To identify candidate non-coding ceRNAs in cytotoxic T cells, we performed Argonaute-2 high-throughput sequencing of crosslinking immunoprecipitation (Ago2 HITS-CLIP, AHC). Integrating sequence reads across different classes of transcribed genomic annotations revealed that lncRNAs are bound extensively by Ago2, but the median lncRNA had 7.8 times fewer aligned AHC sequence reads compared to 3′ UTRs where miRNAs canonically bind to mRNAs (Figure 1A). This difference likely reflects the comparatively low expression of many lncRNAs, and it highlights the relatively low occupancy of Ago2 on most of these transcripts. Nevertheless, there were individual lncRNAs that stood out as highly bound across the transcript. To prioritize lncRNAs for further investigation, we manually curated the transcripts with the largest number of aligned AHC sequence reads. Among the top 10 most highly bound transcripts, 3 overlapped with protein-coding genes, 2 were on the mitochondrial chromosome, and 3 were repetitive annotations. Strikingly, all of the top 10 transcripts were either annotated as rRNA or contained an rRNA repeat element, except the eighth most bound transcript, Malat1 (Table 1). When the same analysis was repeated with rRNA repeats masked, Malat1 was the second most highly bound transcript (Table 2).

Malat1 has been proposed to inhibit miRNAs as a ceRNA (Chen et al., 2017; Luan et al., 2016; Qiao et al., 2018; Wang et al., 2019; Wu et al., 2018; Xiao et al., 2015; Xie et al., 2017). We used the Piranha peak calling algorithm to identify sites with the highest degree of miRNA binding, as indicated by AHC sequence read number and density (Figure 1B). Within cytotoxic T cells, the algorithm identified 55 AHC peaks in Malat1, the largest of which was extremely pronounced (Figure 1C). Compared with other lncRNA-binding peaks, this peak had the 36th most total aligned reads and the 142nd highest read density (15th and 16th, respectively, when rRNA reads are masked). Even when compared to peaks in 3′ UTRs of mRNAs, the largest peak in Malat1 was the 100th most (98th percentile) bound peak in terms of read density (Figure 1B).

miRTarget, the custom miRNA-binding prediction algorithm (Liu and Wang, 2019), identified an 8-mer seed-binding sequence of the miR-15/16 family centered within the most densely bound region of the called peak (Figure 1C). miR-15/16 has not been previously shown to interact with Malat1 in mice, so to determine whether this peak is miR-15/16 dependent, we performed AHC with cultured CD8⁺ miR-15/16^Δ/Δ T cells. In the absence of miR-15/16, Ago2 binding to Malat1 was preserved throughout the whole transcript except for the peak containing the predicted miR-15/16 8-mer seed-binding sequence (Figure 1C). We conclude that miRNAs of the miR-15/16 family bind abundantly to Malat1. Furthermore, miR-15/16 binding occurred in a region of high evolutionary conservation (Figure 1D). The corresponding region containing the miR-15/16-binding site in human MALAT1 was highly enriched in two publicly available AHC datasets obtained using the 293 human embryonic kidney (HEK) cell line, consistent with Malat1 regulating miR-15/16 in a colorectal cancer cell line (Karginov and Hannon, 2013; Ji et al., 2019; Li et al., 2018; Figure 1E). We conclude that miR-15/16 bind abundantly to Malat1 in mouse and human cells.

We hypothesized that Malat1 may inhibit the function of miR-15/16 in cytotoxic T cells. To directly address this question in mice, we used CRISPR-Cas9 with homology directed repair to generate mice in which five nucleotides of the miR-15/16 seed-binding sequence within Malat1 were scrambled (Figure 1F). Mice homozygous for this mutation are subsequently referred to as Malat1^scr/scr. To confirm the targeted functional outcome of this mutation, AHC was performed on cultured CD8⁺ T cells isolated from Malat1^scr/scr mice. Ago2 binding was preserved across the Malat1 transcript except at the mutated miR-15/16-binding site, where AHC sequence read density was greatly reduced (Figure 1C). This finding confirmed that miR-15/16 binds to Malat1 at this site in a sequence-dependent manner, and provided us with a tool for highly specific investigation of the functional consequences of Malat1:miR-15/16 interaction.

