The Tox Gene Encodes Two Proteins with Distinct and Shared Roles in Gene Regulation

This article is featured in Top Reads, p. 1851

Thymocyte selection–associated HMG-box protein (TOX) is a DNA-binding protein with essential roles in lymphocyte development and function. TOX is required for development of the CD4⁺ T cell lineage in the thymus and development of most innate lymphoid cell subsets in the bone marrow (1–4). Although TOX is not expressed in most naive T cells, its expression induces and enforces a developmental exhaustion program in CD8⁺ T cells experiencing chronic Ag exposure (5–7). Exhausted T cells have decreased cytotoxic function, develop at the expense of memory cell formation, and are a major barrier to immune system control of tumors and chronic infections. Although there has been much recent interest in TOX function, very little is known about TOX at the protein level.

We show here that the Tox gene encodes two proteins, despite a single mRNA. Mechanistically, the diversification of the TOX protein occurs via an alternative translation mechanism known as “leaky ribosomal scanning.” We also find that in a model of chronic stimulation of CD8⁺ T cells leading to phenotypic exhaustion, overexpression of the long form of TOX induces the expression of a unique gene set associated with regulation of the cell cycle.

OTI [C57BL/6-Tg(TcraTcrb)1100Mjb/J] mice and OTII [B6.Cg-Tg(TcraTcrb)425Cbn/J] mice were bred at Cedars-Sinai Medical Center or ordered from The Jackson Laboratory (Bar Harbor, ME) and kept under specific pathogen-free conditions. Mice used for experiments were 4–12 wk of age and of either sex, with the exception of OTII mice, which were solely males as the transgene integrated on the Y chromosome. All animal procedures were performed in accordance with protocols approved by the Cedars-Sinai Medical Center Institutional Animal Care and Use Committee.

Mutant constructs were generated using the Invitrogen GeneArt Site-Directed Mutagenesis Kit or the QuikChange II XL Site-Directed Mutagenesis Kit (Agilent, Santa Clara, CA) according to the manufacturer’s instructions. HEK293T or HEK293F cells were transfected using Xtreme Gene HD (Roche) or FuGENE HD (Promega, Madison, WI) at a 1:3 plasmid DNA to reagent ratio.

Cell sorting was conducted by the Cedars-Sinai Flow Cytometry Core on a BD FACSAria III or FACSAria II, and FACS analysis was performed on an LSRII (BD Biosciences, San Jose, CA). Abs against TOX (TXRX10), CD4, CD8, CD3e, CD69, TIGIT, and LAG3 were from Thermo Fisher Scientific (San Diego, CA), anti-histone 3 was from Cell Signaling Technology (Danvers, MA), and anti-PD-1, isotype control, and Zombie live/dead stain were from BioLegend (San Diego, CA). The polyclonal TOX_FL Ab was raised against the N-terminal peptide sequence (MDVRFYPPPAQPAAAPAAPC) and produced by ProSci (Poway, CA). Data analysis was completed with FlowJo version 10 software (BD Biosciences, Ashland, OR).

Chronic stimulation of CD8⁺ T cells in culture was performed as described previously (8). Briefly, on day 0, naive CD8⁺ T cells were obtained via negative selection of OTI splenocytes using the EasySep Mouse Naive CD8⁺ T Cell Isolation Kit (STEMCELL Technologies, Vancouver, BC, Canada). Cells were cultured in RPMI 1640, penicillin/streptomycin/glutamine, 1% HEPES, 1% 100 mM sodium pyruvate, 1% nonessential amino acid (all from Life Technologies, Thermo Fisher Scientific), 10% FBS (Omega Scientific), 0.05 mM 2-ME (Sigma-Aldrich, St. Louis, MO), 5 ng/ml IL-7 (PeproTech, Cranbury, NJ), and 5 ng/ml IL-15 (BioLegend). Treatment groups included naive, acutely stimulated, and chronically stimulated cells. Naive cells were cultured in complete media only. Acutely stimulated cells were treated on day 0 with 10 ng/ml OVA_(257-264) peptide (Sigma-Aldrich), washed on day 2, and cultured until day 5 in complete media. Chronically stimulated cells were treated with 10 ng/ml OVA peptide on days 0 and 1, washed on day 2, and treated with OVA on days 2–4 (total of five stimulations). All cells were collected on day 5 for analysis.

