Translational repression of 5’ TOP mRNAs in chronically activated CD8+ T cells in vitro

In vitro chronically activated CD8+ Tcells show hallmarks of Tcell exhaustion

We first sought to establish a method for generating exhausted CD8+ T cells in vitro. We surveyed the literature for simple protocols that reproducibly produce large quantities of T cells that exhibit the molecular and phenotypic features of T cell exhaustion. An approach used by numerous groups is to chronically stimulate T cells with antibodies targeting the T cell receptor (TCR), bypassing the need for antigen stimulation [15–19]. Following this strategy, we purified CD8+ T cells from human peripheral blood mononuclear cells (PBMCs) of four healthy donors and activated them using anti-CD3 and anti-CD28 antibodies over the course of 2 weeks (Figure 1(a)). Our protocol employs a series of four 2-day activations separated by 2 days of rest, during which we remove the anti-CD3 and anti-CD28 antibodies to maintain cell viability and promote expansion (Figure 1(a), chronic). For comparison, we also activated CD8+ T cells for 2 days and either let them rest for the remaining 12 days or re-activated them on the last 2 days (Figure 1(a), acute and re-activated). These three conditions are herein referred to as T_chron, T_re-act, and T_acute cells.

Next, we sought to determine if our T_chron cells exhibit signs of exhaustion, such as impaired cytokine production and expression of inhibitory receptors. We observed an almost complete loss of IFNγ and TNFα cytokine secretion by the fourth activation in T_chron as compared to T_re-act (Figure 1(b)). Notably, while the viability of T_chron cells modestly declined with each activation, the magnitude of cytokine loss was significantly greater than the loss of viability (Supplementary Figure S1a,b). We also observed high expression of the inhibitory receptors LAG3, TIM3, PD1, and CD39 in T_chron cells (Figure 1(c)). Of note, T_re-act cells also expressed high levels of these markers, which are known to be expressed transiently during T cell activation (Figure 1(c)). Interestingly, there was heterogeneity in expression dynamics amongst the inhibitory receptors in T_chron cells, with some oscillating between high and low expression levels with each activation and rest interval (PD1), while the others gradually increased in expression throughout the course of the 2 weeks (LAG3, TIM3, and CD39) (Supplementary Figure S1c).

Exhausted CD8+ T cells possess a transcriptional programme that is distinct from effector, naive, and memory T cells. We therefore performed RNA sequencing (RNA-Seq) on T_chron, T_re-act, and T_acute cells in four healthy donors to determine if T_chron cells exhibited canonical features of exhaustion. Principal component analysis (PCA) revealed that T_acute cells are transcriptionally distinct from T_re-act and T_chron cells (Figure 1(d)). T_re-act cells also form a well-defined cluster. Interestingly, in two donors, T_chron cells segregated from both T_acute and T_re-act cells, while in the other two donors they cluster near T_re-act. We wondered if this meant the in vitro exhaustion protocol did not induce chronic activation to the same degree in the latter two donors. CD8+ exhausted T cells are heterogeneous, gradually differentiating from a more stem-like progenitor exhausted state to a terminally exhausted state. Thus, the two donors that cluster near the T_re-act cells may represent a different stage along this exhaustion trajectory compared to the other two donors. To explore this hypothesis, we examined gene expression of exhaustion-associated genes at the level of individual donors to illuminate any differences between them. We compared the two donors clustering near T_re-act cells to the two donors segregating from both T_re-act and T_acute cells (Supplementary Figure S2a, dashed and solid lines, respectively). We refer to these donors as T_chron1 and T_chron2, respectively (Supplementary Figure S2a). The transcription factor TCF1, encoded by the TCF7 gene, is the key regulator for maintaining the progenitor-like exhausted T cell population (Z. Chen et al., [20]; Im et al., [21]; Utzschneider et al., [22]; Wu et al., [23]). Interestingly, TCF7 expression was elevated in T_chron1 donors relative to T_chron2 donors under chronic stimulation (Supplementary Figure S2b). In contrast, the mRNAs for transcription factors and transcription regulators TOX2, NR4A2, and ID2 (TOX2, NR4A2, ID2), which have been associated with exhausted and terminally differentiated states, are more highly expressed in T_chron2 donors (Supplementary Figure S2b) (J. Chen et al., [24]; Knell et al., [25]; Omilusik et al., [26] [27]). Similarly, inhibitory receptors (LAG3, HAVCR2, TIGIT) were more highly expressed in T_chron2 donors in the chronic condition (Supplementary Figure S2b). While the cytokine TNF was downregulated in the chronic condition in all donors, the degree of downregulation was significantly greater in T_chron2 donors (Supplementary Figure S2b). Taken together, these data suggest a more progenitor-like state for the T_chron1 donors compared to the T_chron2 donors. Importantly, however, not all exhaustion markers were enriched in T_chron2 donors. Some were equally expressed in all T_chron donors (PDCD1, CCR7, SOX4), while SELL and ID3, generally associated with less differentiated T cell states, were more highly expressed in T_chron2 donors (Supplementary Figure S2b) (Gago Da Graça et al., [28]; Tsui et al., [29]). Taken together, these data suggest that chronic activation induced heterogeneous exhaustion-associated transcriptional states in T_chron1 and T_chron2 donors.