Malat1:miR-15/16 interaction could lead to regulation and/or degradation of the lncRNA, the miRNAs, or both. To assess whether miR-15/16 degrades Malat1 we compared the expression of Malat1 by mRNA sequencing in primary mouse CD8⁺ T cells. Malat1 expression was unchanged in miR-15/16^Δ/Δ cells as well as in the Malat1^scr/scr cells (Figure 2A). Previous studies investigating Malat1 and other lncRNAs as ceRNA inhibitors of miRNAs have suggested that inhibition occurs by either target RNA-directed miRNA degradation (TDMD) or stoichiometric sequestration of the miRNA from protein-coding mRNA targets (Han et al., 2020; Su et al., 2021). Therefore, we also tested the possibility that Malat1 lowers miR-15/16 abundance by TDMD or a related mechanism. However, in freshly isolated mouse CD8⁺ T cells, miR-15b and miR-16 were unchanged and only miR-15a, a family member with lower expression, was modestly decreased in Malat1^scr/scr cells (Figure 2B). We conclude that Malat1 and the miR-15/16 family do not influence each other’s absolute abundance in this setting. To test whether Malat1 affects miR-15/16 function, we examined miR-15/16 target binding in our AHC data. We first defined an experimentally supported list of TargetScan predicted miR-15/16-binding sites with at least one AHC read in both WT and Malat1^scr/scr cells. Using this list, we then examined the read depth at these sites in both Malat1^scr/scr and WT cells (603 sites contained in 479 genes). In WT cells, these sites constituted on average 3.2% of the binding in each 3′ UTR, and this figure increased to 3.4% in Malat1^scr/scr cells, indicating that Ago2 occupancy preferentially increased at these sites when miR-15/16 binding to Malat1 was eliminated (Figure 2C). To confirm that binding to these target sites was miR-15/16 dependent, we assessed their AHC read depths in miR-15/16^Δ/Δ cells as well. As predicted, binding was greatly reduced in miR-15/16^Δ/Δ cells, representing, on average, 1.0% of binding to a given 3′ UTR, a 69% reduction in Ago2 binding compared to WT (Figure 2D). In contrast, no significant differences were observed for binding at predicted sites for the highly expressed miRNA families of miR-101, Let-7, miR-21, and miR-142, (Figure 2—figure supplement 1). Thus, we conclude that the Malat1^scr allele specifically negatively regulates the first requirement of miR-15/16 function – binding to mRNA targets.

We next sought to investigate whether this increased binding resulted in decreased target mRNA expression. To do so, we analyzed mRNA sequencing data from primary CD8⁺ T cells 24 hr after stimulation. We generated empirical CDF plots from these data comparing target gene expression in miR-15/16^fl/fl to miR-15/16^Δ/Δ cells as well as WT to Malat1^scr/scr cells. For each of these comparisons, we then compared the distribution for miR-15/16 target genes, as defined above, to the distribution for all other expressed genes. In the case of WT to Malat1^scr/scr cells, the distribution was shifted in favor of decreased target gene expression in the Malat1^scr/scr cells (Figure 2E). In the case of miR-15/16^fl/fl to miR-15/16^Δ/Δ, the distribution was shifted in favor of increased target gene expression in the miR-15/16^Δ/Δ cells, indicating that this gene set is relieved of miRNA-induced repression (Figure 2F). These data indicate that the increased availability of miR-15/16 leads to increased repression of mRNA target genes when Malat1:miR-15/16 interaction is ablated.

We further investigated genes that displayed reciprocal expression changes in miR-15/16^Δ/Δ and Malat1^scr/scr cells. Of 479 genes in the bound target list, 432 (90%) were either downregulated in Malat1^scr/scr cells or upregulated in miR-15/16^Δ/Δ cells, compared to controls. Among these genes, the expression of 298 (62%) were decreased in Malat1^scr/scr cells, and 294 (61%) were increased in miR-15/16^Δ/Δ cells, with 160 (33%) both increased in miR-15/16^Δ/Δ and decreased in Malat1^scr/scr cells (Figure 2G). Gene ontology enrichment analysis revealed multiple pathways associated with growth factor and antigen receptor signaling affected by the Malat1:miR-15/16 circuit in cytotoxic T cells (Figure 2H). Many of these modules, including the T cell activation module, were identified because they share key signaling proteins. For example, gene expression and AHC data support direct miR-15/16 targeting of the costimulatory receptor Cd28, the alpha subunit of phosphatidylinositol 3-kinase (Pik3r1) and c-Jun N-terminal kinase (JNK1), encoded by Mapk8 (Figure 2—figure supplement 2 and Figure 3). Targeting of these genes by miR-15/16 is consistent with their known tumor suppressor role (Cimmino et al., 2005; Gagnon and Ansel, 2019) and with Malat1’s association with cancer cell proliferation and metastasis (Ji et al., 2003), providing new mechanistic insight into those observations.

Given that the Malat1:miR-15/16 circuit regulated genes essential to T cell activation, and in particular CD28, we next investigated the functional consequences of perturbing this circuit in cytotoxic T cells. First, we looked directly at CD28. AHC in WT cells detected a prominent peak in the CD28 3′ UTR at the TargetScan predicted binding site for miR-15/16. This peak was absent in miR-15/16^Δ/Δ cells, whereas other binding peaks were preserved, empirically verifying that this binding event is miR-15/16 dependent. AHC in Malat1^scr/scr cells indicated a modest increase in Ago2 binding at this site compared to WT cells relative to Ago2 binding in the whole 3′ UTR (Figure 3A). These data, along with our previous demonstration that the CD28 3′ UTR is miR-15/16 responsive, indicate that CD28 is part of a module of miR-15/16 target genes that are highly likely to be affected by the Malat1^scr allele (Gagnon and Ansel, 2019).

