PREreview of Cell state and transcription factor modulation during extended ex vivo CD8⁺T-cell expansion

Brief summary of the study - a sentence summarizing the study and general comments that apply across the full paper

The authors generated a bulk RNA-seq dataset of ex vivo expanded, TCR stimulated primary CD8 T cells for 7 days followed by 10-day resting, and integrated this dataset with published scRNA-seq dataset of T cells from healthy blood donors. Upon pathway enrichment and gene regulatory network analysis, the authors discovered a T cells phenotype spectrum spanning from naive, memory states to exhausted T cell states. The authors userRegulatory network inference via CellOracle and in silico perturbation modeling to infer transcription factor (TF) activity and downstream signaling along T cell differentiation trajectory and uncovered AP1 complexes to be important during T cell activation. Over-expression of FOSL1 enhances CD8+ T cells effector phenotype and killing capacity. This paper serves as a valuable resource and provides more granular insights into the driving forces of ex vivo CD8 T cell differentiation thanks to the increased sequencing depth and temporal resolution of the presented bulk RNA-seq dataset. Although there remains room for functional validation of other candidate TFs during T cell differentiation in vivo and in vivo, the knowledge from this manuscript set up a system to study TF regulation regulation in T cell activation trajectory and lead to valuable insights in improving adoptive cell therapy.

Major comments  - Comments on the validity or strength of the methodology, experiments and analyses, strength of the conclusions

• The figure 1D and figure 2C used different gene markers to annotate T cell exhaustion phenotype. It confused readers that day 7 has the most exhaustion genes in figure 1D but the latest exhaustion gene in figure 2C. Maybe the authors could use consistent and authentic markers to annotate the same phenotype.

• In the figure 4, the authors only performed experiments in the early timepoints. However, it is unclear if the FOSL1 overexpression still enhances CD8+ T effector cell phenotype even after 7 days + 10 breaks ex vivo expansions. Maybe the authors could repeat the experiments to include the later time points.

• The authors determined TFs as an alternative method to enhance ACT, however the rationale of the monitoring TFs at different T cell stages have not been discussed thoroughly. The authors can consider adding more background and reasoning into the rationale behind the TFs focus, comparing with other investigated avenues to enhance ACT, in order to justify the major theme and subject of this manuscript. 

Minor comments - Clarifications to statements in the text, interpretation of the results, presentation of the data/figures

• Figure 1A and B. PCAs are provided without percentages for the individual PCs. The authors may want to describe these, because the interpretation of the results could change depending on how much of the variation in the data is explained by each PC. 

• Figure 2A. the dot size from “<0.25” to “<0.05” groups are too similar to each other, so it is hard to distinguish their differences. Maybe the authors could increase the dot size differences between different groups. Alternatively, the authors can use color gradients to indicate the differences.

• The authors could label a few top genes in figure 2B just like figure 3G.

• Figure 2C. The authors should label the starting time of T cell rests.

• Figure 3A. What does each color and cluster indicate? And what is the direction of the trajectory? Maybe the authors could label the beginning and end of the developmental trajectory. And are there overall T cell states that are conserved across naive and memory T cells and is this why some of the color codes overlap?

• Figure 3B. Because many people might not be familiar with CellOracle’s in silico perturbation mode. Can the authors please briefly explain in the figure legend how the random and negative perturbation scores are calculated and what they mean? Is the set of perturbed TFs conserved across memory and naive T cells i.e. equivalent to the top right quadrant in Figure 3C?

• Figure 3E. The authors may want to label the x axis.

• Figure 3G. Although this is the same plot as in Figure 2B (if we are not mistaken), the authors should include the color legend. What does the dot size mean? This should be explained in some sort of legend as well.

• Figure 4. in the text the authors mention they had generated FOSL1-KO Jurkat T cells as well. Why did they not include those in their functional in vitro studies?

• Figure 4 - was the different stimulation assay (compared to anti-CD3/CD28 beads in prior figures) merely used to be able to tailor signaling strength by pre-treatment with anti-CD86 or was there another reason?

• Figure 4D. authors have described color-coded subsets of genes. However, it is not mentioned what each color represents. Moreover, when performing pathway analysis, the authors have already observed marked differences upon FOSL1 over-expression, which they also find upon stimulation. The authors could further resolve whether IL-2 and NFKB signaling (or other potentially interesting pathways) are truly enriched upon stimulation and not solely from over-expression of FOSL1.

• In general, figure labels are often quite small and hard to read. Maybe the authors could enlarge them.

Comments on reporting - information on the statistical analyses or availability of data.

• The authors have included descriptive statistics but they could consider further incorporating inferential statistics to strengthen the conclusions. For example, they could use statistical tests to compare gene expression levels between different time points or groups.

Suggestions for future studies

• In figure 3-4, although the authors did a bulk RNAseq over 17 days of ex vivo T cell expansions, they are only able to predict and validate roles of TFs in the very early time points (within 1 day). Maybe the authors can analyze a scRNA-seq of “7 days” and “7 days + 10 break” either from published datasets or from their own dataset? With CellOracle, it may enable the prediction of TFs that transit exhausted T cells (7days) to less exhausted T cells (7days+10 days).

• If available, perhaps the authors could have integrated additional scRNAseq datasets from different contexts (viral infection, tumor models etc.) to see how ex vivo expanded/stimulated CD8 T cells may phenocopy (or not) physiologically relevant T cell states and which TFs may be differentially regulated in different disease contexts.

• CellOracle can also simulate over-expression. Owing to their observation that FOSL1 KO do not differ from its parental cell line, it would be very interesting to look into in silico overexpression that drive T-cell activation.