Retroviral constructs were created by insertion of TOX, M1T, or M(40,42,44)I sequences into p-MIG-enhanced GFP (EGFP) retroviral vectors. Platinum-E Retroviral Packaging Cells (Cell Biolabs, San Diego, CA) were transfected in a 100-mm dish using FuGENE HD Transfection Reagent (Promega) with 10 μg of plasmid DNA. Viral supernatant was collected after 48 h. On day 0, naive CD8⁺ T cells were obtained via negative selection of OTI splenocytes and cultured in complete RPMI with IL-7, IL-15, and OVA peptide as above. Twenty-four hours later, cells were washed and resuspended in viral supernatant supplemented with IL-7, IL-15, and 8 μg/ml polybrene (Sigma-Aldrich). Cells were spin infected at 34°C for 75 min at 2000 × g. Cells were returned to the incubator for 4 h, washed, and resuspended in complete media with cytokines and OVA as above. Cells were treated with OVA again on day 2, media were changed on day 3, and cells were collected for analysis on day 5.

Whole-cell lysates were made in radioimmunoprecipitation assay buffer containing Pierce HALT Protease and Phosphatase Inhibitors (Thermo Fisher Scientific). For T cells, 250,000 cell equivalents were loaded into each well. For HEK293F lysates, 10 μg of protein was loaded. Lysates were resolved by SDS-PAGE and transferred onto nitrocellulose membranes using the Transblot Turbo Transfer System (Bio-Rad Laboratories, Hercules, CA). Membranes were incubated in 5% milk TBST and incubated with primary Abs overnight. Target proteins were visualized with appropriate HRP-conjugated secondary Abs (Thermo Fisher Scientific) and Pierce ECL Western Blotting Substrate (Thermo Fisher Scientific). Blots were imaged and analyzed using a ChemiDoc XRS (Bio-Rad Laboratories) system.

The Tox gene was cloned into pSP64 Poly(A) Vector (Promega), and protein was produced using the TNT SP6 High-Yield Wheat Germ Protein Expression System (Promega) according to the manufacturer’s instructions.

Raw sequencing data were demultiplexed and converted to fastq format by using bcl2fastq version 2.20 (Illumina, San Diego, CA). Reads were aligned to the transcriptome using STAR (version 2.6.1)/RSEM (version 1.2.28) with default parameters using a custom mouse GRCm38 transcriptome reference downloaded from http://www.gencodegenes.org, containing all protein coding based on mouse GENCODE version 24 annotation (9, 10). Expression counts for each gene in all samples were normalized by a modified trimmed mean of the M-values normalization method. Each gene was fitted into a negative binomial generalized linear model, and the Wald test was applied to assess the differential expression between two sample groups by DESeq2 (version 1.26.0) (11). The Benjamini-Hochberg procedure was applied to adjust for multiple hypothesis testing, and differentially expressed gene (DEG) candidates were selected with a false discovery rate less than 0.05. DEG candidates were used for Gene Ontology/Kyoto Encyclopedia of Genes and Genomes enrichment analysis performed with ClusterProfileR (12).

Mean values, one-way ANOVA, and the p value associated with a post hoc Tukey’s test were conducted for groups of three or more. p Values <0.05 were considered to represent means with a statistically significant difference (throughout all figures, *p < 0.05, ** p ≤ 0.01, and *** p ≤ 0.001). Sample or experiment sizes were determined empirically for sufficient statistical power. Outliers were identified using the interquartile range rule with a 1.5 constant.

The annotated Tox coding sequence is predicted to produce a 526-aa protein (TOX_FL) (13). However, when the annotated cDNA is introduced into HEK293 cells, which do not endogenously express TOX, Western blots reveal two bands using an mAb specific for the C-terminus of murine TOX (aa 336–526) (Fig. 1A). The proteoforms are not alternatively spliced variants, because both forms arise from a single cDNA, and no cryptic splice sites have been identified. Previous studies (not shown) also failed to identify post-translational modifications that could account for such a large shift in electrophoretic mobility. Thus, we hypothesized that the faster-migrating band represented a truncated proteoform.

Alternative protein translation initiation is an important generator of proteome diversification. In healthy eukaryotic cells, most mRNAs are translated canonically by a scanning-dependent mechanism that requires the presence of a 5′ N7-methylated guanosine cap (14, 15). The 5′ N7-methylated guanosine cap also functions as the recruitment site for the 40S ribosomal subunit, a methionine-conjugated initiator tRNA, and a host of eukaryotic initiation factors (eIFs) that comprise a translation “preinitiation complex.” Inherent variability in the 5′ untranslated region requires the small ribosomal subunit to proceed in the 5′-to-3′ direction base by base, scanning the mRNA until it recognizes an AUG start codon. Although the most 5′ AUG is often the favored start codon, the sequence surrounding these initiation codons (Kozak sequence) affects the efficiency of initiation by eukaryotic ribosomes (16, 17).