Indeed, CD28 expression and T cell activation were regulated by the Malat1:miR-15/16 circuit. Flow cytometric measurement in primary splenic CD8⁺ T cells revealed that CD28 protein expression was decreased in Malat1^scr/scr mice and enhanced in miR-15/16^Δ/Δ mice (Figure 3B, C). To investigate activation-induced gene expression, we performed RNA-seq on splenic CD8⁺ T cells stimulated for 24 hr with plate-bound ɑCD3 with or without ɑCD28 crosslinking antibodies diagrammed in (Figure 3D). CD28 stimulation enhances distinct activation-induced gene expression changes in T cells (Martínez-Llordella et al., 2013), and these CD28-responsive genes were altered in miR-15/16^Δ/Δ and Malat1^scr/scr cells (Figure 3E–H). The previously defined set of 164 genes that are upregulated in WT cells stimulated with ɑCD3 + ɑCD28 compared to ɑCD3 alone was also significantly upregulated in the ɑCD3 + ɑCD28 condition compared to ɑCD3 alone for each genotype tested in our experiments (Figure 3—figure supplement 1). Importantly, the Malat1:miR-15/16 circuit affected this gene set in the ɑCD3 alone condition, with Malat1^scr/scr cells exhibiting decreased expression compared to WT cells, and miR-15/16^Δ/Δ cells exhibiting increased expression compared to miR-15/16^fl/fl cells (Figure 3E, G). This trend was preserved, but to a lesser degree, in the ɑCD3 and ɑCD28 conditions (Figure 3F, H). Thus, Malat1:miR-15/16 interaction enhanced expression of a costimulation-responsive gene expression program in activated T cells, and it further enhanced expression of that module even when CD28 costimulation was directly engaged.

Unsupervised hierarchical clustering of these samples based on the expression of the 164 costimulation-responsive gene set further underscored the costimulatory-like effect of the Malat1:miR-15/16 circuit. In comparing gene expression in Malat1^scr/scr and WT samples, three major groups emerged. The group with the lowest average costimulation-responsive gene expression contained only samples stimulated with ɑCD3 alone and primarily Malat1^scr/scr samples. The group with intermediate expression was the largest group with an even representation of Malat1^scr/scr samples and WT samples. While samples in this intermediate group were from both stimulation conditions, the Malat1^sc/scr samples tended to be from the ɑCD3 + ɑCD28 condition and the WT samples had an even representation from both stimulation conditions. The group with the highest average expression contained predominantly ɑCD3 + ɑCD28 stimulated samples with an even representation of Malat1^scr/scr and WT samples (Figure 3I). Thus, Malat1:miR-15/16 interaction and engagement of CD28 signaling additively induced costimulation-responsive genes. This observation is further supported when clustering the miR-15/16^Δ/Δ and miR-15/16^fl/fl samples. The two largest clusters were divided nearly exclusively by genotype. The group with lower average expression of costimulation-responsive genes contained predominantly mir-15/16^fl/fl samples stimulated with ɑCD3 alone. The group with the higher average expression contained a majority of miR-15/16^Δ/Δ samples, and the only miR-15/16^fl/fl samples within the group received ɑCD3 + ɑCD28 stimulation. Within the higher expression group, the miR-15/16^fl/fl samples stimulated with ɑCD3 + ɑCD28 subclustered with miR-15/16^Δ/Δ samples stimulated with ɑCD3 alone (Figure 3J). A third, much smaller cluster was composed of both samples from a single outlier biological replicate. Overall, these data show that miR-15/16 restrict costimulation-responsive gene expression, and that Malat1:miR-15/16 interaction limits this effect.

In addition to the proximal changes in gene expression, downstream functional outcomes of CD28 costimulation were also affected. Early activation genes have been well described in T cells, with CD69 responding to many cues including TCR and CD28 ligation, and Nur77 responding very specifically to TCR signals (Ashouri and Weiss, 2017; Vandenberghe et al., 1993). In similar fashion as above, we assessed the expression of these proteins 2 and 4 hr after stimulation with plate-bound crosslinking antibodies on primary CD8⁺ T cells. CD69 exhibited decreased expression in Malat1^scr/scr cells at both 2 and 4 hr post stimulation (2 hr: p = 0.0014 and 4 hr: p = 0.0598) and increased expression in miR-15/16^Δ/Δ cells at 4 hr post stimulation only (p < 0.0001) (Figure 4A, B), in accordance with costimulation-responsive gene module expression (Figure 3). There was no difference, however, in expression of Nur77 across Malat1^scr/scr, WT, and miR-15/16^Δ/Δ cells, indicating that TCR signals were equivalent (Figure 4A, C). Another key consequence of CD28 ligation is the production of IL-2 (Fraser et al., 1991; Maggirwar et al., 1997; Vandenberghe et al., 1993). Therefore, we assessed cytokine production in the supernatants of these cultures 16 hr after stimulation. In line with previous findings in LCMV-infected mice (Gagnon and Ansel, 2019), miR-15/16^Δ/Δ cells made significantly more IL-2 and TNFɑ after ɑCD3 stimulation both with and without ɑCD28, with a trend toward increased IFNγ as well (Figure 4D–F). Malat1^scr/scr cells stimulated with ɑCD3 + ɑCD28 exhibited a reciprocal trend specifically for IL-2 (23% decrease compared to WT, p = 0.26) (Figure 4D). Together, these data show that the Malat1:miR-15/16 circuit regulates key functional outcomes of CD28-mediated costimulation, from proximal gene expression changes to early activation protein expression and cytokine secretion.