We also compared all T_chron to T_re-act donors to identify differentially expressed genes (Figure 1(e)). In line with our observation that cytokine production is severely hindered in T_chron cells, both the IFNG and TNF transcripts were downregulated. We also observed upregulation of mRNAs encoding proteins previously linked to T cell exhaustion, such as the transcription factors TOX2 and SOX4, membrane glycoprotein LAYN, and ectoenzyme CD38 [27,30–33]. Interestingly, the mRNA for transcription factor Eomes (EOMES) was downregulated in T_chron cells in our data. Although T cell exhaustion is generally associated with elevated expression of Eomes, its expression varies across exhaustion states [34]. Progenitor and intermediate exhausted T cells express lower levels of Eomes, whereas terminally exhausted cells exhibit high Eomes expression [35]. Overall, these data suggest that T_chron cells may not adopt a uniformly exhausted state, but nevertheless exhibit many hallmarks of exhaustion.

Finally, we used gene set enrichment analysis (GSEA) to probe whether certain pathways and biological processes were enriched in T_chron or T_re-act cells. Using the Molecular Signatures Database hallmark human gene set collection, we obtained a broad overview of the transcriptional programmes associated with chronic stimulation (Liberzon et al., [36]) (Supplementary Table S1). We were particularly intrigued by the strong negative enrichment of MYC transcription factor target genes in T_chron cells (Figure 1(f)). MYC is a major driver of metabolic reprogramming during T cell activation and is required for cell growth and proliferation (R. Wang et al., [37]). While the role of MYC in T cell exhaustion is not well understood, MYC mRNA has been reported to be reduced in exhausted T cells (Wherry et al., [38]). Experimental downregulation of MYC is also sufficient to induce exhaustion-associated phenotypes (Liu et al., [39]). We also observed negative enrichment of genes involved in mTORC1 signalling and glycolysis in T_chron cells (Figure 1(f)). MYC and mTOR pathways have extensive crosstalk, as they cooperate to promote anabolic and glycolytic metabolic programmes that support effector T cell function (Chi, [40]; R. Wang et al., [37]). Thus, the coordinated depletion of MYC targets, mTORC1 signalling genes, and glycolytic genes is consistent with previous reports that metabolic dysfunction contributes to T cell exhaustion [41,42], (Zheng et al., 2022).

T_chron cells translationally downregulate TOP mRNAs

To investigate translation dynamics globally during T cell exhaustion, we also performed ribosome profiling in parallel with RNA-Seq on lysates from T_chron, T_re-act, and T_acute cells (Figure 2(a)). Ribosome profiling provides a snapshot of genome-wide ribosome occupancy on mRNAs by capturing and deep sequencing the ~30 nucleotide ribosome-protected fragments (RPFs) left after nuclease digestion [43]. The density of RPFs on a transcript is a proxy for the extent of protein synthesis [44]. Here, we used a simplified ribosome profiling workflow that uses P1 nuclease for the digestion step and ordered two-template relay (OTTR) for library preparation [45]. These modifications to the original method greatly reduce the labour and input required to perform ribosome profiling, making it ideal for experiments where cell material is limited [45].