The murine Tox mRNA coding sequence contains three downstream AUG codons (nucleotide positions 118–120 encoding methionine [M] 40, 124–126 encoding M42, and 130–132 encoding M44) that could potentially act as downstream translation start sites. Translation at these sites would produce N-terminally truncated proteins 487, 485, and 483 aa in length (Fig. 1B). Significantly, these AUGs are also conserved in human and zebrafish TOX genes, consistent with functional and/or structural importance of the encoded residues (Fig. 1C). The apparent m.w. change of the two TOX forms by Western blot analysis is also generally consistent with a loss of ∼40–50 aa.

On the basis of these observations, we investigated whether the slower-migrating protein band was TOX_FL translated from the annotated coding sequence translation initiation sequence (TIS_CDS) and whether the faster-migrating band was an N-terminally truncated proteoform produced by internal translation. In this regard, we used site-directed mutagenesis to create a series of TOX mutant plasmids, which we introduced into HEK293F cells and analyzed the expressed proteins by Western blotting (Fig. 1A, 1D). Elimination of the annotated start codon by mutating the cDNA from ATG to ACG, encoding a threonine (M1T), specifically eliminated expression of the higher m.w. band (Fig. 1A, 1E). When the TIS_CDS was present but codons encoding M40, M42, and M44 were all mutated to encode isoleucine or glycine, only the higher m.w. band was detected (Fig. 1A, 1E). Therefore, we conclude that TOX_FL is translated from TIS_CDS, and the faster-migrating proteoform (hence termed “TOXΔN”) is alternatively translated from an internal AUG.

To ensure that TOXΔN production is not dependent on the presence of TOX_FL or the upstream nucleotide sequence (with the exception of those required for the M40 Kozak sequence), we created two additional mutants to eliminate TOX_FL without eliminating the TIS_CDS (Fig. 1D). With a deletion-induced frameshift mutation at the R4 position, TOX_FL is not produced (although a 41-aa mutant protein could be translated from the TIS_CDS), but expression of TOXΔN is maintained, indicating that the translation machinery does not require the TIS_CDS as a context for downstream initiation (Fig. 1F, 1G). We further eliminated the possibility that TOXΔN is a post-translationally truncated proteoform of TOX_FL by mutating the C20 position to a stop codon. This eliminated production of TOX_FL but not TOXΔN (Fig. 1F, 1G). Finally, we produced an Ab specific for the TOX N-terminal peptide. This Ab detected only TOX_FL (Fig. 1H).

Internal translation initiation of a single reading frame could result from scanning-independent or scanning-dependent mechanisms. In terms of the former, mRNA secondary and tertiary structures can recruit the ribosome directly to an internal site (internal ribosome entry site [IRES]) (18). Alternatively, scanning-dependent mechanisms include ribosome reinitiation, TISU (translation initiator of short 5′ untranslated region)-mediated translation, and leaky ribosomal scanning. Ribosome reinitiation is unlikely because it most commonly occurs on mRNAs that use short upstream open reading frames to regulate the translation of the downstream product (19–21). Additionally, the Tox gene lacks a TISU (22). Given the potential for alternative in-frame starts in a single open reading frame in Tox, leaky ribosomal scanning remained a likely possibility (23).

To definitively distinguish scanning-dependent translation and IRES-mediated translation of TOXΔN, we inserted a 48-bp stable stem and loop structure 5′ of the TIS_CDS in a TOX expression plasmid (Fig. 2A) (24). This plasmid also expresses EGFP downstream of an IRES sequence (Fig. 2A). Complex secondary mRNA structures are sterically blocked from entering the mRNA channel on the solvent side of the small ribosomal subunit, preventing progression of the ribosome, and thus would be expected to inhibit expression of TOX_FL (25, 26). HEK293F cells expressing TOX, the TOXΔN mutant, and the hairpin mutant had similar expression of EGFP by FACS, consistent with IRES-mediated translation of the fluorescent reporter (Fig. 2B). In contrast, protein expression of TOX_FL was inhibited by the hairpin, as expected by scanning-dependent translation (Fig. 2C). However, the hairpin also inhibited expression of TOXΔN, demonstrating that its translation was also scanning dependent and not IRES mediated (Fig. 2C).

