Thread by @pbachi

1/ Excited to share how T cell therapies kill #leukemia!! multi-omics + new #computational #singlecell tools for longitudinal analysis 👉unexpected answer! cell.com/cell-reports/f…

*👏* @elhamazizi! 🙏 @dpeer Cathy Wu @MDAndersonNews @CPRITTexas @ColumbiaBME @sloan_kettering

2/ We studied donor lymphocyte infusion (DLI) - an #Immunotherapy for relapsed #leukemia after #BMT & the #og of #celltherapy. Previously, we showed DLI reversed T cell exhaustion - but didn't know why/how/which T cells were responsible...
ashpublications.org/blood/article/…

3/ To address these ?'s, we modeled intraleukemic T cell dynamics by integrating longitudinal, multimodal data from ~100K T cells (!) during response (R) or resistance (NR: nonresponder) to DLI.

4/ Using Biscuit (doi.org/10.1016/j.cell…), we found 43 distinct T cell states defined by #clinical (timing relative to DLI, responder/nonresponder) & #biological (subtype, function & differentiation) features

5/ Standard techniques for data decomposition failed for gene programs ↔️ DLI outcome.💡! Factor analysis, explaining SHARED #variance, identified Factor 1 in Rs and 2+3 in NRs, correlating w/ activation/exhaustion vs multiple dysfunctional states (hypoxia, anergy, tolerance)...

6/ ...suggesting multiple pathways to DLI resistance. In fact, no cluster consistently enriched in NRs whereas four enriched in pre-DLI Responders. For Leo #Tolstoy fans 👉 #AnnaKarenina principle!

7/ In addition to these 4 activated/cytotoxic T cell states enriched at baseline, we found 2 clusters expanding in DLI responders exhibiting survival and self-renewal properties. But could we track these T cell states over time? How did their #dynamics relate to #leukemia burden?

8/ To account for variability in timing, total cell #, & cluster size on per-sample basis, we built hierarchical #Gaussian Process #regression model to capture temporal dependencies ➡️ 🔓 intraleukemic T cell dynamics for DLI response & resistance in patients! #MachineLearning🤘

9/ #Statistical models defined 2 diverging patterns of DLI response-assoc T cell states: contraction & expansion. scRNA + scprotein + dynamic data all 👉 'terminal' (Tex-like) and 'precursor' (Tpex-like) exhausted subset identities (prev id'd by @EJohnWherry, @UtzschneiderD & co)

10/ Important lesson: resolution of T cell exhaustion w/ DLI response not driven by changes in gene expression, but rather by shifts in cell type composition. Specifically, expansion of Tpex-like cells and contraction of Tex-like ones.

11/ Having identified the transcriptional states associated with DLI outcome, we sought to study their cell state specific regulatory circuitry. Bulk #ATACseq confirmed exhaustion-associated chromatin changes in responders.

12/ To uncover cell state specific gene regulatory networks (GRNs), Cassandra Burdziak, @elhamazizi, @dana_peer developed Symphony, a #Bayesian model that infers GRNs by integrating co-expression patterns of TFs & targets with chromatin accessibility

13/ Symphony unveiled network circuitries and the master regulators coordinating the underlying GRNs - many of these known to drive exhaustion (e.g. TOX, TCF7)!

14/ But from where are these post-DLI, expanding Tpex-like cells coming?? scTCR-seq of the DLI product detected very few of these cells...finding most of them to be pre-existing or recruited de novo to the bone marrow, *not* directly introduced by DLI!!

15/ What did we learn?
1. Terminal/precursor exhausted T cells define DLI response
2. DLI works through 'help' rather than direct transfer
3. #MachineLearning reveals cell dynamics (GPR models) & uncovers GRNs (Symphony) ➡️ template for longitudinal studies beyond #celltherapy

16/ Many many 🙏 to our #colleagues & #patients; to @CellReports @ASTCT @BeTheMatch @DamonRunyon @theNCI @ASH_hematology
👉👉 sites.google.com/view/tcellsdli
for data and computational tools!

17/ If you are interested in similar unbiased discovery approaches using #SingleCell multi-omics to identify #cancer-#immune interactions that impact #patient outcomes, come join @pbachi @MDAndersonNews or @elhamazizi @ColumbiaBME!!

*Major correction: 🙏 to @dana_peer !! (my apologies for glib assumption of non-existent autocorrect 😳)