miR-15/16 restrict memory T cell differentiation, cell cycle, and cell survival during the response to LCMV Armstrong infection (Gagnon and Ansel, 2019). CD28 costimulation is essential for IL-2 production, memory cell formation, and memory recall responses in vivo (Borowski et al., 2007; Eberlein et al., 2012; Fuse et al., 2008; Grujic et al., 2010; Suresh et al., 2001). IL-2 is also essential for these same processes and in particular is required in a CD8⁺ T cell-intrinsic nature for the formation of CD127⁺ KLRG1⁻ memory cells (Blattman et al., 2003; Kahan et al., 2022; Pipkin et al., 2010; Toumi et al., 2022; Whyte et al., 2022). Since Malat1 inhibits miR-15/16 activity and this circuit impacts proper T cell activation and IL-2 production after CD28 costimulation, we hypothesized that Malat1^scr/scr cells would exhibit poor memory formation and survival.

We first examined steady-state memory populations of polyclonal T cells in unchallenged, young mice. Malat1^scr/scr mice had normal naive and central memory T cell populations, but a reduced percentage and number of effector memory (CD44⁺CD62L^–) cells in the spleen (Figure 5—figure supplement 1). Therefore, we investigated the cell-intrinsic nature of the Malat1^scr/scr memory cell impairment during a viral challenge known to induce a large memory response. We bred the Malat1^scr/scr mice with P14 TCR transgenic mice that express an antigen receptor specific for the immunodominant LCMV GP33 peptide. We then transferred Malat1^scr/scr and WT P14 T cells into congenic CD45.1 hosts followed by LCMV Armstrong infection and tracked the acute and memory responses in the spleen and liver, the sites of primary LCMV infection (Figure 5A; Matloubian et al., 1993).

The total numbers of P14 cells were similar in recipients of transferred WT or Malat1^scr/scr P14 cells in both the spleen and liver at day 7, and in the spleen at day 31. However, there was a reduction in Malat1^scr/scr P14 cells in the liver at day 31 (Figure 5B, C). Despite having no effect on cell expansion at the peak of infection and mixed cell number results in the memory phase, the Malat1^scr allele had a distinct effect on the cellular phenotype across organs and time points. Malat1^scr/scr P14 cells preferentially displayed a phenotype associated with terminally differentiated effector memory cells (t-TEM), defined by KLRG1 expression and lack of CD127 expression as shown by Milner et al., 2020, and a corresponding reduction in the percentage of KLRG1⁻ CD127⁺ memory cells (Figure 5D–F). Previous work has shown that substantial heterogeneity exists in the memory pool and that CD27 and CD43 can be useful in delineating functional differences such as recall potential between different memory cells. For instance, CD27⁺ memory cells tend to produce more IL-2 than CD27⁻ cells (Milner et al., 2020), and CD43⁺ CD27⁺ cells are more effective at clearing Listeria upon re-challenge (Hikono et al., 2007; Olson et al., 2013). We therefore assessed these markers on the transferred Malat1^scr/scr and WT P14 cells. Consistent with increased KLRG1⁺ CD127⁻ t-TEM cells there were proportionally more CD27⁻ CD43⁻ cells in both the spleen and liver at day 31 among the Malat1^scr/scr transferred cells compared to WT (Figure 5G, I). CD27⁺ CD43⁻ cells were unchanged. The proportional reduction in CD127⁺ KLRG1⁻ cells corresponded with a decreased proportion of CD43⁺ CD27⁺ cells. Notably, the total numbers of all CD27⁻ subsets were unchanged. The proportional differences we observed were primarily driven by a reduction in the number of CD43⁺ CD27⁺ Malat1^scr/scr P14 cells compared with wild-type P14 cells (45% reduction in liver (p = 0.0058), 30% reduction in spleen (p = 0.113)) (Figure 5G–I). This phenotype is exactly reciprocal to the increase in CD27⁺ memory cells previously documented in miR-15/16^Δ/Δ mice (Gagnon and Ansel, 2019). To test whether the effect of the Malat1^scr allele was epistatic to miR-15/16, we bred Malat1^scr/scr to miR-15/16^Δ/Δ mice to generate triple mutant Malat1^scr/scr miR-15/16^Δ/Δ mice. The poly-clonal LCMV response in these mice phenocopied that of miR-15/16^Δ/Δ with a WT allele of Malat1, indicating that the observed effects of the Malat1^scr allele are epistatic to miR-15/16 (Figure 5—figure supplement 2). We conclude that Malat1 inhibits miR-15/16 in responding CD8⁺ T cells during LCMV infection, leading to fewer CD43⁺ CD27⁺ memory cells. Compared to LCMV, Listeria monocytogenes (LM) infection induces lower expression of multiple costimulatory ligands so the antigen-specific response is more sensitive to the ablation of CD28 costimulation (Welten et al., 2015). Therefore, we hypothesized that cytotoxic T cell response may be more impaired by the Malat1^scr allele during LM infection. We first sought to understand the effect of miR-15/16 on cytotoxic T cells during primary LM infection. To do so, we directly infected polyclonal mir-15/16^fl/fl and miR-15/16^Δ/Δ mice with LM expressing the LCMV GP33 peptide (LM-GP33) (Figure 6A). In contrast to the increased antigen-specific CD8⁺ T cell numbers in LCMV-infected miR-15/16^Δ/Δ mice (Gagnon and Ansel, 2019), we observed similar cell numbers in LM-GP33-infected miR-15/16^Δ/Δ and miR-15/16^fl/fl mice (Figure 6B). However, the effect on memory cell differentiation was still present with decreased percentages of CD127⁻ KLRG1⁺ t-TEMs and increased percentages of CD127⁺ KLRG1⁻ memory cells (Figure 6C, D). Again, miR-15/16^Δ/Δ antigen-specific cells had a significantly lower proportion of CD43⁻ CD27⁻ cells and a trend toward proportionally more CD43⁺ CD27⁺ cells (p = 0.127) (Figure 6E, F).