Our ribosome profiling results showed the expected read length distribution for eukaryotic RPFs, averaging around ~35 nt (Supplementary Figure S3a) [45]. The percentage of reads that mapped to mRNA (as opposed to rRNA, gDNA, etc) ranged from ~10–30% (Supplementary Figure S3b). Principal component analysis of RPFs revealed a similar trend to the RNA-seq data – T_acute and T_re-act cells were distinct populations, while T_chron cells showed two donors that clustered with T_re-act cells while the other two donors formed a unique cluster (Supplementary Figure S3c and Figure 1(d)).

We then used our ribosome profiling and RNA-seq data to investigate gene regulation during T cell activation and T cell exhaustion programmes (Supplementary Tables 2–4). To this end, we performed an integrative analysis of both our RNA-seq and ribosome profiling datasets using the deltaTE method to compare T_re-act to T_acute cells and T_chron to T_re-act cells [46]. Genes were sorted into two regulation classes: forwarded genes, where changes in RPF counts mirrored changes in mRNA counts and therefore indicate transcriptional regulation, and exclusive genes, where changes in RPF counts occur despite no change in mRNA counts and therefore indicate translational regulation [46] (Figure 2(a)). Of note, there are two more regulation classes – buffered and intensified – which we did not observe in our analysis. When comparing T_re-act to T_acute cells, nearly all the differentially expressed genes (n = 773) fell under the forwarded category, suggesting that gene regulation occurs largely at the level of transcription during CD8+ T cell activation (Figure 2(b) and Supplementary Table 4). Consistent with this, gene ontology (GO) analysis revealed that the biological processes enriched among the forwarded genes were those related to immune system regulation and T cell activation (Figure 2(b)). By contrast, comparison of T_chron and T_re-act cells revealed only 26 forwarded genes, which were again enriched in processes related to immune function (Figure 2(c) and Supplementary Table S4). Interestingly, there were 49 genes categorized as exclusive that showed significant translational repression in T_chron relative to T_re-act cells (Figure 2(c) and Supplementary Table 4). GO analysis revealed that this group of translationally repressed genes is highly enriched for processes related to translation and ribosome biogenesis, including ribosomal protein genes (RPS24, RPL32), initiation factor genes (EIF3E), and translation elongation factor genes (EEF1A1, EEF1B2) (Figure 2(c) and Supplementary Table 4).

Next, we were interested in further investigating the group of exclusive genes that were translationally downregulated in T_chron cells. Importantly, the majority of the genes in this category are mRNAs that contain TOP motifs. We therefore hypothesized that all TOP mRNAs may be subject to dynamic translational control in T_chron cells. To test this, we analysed the translation efficiencies (TE) of all genes using deltaTE [46]. TE is defined as the ratio of the number of RPFs mapping to the open reading frame (ORF) over the number of mRNA counts [44,46]. We then compared TEs of T_chron to T_re-act cells and T_re-act to T_acute cells (Supplementary Figure S3d, Supplementary Tables S2–3). Strikingly, TOP mRNAs are significantly repressed relative to all other mRNAs when comparing T_chron to T_re-act cells (Figure 3(a)) [47]. In contrast, TOP mRNAs show no translation regulation during T cell activation (T_re-act vs T_acute) (Figure 3(a)). Notably, the translational repression in T_chron occurs despite a mild increase in transcript abundance, indicating that downregulation is driven entirely at the level of translation (Figure 3(b)).

