TGF-β1 from its complex with LAP. TGF-β1 and LAP synergistically upregulate NEU3 translation leading to a positive feedback loop. DDX3 is inhibited by RK-33, and NEU3 is inhibited by 2-AP and AMPCA, all three of which can disrupt this loop and the rapid 5 min upregulation of NEU3. This figure was created using BioRender elements (https://biorender.com/ (accessed on 16 July 2025)); structures of NEU3 and TGF-β1 were predicted using AlphaFold 3 [75].
Fig 4: Latency-associated peptide upregulates NEU3 at 5 min and functions synergistically with TGF-β1. Human lung fibroblast cells were exposed to the indicated concentrations of (A) LAP alone and (B) suboptimal concentrations of LAP, TGF-β1, and a combination for 5 min, fixed, and stained for NEU3 using immunofluorescence. Blue points indicate the averages for male cell lines and red points indicate the averages for female cells, circles represent healthy cell donors and triangles represent cells from IPF patients. Values are normalized to 0 ng/mL control. Bars are mean ± SEM, n = 5–11 (4–6 male and 1–3 female cells lines). * p < 0.05, ** p < 0.01 (One-way ANOVA, Dunnet’s test compared to 0 ng/mL control).
Supplier Page from BioLegend for Recombinant Human TGF-β1 (carrier-free)