• The authors may want to consider TF Network analysis TF Network Analysis in which they can delve more into the analysis of transcription factor networks, including the use of computational tools and databases for network inference and visualization.

• In ACT, secondary T-cell malignancies induced by CAR T cell treatments are more and more discussed. Maybe the authors could look into the possibilities of their platform in this regard or speculate?

• The authors may want to use an in vivo model to test the enhanced killing capacity of engineered CD8 T cells (perhaps a mouse melanoma model, where the antigen GP100 is expressed in the tumor cells).

• The authors may look into the epigenetic Landscape and metabolic reprogramming: to investigate how epigenetic modifications contribute to the establishment, maintenance of different T-cell subsets, and explore the role of metabolic pathways in T-cell activation, proliferation, and effector function. The authors can also consider looking into other TFs candidates presented from the RNA-seq dataset in this manuscript. 

Conflicts of interest of reviewers

• There are no conflicts of interest to declare.

Inline commenting section

Please add comments on the preprint below via comments. You can add comments on the full paper, sections or only individual fragments. Any comments added here will be reviewed for inclusion in the public review section if relevant, but will not be posted publicly in any way that can identify the commenter for individual comments.

• (Abstract) The claim that producing "super engager-like" T cells is based solely on gene expression signatures and enhanced cancer-cell killing capacity might be premature.  if there is more concrete pieces of evidence of in vivo efficacy or clinical relevance, it would strengthen the conclusion

• (Introduction) Consider reorganizing the introduction in order to improve the narrative flow, and consider rearranging the sections on single-cell RNA-seq limitations and the need for a temporal gene expression map.

• (Introduction - paragraph 2 from ”T-cell optimization has ..” to “for tailoring ACT”) These two sentences are very vague and make it difficult for the reader to understand their meaning. Maybe the authors could:

  • Explain the function of the lost proteins, especially readers outside the T-cell field won't know RASA2 or PTPN2 and the effect of their loss in T cells.

  • Which proteins are used for gains of function? In the cited screen, CRISPR-Cas9 (what's CRISPRa?) was used to knock out proteins. I can find no mentioning of a gain of function. 

  • "modifying TF expression in T cells" - please specify which TFs the authors mean specifically (activating? Promoting proliferation?), or give an example.

• (Introduction - paragraph 3) Why do the authors focus on TFs alone here? Knowledge e.g. from chromatin research could also be helpful in understanding gene transcription. Also, RNA expression patterns from downstream genes often allow for conclusions on the upstream TFs, especially the "main" ones. The purpose of their study is to look into less well-known TFs, I assume, but maybe the other methods should be taken into consideration and/or outlined why TFs are especially important?

• (Introduction - paragraph 4) The authors could expand on clinical implications in that they can emphasize the potential impact of the research on patient outcomes. How could the findings translate into improved ACT therapies? 

• (Introduction - paragraph 4 from “By integrating our bulk …”) The authors should clearly state the expected outcomes of the study. For example, "We hypothesize that by mapping gene expression trajectories during T-cell expansion, we can identify key TFs that regulate critical cell states and that manipulating these TFs will enhance T-cell function and improve ACT efficacy."

• (Result - paragraph 1) The authors should mention the origin of cells also in the results (e.g. extracted from healthy PBMCs).

• (Result - paragraph 1) The authors should rephrase the line “ For each time-point we extracted total RNA and performed bulk RNA-seq in three technical replicate” into “We extracted ...for each time point”

• (Figure 1) Consider complementing it with other visualization techniques like t-SNE or UMAP for a more comprehensive view.

• (Figure 1C Here, the authors could map their transcriptomes to most major clusters in the dendrogram. However, there are several clusters (e.g., CD8 EMRA) that does not resemble any timepoint. The authors could speculate why this could be so.

• (Result 1 - paragraph 2) While the study describes phenotypic changes, delving deeper into the underlying molecular mechanisms would enhance its impact. For instance, analyzing transcription factor binding sites or performing pathway enrichment analysis could provide clues about the regulatory networks involved.

• (Result 2 - paragraph 1) The phrasing “signalling components (e.g., TCR signalling, MAPK signalling, JAK-STAT signalling, Hippo signalling, NF-kappa B signalling) were significantly upregulated” is misleading. Like this, it could be interpreted that ALL signalling components are upregulated at 3h, for example TCR signalling etc. However, e.g. p53 signalling is more upregulated at later time points and we wonder if the authors investigate all signalling pathways that come to mind, not all will be upregulated. Just as an example, did the authors look into DNA damage signalling and cGAS STING? 

• (Result 2 - paragraph 3) Functional validation in correlating gene expression changes with functional outcomes is crucial. Validating the identified gene expression patterns through functional assays (e.g., cytokine production, cytotoxicity) would strengthen the conclusions. For instance, overexpressing or inhibiting key genes identified in the gene expression patterns could help elucidate their functional roles.

• Figure 3G needs a better description, what are the regulators there, different classes by different colors, size or shape ?

• (Discussion paragraph 4) The authors wrote “we show that FOSL1 over-expressing cells are not cytotoxic without engagement of the TCR, offering reassurance that FOSL1 would not engender non-specific hyperactivity”. Could the authors speculate on the application of this knowledge? Should future ACT cells be genetically manipulated to overexpress FOSL1, or should one select T cells with increased FOSL1 for treatment?

• (Discussion paragraph 5) Exploring transcription factor networks, epigenetic modifications, and post-translational modifications can provide valuable insights into the regulatory circuitry governing T-cell differentiation.

• (Limitations) The use of anti-CD3/CD28 beads for T-cell activation could introduce biases. Discussing potential limitations and alternative activation methods would enhance the study's credibility.

Competing interests

The authors declare that they have no competing interests.