A recent study by Liu and colleagues used an in vitro chronic antigen stimulation model combined with a modified ribosome profiling approach to investigate translational dynamics during T cell exhaustion [15]. We were therefore interested in assessing the similarities and differences observed in their results compared to those presented here. They observed hypertranslation of OXPHOS mRNAs driven by the RNA-binding protein LARP4 in their chronically stimulated CD8+ murine T cells (Figure 3(c), left). However, they did not observe the translational repression of TOP mRNAs we observed (Figure 3(c), right). In contrast to Liu et al., we did not observe the increase in translation efficiency of OXPHOS mRNAs in our T_chron cell data. In fact, we observed a mild repression in the translation of these mRNAs (Figure 3(d)). We explore possible reasons for this seeming discrepancy in the Discussion.

mTOR activity is elevated in T_chron cells

TOP mRNA translation is highly sensitive to the activity of mTOR. Given our observation that TOP mRNAs are translationally downregulated in T_chron cells, we sought to investigate the role of mTOR signalling during this process. To do so, we first assessed the state of mTOR activity using phosphorylation of ribosomal protein S6 (pRPS6) as a readout. We hypothesized that reduced mTOR activity in T_chron relative to T_re-act cells underlies the translational repression of TOP mRNAs that we observed in our ribosome profiling and RNA-seq data. This would also match the negative enrichment of mTORC1 signalling pathway genes that we observed in T_chron cells from GSEA of our RNA-seq data (Figure 1(f)). Unexpectedly, we observed the opposite effect – pRPS6 levels were consistently higher in T_chron than in T_re-act cells throughout activations 2, 3, and 4 (Figure 4(a,b)). Notably, T_chron cells have greater levels of pRPS6 even after the last activation (activation 4), during which T_re-act cells were also subjected to the same activation (Figure 4(a,b)).

Next, we used an inhibitor of mTOR, rapamycin, to test two alternative hypotheses based on our seemingly contradictory observations. First, given our observation that TOP mRNAs are repressed in T_chron cells, we hypothesized that inhibiting mTOR activity via rapamycin treatment would push these T cells towards an exhausted state by further repressing translation of TOP mRNAs. Alternatively, given our observation that mTOR activity is constitutively high in T_chron cells, we hypothesized that rapamycin treatment would put a brake on mTOR activity and slow down the progression of T cell exhaustion. We limited rapamycin treatment to the first three activations (Figure 4(a)), consistent with studies showing that rapamycin treatment improves outcomes when administered to mice during the initial phase of chronic infection, but has the opposite effect when administered after the establishment of exhaustion [48]. Notably, rapamycin treatment reduced pRPS6 levels compared to the DMSO treated control in T_chron cells (Figure 4(c)), consistent with mTOR inhibition. However, after the last activation, at which point rapamycin had been removed from the media, pRPS6 levels in T_chron cells returned to the same level as the DMSO treated control T_chron cells (Figure 4(c)). We also observed a smaller cell size during rapamycin treatment of T_chron cells as measured by forward scatter (FSC), consistent with the role of mTOR in regulating cell growth (Supplementary Figure S4a, timepoint 2). Once again, this effect was transient and disappeared upon rapamycin removal (Supplementary Figure S4a, timepoint 4). Notably, rapamycin treatment improved cell viability throughout the 2-week activation period as compared to T_chron cells treated with DMSO, as observed by periodic cell counting. We also used inhibitory receptor expression to assess the effect of rapamycin treatment on T_chron cells. Intriguingly, we observed a reduction in LAG3 and TIM3 expression, but not PD1 expression, in the rapamycin treated sample as compared to the DMSO treated control after the final activation (Figure 4(d) and Supplementary Figure S4b). This result better matches our second hypothesis – that rapamycin treatment effectively restrains mTOR activity to slow the progression of T cell exhaustion.

Cell culture

Human primary T cells were maintained in ImmunoCult™-XF T cell expansion medium (STEMCELL Technologies). Recombinant human IL-2 (Peprotech) was added to the media at the indicated concentrations. Cells were maintained between 2 × 10⁵ − 1 × 10⁶ cells/mL.

Isolation of human primary T cells

Peripheral blood mononuclear cells (PBMCs) were isolated from human peripheral blood leukopaks (STEMCELL Technologies). Briefly, the contents of the leukopak were combined with an equal volume of MACS buffer (PBS, 2% FBS, 1 mM EDTA) and spun down (15 min, 100 × g, Dec = 1), following which the pellet was washed 1–2 times (5 min, 400 × g) in the MACS buffer. Bulk (CD3⁺) T cells or CD8⁺ T cells were isolated from the PBMCs by magnetic negative selection using the EasySep™ Human T cell isolation kit or Human CD8⁺ T cell isolation kit (STEMCELL Technologies), respectively, following the manufacturer’s protocol. For future use, cells were frozen in CryoStor® freezing medium (BioLife Solutions) and treated in the same manner as freshly isolated cells once thawed.