Given that antigen-specific miR-15/16^Δ/Δ T cells exhibited enhanced memory cell differentiation during LM-GP33 infection, we further tested the role of Malat1:miR-15/16 interaction in this model using the P14 adoptive transfer system (Figure 6F). Transferred Malat1^scr/scr and WT P14 cells accumulated in the spleen and the liver to similar numbers at both days 7 and 31 following LM-GP33 infection (Figure 6H). In contrast to LCMV, there was no change in the proportion or number of CD127⁻ KLRG1⁺ Malat1^scr/scr P14 cells at day 7 in the spleen or liver (Figure 6I). However, in LM-GP33 infection at day 7, CD43 and CD27 expression did not explicitly mirror the phenotype observed by CD127 and KLRG1 expression. In the spleen, there was a proportional and numerical increase in CD43⁻ CD27⁻ P14 cells with the Malat1^scr/scr genotype (Figure 6J). There was a decreased proportion of CD43⁺ CD27⁺ cells (p = 0.0531), but this was despite a trend toward increased numbers of these cells (p = 0.1056) in the Malat1^scr/scr P14 cells (Figure 6K). In the liver, there were no significant trends in any of these populations at day 7 (Figure 6J, K).

Although some of the phenotypes observed with LCMV were attenuated or absent during acute infection with LM-GP33 at day 7, Malat1:miR-15/16 interaction had more pronounced effects on memory cell populations at later times post infection (Figure 6L, M). On day 31, KLRG1⁺ cells were increased in proportion and numbers in the Malat1^scr/scr P14 cells (Figure 6N). In the liver, a similar trend in KLRG1⁺ cell proportion existed (p = 0.233), but the numeric effect was entirely absent (Figure 6N). Surprisingly, this effect was restricted to the KLRG1⁺ populations, as CD127⁺ KLRG1⁻ memory cells were unaffected in proportion and number in both the spleen and liver (Figure 6O). In both organs CD43⁻ CD27⁻ t-TEMs were proportionally increased and CD43⁺ CD27⁺ memory cells were proportionally decreased in Malat1^scr/scr P14 cells (Figure 6P, Q). The numeric underpinnings of these proportional changes were different in each organ, with a 94% increase in CD43⁻ CD27⁻ cells in the spleen, and a 40% loss of the CD43⁺ CD27⁺ population in the liver (Figure 6P, Q). Taken together, these data demonstrate that miR-15/16 restrict memory cell differentiation, and reveal the ability of Malat1:miR-15/16 interaction to enhance memory cell differentiation across infection contexts.