Leaky scanning occurs when the most 5′ TIS is embedded in a Kozak sequence of reduced efficiency (16, 27). When sampled, the preinitiation complex can either initiate at the suboptimal TIS or scan past it (“leak”) and initiate at a downstream site. As such, the most proximal start site is typically highly efficient to prevent leaky scanning (27). Recently, Noderer et al. determined the translation initiation efficiency for all possible AUG initiation sequences in mammalian cells with an approach combining FACS and high-throughput DNA sequencing of a reporter library (17). They determined the high-efficiency initiation motif as RYMRMVAUGGC, where M = A or C and V = A, C, or G (17). Using their data, we determined the efficiency of each TIS contained in Tox (Fig. 3A) (17). Consistent with the potential for leaky scanning, the Kozak sequences surrounding M40 and M44 were predicted to be more efficient than the M1 TIS_CDS, whereas M42 was predicted to be the least efficient (Fig. 3A). When we mutated the Kozak sequence surrounding M1 to be more efficient, which would be predicted to reduce leaky scanning, expression of TOXΔN was reduced (Fig. 3A–3C). Changing the M1 TIS_CDS to a Kozak sequence of an equivalent strength did not significantly alter the ratio of the proteoforms (Fig. 3B, 3C).

Although the highly efficient M40 is the likely natural translational start for TOXΔN, a scanning mechanism would be predicted to use a downstream AUG in its absence. Thus, we tested the effect of various combinations of mutations of the downstream potential starts [M40G; M(42,44)G; M(40,44)I] (Fig. 3D). Consistent with leaky scanning, translation of TOXΔN still occurs when M40 is eliminated (Fig. 3E, 3F). Eliminating both of the high-efficiency initiation sites [M(40,44)I mutant] resulted in lower translation of TOXΔN than the other mutants, demonstrating that the efficiency of the TISs indeed regulates the amounts of proteoform production (Fig. 3E, 3F). Interestingly, the N-terminus of TOX_FL is highly conserved in some forms of the TOX family genes TOX2 and TOX3, and those forms contain potential downstream initiation sites at a position similar to those in TOX (Supplemental Fig. 1). The M1s from TOX2 and TOX3 have an efficiency score similar to that of TOX M1. However, the TOX2 downstream methionine (M43) has an equivalent efficiency to TOX2 M1. TOX3, in contrast, has two downstream methionines, with M38/M39 (mouse/human) having a higher efficiency score than M1, a more favorable configuration for leaky scanning (Supplemental Fig. 1). However, the contribution of these conserved sequences to the regulation of translation remains to be determined, especially considering the myriad of reported transcripts which indicate that Tox2 and Tox3 loci may use other protein diversification mechanisms.

To determine whether the efficiency of leaky scanning is entirely dependent on the mRNA sequence or whether it is also regulated by the cellular context, we used an in vitro wheat germ lysate-based transcription-translation system expressing TOX under the control of an SP6 promoter. Surprisingly, this yielded only TOX_FL (Fig. 4A). Because of specific eIF composition, leaky scanning may not be favored in this highly efficient system. But this also suggested that TOX proteoform expression might be a regulated process and highly dependent on cellular context.

To address this, we analyzed proteoform production during positive selection of CD4⁺ T cells in the thymus, a TOX-dependent process. Western blots of lysates from OTII TCR-transgenic thymocyte subpopulations showed that the ratio of TOX proteoforms is indeed regulated during positive selection. Expression of TOX is low in preselection CD4⁺8⁺ (double-positive) thymocytes and favors expression of TOX_FL (Fig. 4B–4D). TOX is upregulated as positive selection precedes, with peak expression in CD4⁺CD8^lo transitional cells (Fig. 4C), where lineage commitment occurs. This transition is accompanied by a switch to favoring TOXΔN over expression of TOX_FL (Fig. 4C, 4D). TOXΔN remains predominant as overall TOX expression declines in CD69⁺CD4⁺CD8⁻ (CD4SP) cells.

TOX also acts as an oncogenic factor in a number of hematological malignancies, including T cell acute lymphoblastic leukemia (T-ALL) (28) and cutaneous B and T cell lymphoma, particularly in the mycosis fungoides and Sézary syndrome subtypes (29–33). We found that corresponding tumor cell lines also produce varying ratios of TOX forms (Supplemental Fig. 2). Lysates from cutaneous T cell lymphoma–derived HH cells, Sézary syndrome–derived Hut78 cells, T-ALL–derived Molt4s, and Raji cells from a B cell lymphoma were analyzed by Western blotting for TOX. Hut78 cells expressed more TOXΔN, whereas the other cell lines produced more TOX_FL. Together, these data demonstrate that although the intrinsic efficiency of Kozak sequences controls the potential for leaky scanning, the particular cellular context regulates the quantitative pattern of TOX proteoform formation.