In vitro chronic activation of CD8+ T cells

To model T cell exhaustion in vitro, CD8⁺ T cells were cultured under repeated stimulation conditions, as shown in Figure 1(a). Cells were stimulated using plate-bound anti-CD3 (clone UCHT1) (Tonbo Biosciences (now Cytek)) at 5 µg/mL and soluble anti-CD28 (clone CD28.2) (Tonbo Biosciences (now Cytek)) at 2.5 μg/mL antibodies for 48 h in ImmunoCult™ T cell expansion medium (STEMCELL Technologies) supplemented with recombinant human IL-2 (200 U/mL). After stimulation, cells were centrifuged and resuspended in fresh medium containing IL-2 (200 U/mL) and expanded for 48 h. This stimulation and rest cycle was repeated four times over the course of 2 weeks. Cell lysates were collected 48 h after the final stimulation. This condition is referred to as the ‘chronic’ stimulation group and models T cell exhaustion.

In parallel, two additional conditions were maintained. The ‘acute’ condition received only a single 48-h stimulation at the beginning of the culture period, followed by expansion in IL-2-containing medium without further stimulation. The ‘re-activated’ condition received an initial stimulation followed by expansion, and then a final stimulation at the end of the 2-week period. All cultures were maintained under identical IL-2 concentrations (200 U/mL) and medium conditions throughout.

Rapamycin treatment

Rapamycin was added to the media during both the stimulation and rest stages at 10 nM during the first three activations in T_chron. As a control, we included a condition in which an equal volume of DMSO as in the rapamycin treatment was added to the media during the first three activations.

Enzyme-linked immunosorbent assay (ELISA)

We used ELISAs to quantify cytokine production (Interferon gamma (IFN-γ) and Tumor Necrosis Factor alpha (TNF-α)) by T_chron and T_re-act cells following each stimulation. At each stimulation, cells were counted and plated in triplicate at 2 × 10⁵ cells/well in a 96-well flat bottom plate coated with anti-CD3 antibody (5 µg/mL). Anti-CD28 antibody (2.5 µg/mL) and IL2 (200 U/mL) were added directly to the media containing the cells. After 48 h, the cell supernatants from these wells were collected and stored in the −80°C freezer until use. Cell supernatants were diluted appropriately and the ELISA MAX™ Human IFN-γ and TNF-α Sets (BioLegend) were used to determine IFN-γ and TNF-α concentrations following the manufacturer’s protocol. Absorption at 450 nm was measured using a TECAN Spark plate reader. The IFN-γ and TNF-α standards provided in the kits were used to generate a standard curve. GraphPad Prism was used to fit the standard curve.

Flow cytometry

Flow cytometric analysis was performed on an Attune NxT Acoustic Focusing Cytometer (ThermoFisher). Briefly, cells were pelleted in a 96-well round bottom plate and washed with PBS before staining. For intracellular staining, cells were fixed and permeabilized using the Cytofix/Cytoperm Fixation/Permeabilization kit (BD Biosciences) according to the manufacturer’s instructions. Cells were stained with the following antibodies and live/dead stain diluted in PBS: LIVE/DEAD fixable near-IR (Invitrogen), CD8 (BioLegend, clone SK1), PD-1 (BioLegend, clone EH12.2H7), TIM3 (BioLegend, clone A18087E), LAG3 (BioLegend, clone 11C3C65), CD39 (BioLegend, clone A1), and RPS6 phospho Ser235/236 (BioLegend, clone A17020B). Cells were stained for surface markers on ice for 20 min in the dark. Cells were washed twice in PBS and resuspended in 150 µl of PBS for analysis. Flow cytometry data was analysed using FlowJo software.