Finally, we investigated how the Malat1:miR-15/16 RNA circuit regulates memory cell differentiation and accumulation, giving consideration to the many direct targets of miR-15/16 and the emergent indirect effects on IL-2 and other costimulation-responsive gene expression in CD8 T cells. The pro-survival protein Bcl2 is the first characterized target of miR-15/16 (Cimmino et al., 2005). Higher abundance of Bcl2 in memory cells (as compared with short-lived effector cells) aids their survival and persistence, counteracting their increased expression of pro-apoptotic factors such as Bim (Kurtulus et al., 2011). As such, changes in the balance of pro- and anti-apoptotic factors can have selective effects on the accumulation of memory T cells. Bcl2 expression in Malat1^scr/scr P14 cells was reduced at day 31 in the CD43⁻ CD27⁻ population in both LCMV and LM-GP33 infection, and also in CD43⁺ CD27⁺ cells in LCMV infection (Figure 7A–C). This reduction is consistent with increased direct miR-15/16 action on the Bcl2 3′ UTR in the Malat1^scr/scr cells, but it may also be affected by IL-2 and other costimulation responsive factors. Also consistent with reduced Bcl2 expression, increased proportions of dead Malat1^scr/scr cells were observed in both the KLRG1⁺ and KLRG1⁻ cell fractions at day 7 during LM-GP33 infection (Figure 7D, E). Very few dead cells were detected among P14 cells of either genotype at day 31. IL-2 is critical for the induction of Bcl2 in memory T cells to promote their survival (Toumi et al., 2022). Furthermore, IL-2 production by cytotoxic T cells predisposes them to differentiate into a memory phenotype, likely by an autocrine/paracrine mechanism (Kahan et al., 2022; Kurtulus et al., 2011; Wojciechowski et al., 2007). Given the enhanced ability of miR-15/16^Δ/Δ cells to produce IL-2 and the trend toward impaired IL-2 production in Malat1^scr/scr cells in vitro (Figure 4D), we assessed IL-2 production by Malat1^scr/scr T cells in LCMV and LM-GP33-infected mice. Corroborating our in vitro observations, a lower proportion of Malat1^scr/scr P14 cells produced IL-2 when stimulated ex vivo on day 7 in both infection models (Figure 7F, G). This defect was also shared by KLRG1⁺ and KLRG1⁻ populations and limited to this critical early time point. Higher proportions of memory cells were able to produce IL-2 ex vivo at day 31 after infection with either LCMV or LM-GP33 infection, and Malat1^scr/scr P14 cells produced equivalent amounts of IL-2 (Figure 7H).

As CD28 costimulation is key to IL-2 production, the observed early defect is consistent with less robust activation while antigen is still present (Shapiro et al., 1997). Taken together, these results indicate that Malat1^scr/scr cells in the context of a viral or bacterial infection receive relatively poor activating cues and subsequently produce less IL-2 early during infection, contributing to a less robust pro-survival and pro-memory state. These findings illustrate how Malat1 regulation of miR-15/16 and its large target gene network can act through multiple connected nodes to coordinate gene expression programs essential to cytotoxic T cell responses (Figure 8).

WT C57BL/6 mice were bred in our facility. miR-15/16^Δ/Δ and miR-15/16^fl/fl mice were derived as described in Gagnon and Ansel, 2019. Malat1^scr/scr mice were generated from WT C57BL/6 mouse zygotes electroporated with CRISPR-Cas9 RNPs and HDR template (guide and template sequences below) as described previously (Chen et al., 2016). B6.SJL-Ptprca Pepcb/BoyJ (CD45.1) strain #002014 were purchased from the Jackson Laboratory. WT P14 mice carrying the TCR transgene specific to the LCMV GP33 peptide were obtained from the Waterfield lab and were backcrossed to C57BL/6 to maintain the line. Malat1^scr/scr mice were bred to the P14 line to generate Malat1^scr/scr P14 mice. Male and female age- and sex-matched mice were used between 5 and 12 weeks of age. All mice were housed and bred in specific pathogen-free conditions in the Animal Barrier Facility at the University of California, San Francisco. Animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) of the University of California, San Francisco (Protocol AN200003).