Chronic TCR-mediated stimulation induces sustained expression of TOX, which is required to induce many features of CD8 T cell exhaustion (5, 7, 28, 34). This can be modeled in culture, where repeated stimulation of OVA-specific TCR transgenic OTI CD8⁺ T cells leads to an exhausted phenotype (Fig. 5A) (8). Indeed, chronically but not acutely stimulated T cells in this system highly upregulate TOX and PD1, hallmarks of in vivo exhausted CD8 T cells (Fig. 5A). Interestingly, the ratio of proteoforms of TOX in chronically stimulated CD8⁺ T cells was similar to preselection double-positive thymocytes and lacked the predominance of TOXΔN as seen in thymocytes undergoing positive selection (Supplemental Fig. 3A).

We took advantage of this system to ask if the individual forms of TOX could substitute for chronic stimulation. Naive CD3⁺CD8⁺ splenocytes were isolated from OTI mice and stimulated with OVA peptide for 3 d to drive initial proliferation of the cells and allow retroviral transduction. Because this was insufficient to fully induce an exhausted phenotype, cells were also transduced to express TOX_FL or TOXΔN (and IRES-GFP) at 24 h after initiation of culture to sustain TOX expression (Fig. 5B). Interestingly, using GFP expression as a surrogate marker for TOX expression, both long and short forms of TOX quantitatively induced PD1 and Tigit at roughly equivalent levels (Supplemental Fig. 3B, 3C).

Because known cell surface markers of the exhausted phenotype are limited, we also compared isolated GFP⁻ (exogenous TOX⁻) and the top 20% of GFP⁺ (exogenous TOX⁺) populations by RNA-seq. DEG analysis was performed, comparing GFP⁺ cells from TOX_FL or TOXΔN cultures with GFP⁻ counterparts. Cells expressing exogenous TOX_FL contained a much higher number of DEGs than cells expressing TOXΔN (Fig. 5C). DEGs of note specifically regulated by TOX_FL but not TOXΔN include upregulated activation marker Kit and downregulation of effector- and memory-associated genes, including Bhlhe40, Klf2, and its target S1pr1 (35–37). Pathway analysis of specific TOX_FL DEGs (adjusted p value ≤0.05 and fold change ≥2) showed enrichment for genes involved in the cell cycle and DNA replication, most of which were upregulated (Fig. 5D, 5E). Other significant pathways identified were the FOXO pathway and lysine degradation, which were overall downregulated.

Of the DEGs, two genes altered by both TOX_FL and TOXΔN, Vipr1 and Tubb6, were previously reported as direct Tox gene targets in CD8⁺ T cells by chromatin immunoprecipitation sequencing (Fig. 5F) (34). Additionally, the gene target Lrrk1 was identified in CD8⁺ T cells by chromatin immunoprecipitation sequencing but was only differentially expressed by mRNA in our samples expressing TOX_FL (Fig. 5C) (34). Although no direct known gene targets were identified as unique to the TOXΔN sample, Batf3 was predicted to be a binding site by expression of TOX fused to bacterial DNA adenine methyltransferase combined with deep sequencing (38). Thus, we conclude that both forms of TOX are transcriptionally active, likely contribute to T cell exhaustion, and regulate overlapping but distinct gene targets in isolation. However, TOX_FL has greater influence over the transcriptional landscape during chronic stimulation in culture.

Here we describe the mechanism of diversification of the TOX protein via an alternative translation mechanism known as “leaky ribosomal scanning.” This is due to a less than optimal Kozak sequence surrounding M1, leading to production of two distinct TOX proteins from a single mRNA. The Tox locus is not unique in this regard. Deep sequencing of ribosome-protected mRNA fragments allows quantitative measuring of protein synthesis and ability to detect translation initiation sites at the codon level (39, 40). Studies using permutations of this technique have revealed widespread use of alternative TISs, including those that encode unannotated products from upstream open reading frames, N-terminal truncations and extensions, out-of-frame products, and use of near-cognate start codons (27, 39). In fact, ribosome profiling and N-terminal mass spectrometry estimate that ∼20% of human and mouse genes produce N-terminally truncated proteoforms via translation from internal start sites (41). It is interesting that not only M40, the likely downstream translational start based on Kozak efficiency and a 5′-to-3′ scanning mechanism, but also nearby downstream M42 and M44 are evolutionarily conserved. Whether these help to ensure efficient TOXΔN production and/or are important structural elements of the TOX protein remains to be determined. Likewise, whether the first 40 aa of TOX_FL serve as a structural or functional domain requires more investigation.