RNA-seq

Total RNA was extracted from the same lysates collected for ribosome profiling using the Direct-zol RNA miniprep kit (Zymo Research). Library preparation and sequencing were carried out by Novogene using an Illumina NovaSeq X Plus 150PE platform. The FASTQ files provided by Novogene were already trimmed and filtered. Reads were aligned to a custom transcriptome that includes both the NCBI RefSeq MANE v0.95 transcript set and the 13 mitochondrial mRNA sequences from Ensembl [45]. We used Salmon to quantify transcript-level abundances [65]. PyDESeq2 was used for differential expression analysis [66]. Gene set enrichment analysis (GSEA) was performed using GSEA software, with the hallmark human gene sets from the Molecular Signatures Database (MSigDB) (Hallmark Myc Targets V2, Hallmark mTORC1 signalling, Hallmark Glycolysis) (Liberzon et al., 2015; Mootha et al., [67]; Subramanian et al., [68]).

Ribosome profiling

Samples were prepared according to the protocol described in Ferguson et al. and on the Karnateq website [45]. Briefly, on the day of collection, T cells were pelleted by centrifugation and washed once in PBS. The pellet was then resuspended thoroughly in 1 mL of lysis buffer (20 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM MgCl2, 1 mM DTT, 100 μg/mL cycloheximide, 1% v/v Triton X-100, 25 U/mL Turbo DNase I (Invitrogen)) and digested on ice for 30 min. The cell lysate was triturated through a 26 G needle 10 times and clarified by centrifugation (10 min, 4°C, 20,000 × g). Lysates were aliquoted, snap-frozen in liquid nitrogen, and stored at −80°C until ready for use.

RNA concentration of the lysates was determined using the QuantIt RiboGreen RNA kit (Invitrogen). Lysates were subsequently digested by P1 nuclease (New England Biolabs) at 20 U per µg of RNA for 1 h at 37°C on a tube rotator. Ribosomes were pelleted by sucrose cushion in a TLA 110 rotor at 4°C for 1 h at 100,000 rpm. Trizol was added to the ribosome pellets and RNA was extracted using a modified method called mirRICH to preferentially enrich for small RNA fragments [45,69]. Library generation was carried out using the OTTR cDNA library construction kit (Karnateq and a generous gift from the Collins lab) following the manufacturer’s protocol. 40 ng of RNA per sample was used as input for the OTTR reaction. In parallel, 30 and 40 nucleotide control RNA oligos (a gift from the Collins lab) were used as input for the OTTR reaction to use as size selection markers [45]. cDNA cleanup was performed following RNase A and RNase H digestion using an Oligo Clean & Concentrator kit (Zymo Research). cDNAs were size selected on a 0.6x TBE 8% Urea-PAGE and visualized using the fluorescence of the IR800 dye included in the 5’ adapter. The region between the 30 and 40 nucleotide controls was excised and the cDNA was eluted. Following the protocol described in McGlincy et al., cDNA concentrations were measured by qPCR to determine the number of PCR cycles necessary to amplify each library using Illumina primers [70]. Libraries were quantified using the NEBNext Library Quant Kit, pooled at equimolar concentration, and sequenced by Novogene on a NovaSeq X Plus 150PE platform.

Ribosome profiling computational analysis was carried out as described by Ferguson et al., with minor modifications [45]. First, the adapters were trimmed, unique molecular identifiers (UMIs) were annotated, and reads shorter than 20 nucleotides or of poor quality were discarded using Cutadapt: -m 2 -a AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC -A AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT -u 7 -U −7 –rename = ’{id}_NTA = {r1.cut_prefix}’ -m 20 -q 10.

Of note, although the libraries were sequenced with paired-end sequencing, we only used Read 1 for our analysis after this point. Reads were sequentially mapped to rRNA, ncRNA, mtDNA, mRNA, gDNA, size-selection oligos, and OTTR adapter references using Bowtie2. For mRNA, reads were aligned to a custom transcriptome that includes both the NCBI RefSeq MANE v0.95 transcript set and the 13 mitochondrial mRNA sequences from Ensembl [45]. We proceeded with further analysis using only the reads that mapped to mRNA. PCR duplicates were collapsed using UMI-tools [71]. Reads mapping outside of the coding sequence (+15^th to −10^th codon) were removed using Bedtools intersect [72]. The remaining reads were counted using Salmon [65]. Differential translation efficiency analysis was carried out using DeltaTE [46].