CD8⁺ T cells were isolated as below and stimulated for 3 days with αCD3 and αCD28 antibodies and grown with kool aid complete media. Subsequently, the cells were rested and expanded in kool aid complete media for 2 days supplemented with 100 U/ml recombinant Human IL-2 (R&D Systems Cat# 202-IL-010/CF). Subsequently, 1 × 10⁶ cells were used to prepare NGS libraries as described in Gagnon and Ansel, 2019 and Loeb et al., 2012. Samples were sequenced on HI-Seq 2500 (Illumina). Eleven nucleotide adaptors were trimmed from each read and resultant sequences were aligned to the mm10 genome using bowtie2 (Langmead and Salzberg, 2012). To assure lack of miR-15/16 binding in Malat1^scr/scr mice was not do to errors in alignment, reads from Malat1^scr/scr cells were aligned to a mm10 genome that contained a single modification changing Malat1 from the WT allele to the Malat1^scr allele. To determine maximum binding depth across the genome and to manipulate aligned files the samtools package was used (Li et al., 2009). To assess Ago2 binding, aligned HITS-CLIP reads were integrated across the follow genomic annotations: lncRNA genes from mouse Noncode v6 (http://www.noncode.org/download.php); miRNA target-binding sites from TargetScan V7.2 (https://www.targetscan.org/cgi-bin/targetscan/data_download.mmu80.cgi); mm10 introns, exons, 3′ UTRs, and coding genes from UCSC genome table browser (http://genome.ucsc.edu/cgi-bin/hgGateway?db=mm10). To assess differences in target binding at specific 3′ UTR sites between WT and Malat1^scr/scr cells, reads within each 3′ UTR were normalized to the total reads contained in the given 3′ UTR for each genotype. To identify regions of significant Ago2 binding above background, so-called ‘peaks’, we used the Piranha algorithm (Uren et al., 2012) (https://github.com/smithlabcode/piranha) on our HITS-CLIP. To remove sites from analysis where confounding features were present, such as miRNA expression sites or rRNA repeats, bedtools intersect was used to remove these features from annotation files (Quinlan, 2014). Human AHC data from publicly available datasets were trimmed and aligned to the Hg38 genome as previously described (Karginov and Hannon, 2013; Li et al., 2018).

Spleens were harvested from WT or Malat1^scr/scr P14 CD45.2 mice into phosphate-buffered saline (PBS) and passed through 70 μm strainers to generate a single-cell suspension. Samples were then stained for live/dead (apc-cy7), Thy1.2 (BV605), CD8ɑ (apc), and TCRVɑ2 (pe). Whole splenocytes were then transferred retro-orbitally (r.o.) into WT BoyJ (CD45.1) 7-week-old male recipients such that each recipient received 20,000 Thy1.2⁺ CD8ɑ⁺ TCRVɑ2⁺ cells in 200 μl PBS. To initiate LCMV infections, mice were injected intraperitoneally (i.p.) with 2 × 10⁵ plaque forming units (p.f.u.) LCMV armstrong in 200 μl plain RPMI. LCMV virus was produced in house as described in Shehata et al., 2018. To initiate L. monocytogenes infection, mice were injected r.o. with 2 × 10⁵ colony forming units (c.f.u.) Listeria monocytogenes-GP33 in 100 μl PBS. Listeria monocytogenes-GP33 was prepared in house as described in Allen et al., 2020. Blood was collected via submandibular bleeds with goldenrod 4 mm lancets collected into sample tubes coated with K2 EDTA (BD Ref# 365974) and 60 μl of blood was lysed with 500 μl of ACK lysis buffer. To assay spleens and livers, mice were sacrificed and organs harvested. To assay intracellular cytokine production, splenocytes were plated into 96-well u-bottom plates in complete kool-AID media and stimulated for 6 hr with a final concentration of 0.2 mg/ml GP33-41 (KAVYNFATM) in the presence of Brefeldin A.

Spleens and livers were harvested into 2% fetal bovine serum (FBS) PBS with and passed through 70 μm strainers to generate a single-cell suspension. Samples were then spun at 450 r.c.f. for 5 min and livers were resuspended in 20% Percoll. Percoll suspensions were spun at 741 r.c.f. for 20 min at 25°C and the supernatants discarded. Cell pellets for both spleens and livers were resuspended in 1 ml ACK lysis buffer and incubated at 4°C for 5 min. Lysis was stopped with 5 ml 2% FBS PBS and samples again spun at 450 r.c.f. for 5 min. Samples were resuspended in 2% FBS and aliquoted into v-bottom 96 well plates. Live dead staining was then performed using the fixable viability dye at 1:2000 in PBS. Subsequently, cells were blocked with mouse Fc block 1:100 in 2% FBS PBS and stained for surface proteins with directly conjugated antibodies diluted 1:100 in 2% FBS PBS. Stains were incubated for 20 min at 4°C protected from light. For surface stains alone, LCMV samples were fixed with 4% paraformaldehyde (PFA) for 5 min at 4°C. For intracellular stains, samples were fixed and permeabilized according the the Transcription Factor FIxation Kit (Invitrogen Cat#00-5521-00). Intracellular antibodies were diluted in 1:100 in permeabilization buffer and incubated at 4°C for 30 min. All samples were spun at 821 r.c.f. for 5 min prior to being resuspended in 2% FBS PBS with 1:10 AccuCount beads (spherotech Cat #ACBP-100–10) for analysis on either the BD LSR II or the BD LSRFortessa flow cytometry analyzer.

