Researchers at Columbia University’s Sadelain lab and University Hospital Tubingen used CRISPR gene editing to knock out the transcription factor NFIL3 in engineered CAR T cells, and the modified cells avoided the exhaustion that typically blunts their anti-tumor activity in animal models. The finding, reported in Cancer Discovery, targets an internal driver of T-cell dysfunction rather than a surface checkpoint receptor, offering a distinct strategy at a time when CAR T therapies against solid tumors still fail primarily because the cells burn out too quickly.
Why an Internal Exhaustion Driver Changes the CAR T Calculus
CAR T-cell therapy has produced durable responses in blood cancers, but solid tumors remain a different problem. Tumor microenvironments push T cells into a dysfunctional state marked by inhibitory receptors and suppressive cytokine production. Most efforts to counteract that process have focused on blocking surface molecules like PD-1. A preclinical study in glioblastoma models, for instance, showed that CRISPR disruption of PD-1 improved CAR T-cell activity against EGFRvIII-expressing tumors. The NFIL3 approach works further upstream, inside the cell’s own gene-regulatory machinery, which means it could sidestep some of the limitations of receptor-level interventions.
NFIL3 sits at the end of a signaling chain that starts with the cytokine IL-27. Earlier mechanistic work established that this IL-27–dependent pathway drives expression of Tim-3 and IL-10, both associated with impaired T-cell function in vivo. By removing NFIL3 entirely, the Columbia and Tubingen teams aimed to cut off that dysfunction program at its transcriptional source rather than blocking individual downstream molecules one at a time.
A key question is whether the benefit of NFIL3 knockout depends on how the CAR itself is built. CAR constructs typically include either a CD28 or a 4-1BB costimulatory domain, and those two designs push T cells toward distinct exhaustion trajectories with separate chromatin states. The institutional summary from Columbia noted that the NFIL3 findings showed distinct effects across these two costimulatory architectures. That observation raises the possibility that NFIL3 knockout may interact more strongly with 4-1BB-containing CARs. The 4-1BB domain already biases cells toward chromatin configurations that could amplify NFIL3-dependent exhaustion programs, so removing the transcription factor in that context might produce a larger rescue effect. Confirming or refuting that hypothesis will require head-to-head data comparing the two designs with and without the knockout, data that has not yet been made publicly available in granular form.
Genome-Wide Screens and the CELLFIE Platform Built the Road Here
The NFIL3 finding did not emerge in isolation. It belongs to a growing wave of screen-driven gene editing strategies aimed at identifying which genes, when disabled, allow T cells to persist longer and kill more effectively. A genome-wide CRISPR screen published in Cancer Cell used an in vitro exhaustion model to identify chromatin regulators that limit T-cell persistence, then validated top hits in tumor-bearing mice. That work demonstrated the basic logic: screen thousands of gene knockouts, find the ones that resist exhaustion, and test them in animals.
A separate large-scale platform called CELLFIE, described in Nature, took the concept further by performing high-content CRISPR screening directly in primary human CAR T cells and validating hits in xenograft models. Together, these efforts have created an expanding catalog of genetic targets that could be disabled to make CAR T cells more durable. NFIL3 joins that catalog with a specific mechanistic rationale tied to the IL-27 signaling axis, distinguishing it from chromatin-level targets identified through unbiased screens alone.
The practical significance is that researchers now have multiple, potentially combinable edits. A future CAR T product might carry not just NFIL3 knockout but also disruptions to chromatin remodeling factors or checkpoint receptors, stacking interventions that address different layers of the exhaustion problem. Whether such combinations improve outcomes without introducing safety risks is an open experimental question.
Missing Data and the Distance Between Mouse Models and Patients
Several gaps limit how far the current evidence can be pushed. No primary data tables showing exact tumor volumes, survival curves, or statistical comparisons from the NFIL3 knockout experiments have been released publicly beyond the Cancer Discovery publication and Columbia’s institutional announcement. Without those numbers, independent assessment of effect size is not possible. The reported animal studies indicate delayed tumor growth and improved survival, but the magnitude of benefit, variability between animals, and any dose–response relationships remain unclear outside the published figures.
Direct statements from the Sadelain lab or Tubingen collaborators have emphasized that the work is preclinical. The modified CAR T cells have so far been tested in mouse models and ex vivo human systems, not in patients. Mouse tumors, even when engineered to mimic human cancers, differ in antigen expression, immune contexture, and stromal architecture. Many interventions that reinvigorate T cells in mice lose potency in the setting of human solid tumors, where antigen heterogeneity, physical barriers, and systemic immunosuppression all conspire against durable responses.
There are also safety questions that animal data alone cannot fully resolve. NFIL3 participates in broader immune regulation, including natural killer cell biology and circadian-linked transcriptional programs. Knocking it out in therapeutic T cells could, in principle, alter cytokine production, tissue homing, or interactions with other immune subsets in ways that increase the risk of cytokine release syndrome, neurotoxicity, or off-tumor inflammation once the cells are infused into patients. Early mouse studies are not powered to detect rare but severe toxicities that might emerge in a larger clinical population.
Regulators and clinicians will therefore look for careful stepwise development. Before a first-in-human trial, researchers will need toxicology studies in more than one animal model, detailed pharmacokinetics on CAR T-cell expansion and persistence, and assays that track NFIL3-independent exhaustion pathways that might compensate over time. If NFIL3 knockout simply delays, rather than prevents, dysfunction, patients could still experience relapse after an initial response.
How NFIL3 Knockout Fits into the Future of Engineered T Cells
Even with these caveats, the NFIL3 story illustrates how the field is moving from single-gene tweaks toward rationally layered engineering. One can imagine a CAR T product built on a 4-1BB backbone, edited to remove NFIL3 and a selected chromatin regulator, and paired with a logic-gated CAR design that restricts activation to tumor tissue. Such a cell could, in theory, resist exhaustion, maintain effector function in hostile microenvironments, and limit collateral damage to normal tissues.
Realizing that vision will demand more than clever molecular biology. Manufacturing workflows will have to accommodate multiplex CRISPR editing, rigorous off-target analysis, and potency assays that capture subtle differences in exhaustion resistance. Clinical trial design will need to separate the contributions of each edit, at least in early-phase studies, so that unexpected toxicities or failures can be traced back to specific engineering choices.
For now, NFIL3 knockout is best viewed as a promising new lever rather than a solved solution. It adds mechanistic depth to the understanding of how cytokine signaling imprints exhaustion programs on CAR T cells and offers a tractable target for intervention. Whether that intervention will translate into longer-lasting, safer responses in patients with solid tumors is a question that only carefully staged clinical testing can answer.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