Thymic positive selection was associated with not only an increase in TOX expression but also a switch to favoring TOXΔN production. Because the mechanisms involved are not likely to be TOX specific, this suggests that global changes in the control of translation site selection have the potential to be an important regulatory mechanism during T cell development, allowing greater leaky scanning. Leaky scanning is in part regulated by eIFs, with the fidelity of start codon recognition governed by eIF1 and eIF5 antagonism (42, 43). Competition between eIF1 and eIF5 stabilize the stringency of start codon recognition, and eIF1 and eIF5 are also engaged in a regulatory feedback loop that not only controls global TIS stringency but also their own translation. Whether the complex interplay between eIF1 to eIF5 also regulates T cell development and T cell function remains to be determined.

Interestingly, TOXΔN does not predominate in chronically stimulated CD8⁺ T cells as it does in differentiating thymocytes. In some instances, alternative translation mechanisms evolved to ensure the production of essential proteins during times of stress, when canonical translation is globally downregulated. The metabolic stress CD8⁺ tumor-infiltrating lymphocytes experience alters translation by changing the balance of key nutrients in the cell (44–46). The metabolic changes induced by the tumor microenvironment or during chronic stimulation could also affect levels of leaky scanning and thus TOX proteoform expression, with as yet unknown consequences on gene expression.

We also found that PD-1 and other markers of an exhausted phenotype are upregulated in the context of either form of TOX in previously activated cells, but whether this is dependent on initial upregulation of endogenous TOX in this system remains to be determined. The Pdcd1 transcript had an average fold change of 1.6 in DEG analysis but failed to reach significance using an adjusted p value. It is possible that there are some nontranscriptional effects of TOX, or, alternatively, Pdcd1 upregulation, which appears to be dependent on very high levels of TOX, may be less prominent in the heterogeneous GFP⁺ population used for bulk RNA-seq (Supplemental Fig. 3B, 3C). Nevertheless, TOX_FL on its own had a greater effect on other gene regulation during the enforcement of a model of T cell exhaustion than did TOXΔN. This is consistent with the maintenance of greater expression of TOX_FL in chronically stimulated T cells than some thymocyte subpopulations. The data also indicate that the two forms of TOX may have distinct if somewhat overlapping roles in gene regulation. Because both forms of TOX contain the identical DNA-binding domain, C-terminal domain, and the majority of the N-terminal domain, it remains to be seen how the N-terminal peptide might influence TOX function. The N-terminal sequence is also highly conserved in some proteoforms of TOX family members TOX2 and TOX3, consistent with functional or structural importance. It is also possible that the two forms of TOX act synergistically to control cell fate, because we have not observed expression of one form without the other in cells.

Cells overexpressing TOX_FL uniquely upregulated cell cycle genes. TOX also controls the hierarchical development from a proliferative progenitor-like state to one of terminal exhaustion in conjunction with TCF1 and T-bet (47). The induction of cell cycle genes induced by TOX_FL may mimic this developmental process. Because both resident and proliferating progenitor exhausted cells express TOX, it would be interesting to know whether the conversion to proliferating cells is driven by TOX_FL. Additionally, TOX expression has been shown to drive oncogenesis in T-ALL by altering the cell cycle and inducing proliferation of the tumor cells (28). The downregulation of FoxO3, Klf2, S1pr1, and other FoxO signaling genes by TOX_FL is also intriguing. Although the FoxO family plays significant roles in T cell homeostasis and differentiation, FoxO3 has been shown to limit proliferation and promote the apoptosis of T cells during infection (48). Both of these findings are consistent with the requirement for TOX in the survival of exhausted CD8⁺ T cells (6, 49).

Exhausted T cells are a major barrier to immune system control of tumors and chronic infections, and prevention of their development and reinvigoration of existing exhausted cells is a major goal of immunotherapies. Because knockdown of TOX improves T cell function in tumors and infection, it could be a viable pharmaceutical target (5–7, 50). This work helps to ensure that future studies are seeing the whole of TOX.

The authors have no financial conflicts of interest.

We thank the Cedars-Sinai Medical Center Flow Cytometry and Applied Genomics core facilities for their expert technical assistance.