CD8⁺ T cells were isolated from spleens using negative selection from the EasySep Mouse CD8⁺ T Cell Isolation Kit (cat# 19853). Cells were counted using trypan blue staining and the nexelcom cellometer spectrum. Cultures were started via stimulation with plate-bound αCD3 (1 μg/ml, clone 2C11) and αCD28 (1 μg/ml, clone 37.51); plates coated overnight in PBS with Ca²⁺ and Mg²⁺ at 4°C. Cells were plated in Kool AID complete media (Dulbecco's Modified Eagle Medium (DMEM) high glucose media supplemented with 10% FBS, pyruvate, nonessential amino acids, minimum essential medium (MEM) vitamins, l-arginine, l-asparagine, l-glutamine, folic acid, beta mercaptoethanol, penicillin, and streptomycin) and spun at 450 r.c.f. for 5 min at 25°C to begin stimulation. For functional assays cells were harvested 2 or 4 hr after stimulation for flow cytometry, or 24 hr after stimulation for mRNA-seq. For supernatant cytokine expression, plates were spun at 450 r.c.f. for 5 min at 4°C 16 hr after stimulation and cell-free culture medium was collected and frozen at −80°C. TNFα (Cat #), IFNγ (Cat #), and IL-2 (Cat #) were analyzed by ELISA.

Spleens were harvested and single-cell suspensions generated by passing through 70 μm strainers. CD8+ T cells were isolated as above, and 2 × 10⁶ cells were pelleted at 450 r.c.f. for 5 min at 4°C. Cell pellet was resuspended in 700 μl Trizol reagent (Ambion cat #15596018) and kept at –80 °C. RNA was was isolated using using the Direct-zol-96 RNA Kit (zymogen cat #R2054). This RNA was used as input to the Mir-X miRNA qRT-PCR TB Green Kit (Takara cat# 638316) to generate miRNA cDNA. Specific primers to miR-15a, miR-15b, and miR-16 were used to quantify those miRNA species on the QPCR MACHINE (eppendorf realplex²). ribosomalRNA 5.8 s was used as a housekeeping control for each sample. Each sample was run in duplicate. To quantify miRNA expression technical duplicates were averaged and then normalized to rRNA 5.8 s by subtracting the rRNA 5.8 s Ct value from miRNA Ct value (ΔCt). Expression values reported are generated by 2^−ΔCt.

1 × 10⁶ CD8⁺ T cells were harvested 24 hr after αCD3 ± αCD28 stimulation as described above. Cells were pelleted at 450 r.c.f. for 5 min at 4°C. Cell pellet was resuspended in 700 μl Trizol reagent (ambion ref #15596018) and kept at −80°C. RNA was was isolated using using the Direct-zol-96 RNA Kit (zymogen cat #R2054). The integrity of total RNA was checked on Fragment Analyzer (Agilent, Cat. No. DNF-472), only RNA with RQN number of above 7 was used for library construction. The starting quantity of 100 ng of total RNA was used according to vendor instructions with Universal plus mRNA with Nu Quant (TECAN, Cat. No. 0520), final library PCR amplification was 17 cycles. After library completion, individual libraries were pooled equally by volume, and quantified on Fragment Analyzer (Agilent, Cat. No. DNF-474). Quantified library pool was diluted to 1 nM and further diluted as per protocol and sequenced on Illumina MiniSeq (Illumina, Cat. No. FC-420-1001) to check for quality of reads. Finally, individual libraries were normalized according to MiniSeq output reads, specifically by % protein-coding genes and were sequenced on one lane of NovaSeq6000 S4 PE100 (Illumina, Cat. No. 20028313). Reads were aligned to the mouse genome GRCm38 and quantified using the STAR aligner software version 2.7.2b. Read normalization and differential expression analysis were performed in the R computing environment version 3.6.1 using the software DESeq2 version 1.26. For RNA sequencing analysis, FDR-corrected p-values were used to evaluate significant differences between experimental groups using a significance threshold of 0.05. Lowly expressed genes that had fewer than 2 reads per million on average across all samples were removed from the analysis. Empirical cumulative density plots were made using the ggplot package for R. Heatmaps and hierarchical clustering were performed via the gplots package for R. Gene ontology analysis was performed using the statistical overrepresentation test for Panther pathways (Mi et al., 2019).

Malat1^scr/scr mice will be provided upon request.

All flow cytometry data were analyzed using FlowJo (version 10). Statistical analyses and plotting were performed using GraphPad Prism (Version 9.2.0) and R (version 4.2.1).

Ago2 HITS-CLIP data uploaded to NCBI GEO accession #GSE216565.

mRNA sequencing data uploaded to NCBI GEO accession #GSE216113.

Code used to analyze HITS-CLIP data summed over annotations and peaks in WT cells can be found at https://github.com/AnselLab/WT_HITS_CLIP_Analyses, (copy archived at Wheeler, 2023a).

Code used to analyze relative binding densities of HITS-CLIP reads can be found at https://github.com/AnselLab/Malat1_miR1516_AGO2_HITS_CLIP, (copy archived at Wheeler, 2023b).

Code used to analyze mRNAseq data for both miR-15/16 target expression and CD28 responsive genes can be found at https://github.com/AnselLab/Malat1_miR1516_CD3CD28_RNAseq, (copy archived at Wheeler, 2023c).