Complex Interplay Between MAZR and Runx3 Regulates the Generation of Cytotoxic T Lymphocyte and Memory T Cells

The BTB zinc finger transcription factor MAZR (also known as PATZ1) controls, partially in synergy with the transcription factor Runx3, the development of CD8 lineage T cells. Here we explored the role of MAZR as well as combined activities of MAZR/Runx3 during cytotoxic T lymphocyte (CTL) and memory CD8+ T cell differentiation. In contrast to the essential role of Runx3 for CTL effector function, the deletion of MAZR had a mild effect on the generation of CTLs in vitro. However, a transcriptome analysis demonstrated that the combined deletion of MAZR and Runx3 resulted in much more widespread downregulation of CTL signature genes compared to single Runx3 deletion, indicating that MAZR partially compensates for loss of Runx3 in CTLs. Moreover, in line with the findings made in vitro, the analysis of CTL responses to LCMV infection revealed that MAZR and Runx3 cooperatively regulate the expression of CD8α, Granzyme B and perforin in vivo. Interestingly, while memory T cell differentiation is severely impaired in Runx3-deficient mice, the deletion of MAZR leads to an enlargement of the long-lived memory subset and also partially restored the differentiation defect caused by loss of Runx3. This indicates distinct functions of MAZR and Runx3 in the generation of memory T cell subsets, which is in contrast to their cooperative roles in CTLs. Together, our study demonstrates complex interplay between MAZR and Runx3 during CTL and memory T cell differentiation, and provides further insight into the molecular mechanisms underlying the establishment of CTL and memory T cell pools.


INTRODUCTION
CD8 + T cells play a central role during immune responses against viruses, intracellular bacteria and protozoan parasites and are also key regulators of anti-tumor immunity (1)(2)(3). Upon activation, antigen-specific CD8 + T cells proliferate and differentiate into a heterogenous population of "armed" cytotoxic T lymphocytes (CTLs), which release a large amount of cytolytic proteins as well as various inflammatory cytokines (1)(2)(3). While the majority of CTLs undergo apoptosis after the clearance of pathogens or tumors, a fraction of the cells is programmed for long-term survival and becomes memory T cells, which consist of various subsets distinguished by phenotypes, functions and locations (4)(5)(6)(7). During the last two decades, key transcriptional regulators controlling CTL and memory T cell differentiation have been identified, leading to an understanding of the basic transcriptional networks of the differentiation processes (8)(9)(10). However, the gene regulatory networks connecting cooperative as well as antagonistic interplay between these transcription factors are still not fully understood.
Myc-associated zinc finger-related factor (MAZR; also known as PATZ1 and encoded by the Patz1 gene) has been identified as an important transcriptional regulator controlling the development of CD4 + regulatory and CD8 + T cells (11)(12)(13)(14). MAZR belongs to the broad-complex, tramtrack and bric-à-brac (BTB) and zinc finger (ZF) containing transcription factor family, a family known to function both as transcriptional repressors and activators in a context-dependent manner (15,16). In double-negative (DN) thymocytes MAZR binds to multiple Cd8 enhancer regions and thereby negatively regulates CD8 expression (11). In addition, MAZR promotes cytotoxic lineage differentiation of MHC class I-selected double-positive (DP) thymocytes via repressing the expression of T-helperinducing POZ/Krueppel-like factor (ThPOK, encoded by the Zbtb7b gene) (12). Further, MAZR is required to maintain ThPOK repression also in CD8 + T cells (13). MAZR interacts with Runt-related transcription factor (Runx) proteins (i.e. Runx1 and Runx3) and MAZR/Runx1 and MAZR/Runx3 complexes synergistically repress ThPOK expression in preselected DP thymocytes and CD8 + T cells, respectively (13). Thus, these studies identified MAZR as an important regulator of CD8 lineage differentiation, which acts in synergy with Runx proteins. However, the role of MAZR during CTL and memory T cell differentiation remains unexplored. Since recent studies have demonstrated an essential role of Runx3 for CTL function and the generation of memory T cell precursors during LCMV infection (17)(18)(19), we hypothesized that an interplay between MAZR and Runx3 controls these processes. In the present study we examined the impact of loss of MAZR and/or Runx3 on CTL and memory T cell differentiation by generating and analyzing T cell lineage-as well as cytotoxic lineage-specific MAZR-, Runx3and MAZR/Runx3 double-deficient mice. Our study demonstrates that, while MAZR plays a compensatory role for Runx3-mediated transcriptional programs in CTLs, the two molecules exhibit distinct functions in the generation of memory T cell subsets. This suggests the differentiation stagespecific interplay between MAZR and Runx3, and thereby provides further insight into the transcriptional mechanisms governing the differentiation process of naïve CD8 + T cells into CTLs and memory T cells.

Retroviral Vector-Mediated Gene Transduction
Murine stem cell virus (MSCV)-based retroviral vectors containing the internal ribosome entry site (IRES)-GFP cassette were used for the overexpression of MAZR and distal promoter-derived Runx3 proteins (11,29). Retroviral vectors (20 mg) were transiently transfected into Phoenix-E packaging cells plated in 10-cm culture dish by using standard calcium phosphate precipitation. One day after transfection, the medium was changed into complete RPMI 1640 medium. Viral supernatants were collected on the following day, filtrated through a 0.45-mm filter, and used for the infection of CD8 + T cells. Sorted naïve CD8 + T cells were stimulated with platebound anti-CD3 and anti-CD28 as described above. On the next day CD8 + T cells were suspended in 1 ml viral supernatants containing 10 mg polybrene (Sigma) and centrifuged at 600 g for 2 hours at 32°C. After spin infection, cells were placed into 1 ml of fresh complete RPMI medium containing 20 U/ml rhIL-2. Cells were split 1:2 on the next day, cultured as described above and analyzed by a flow cytometer 6 days after activation.

Antibodies and Flow Cytometry
Antibodies used in this study are listed in Supplementary Table 1. Foxp3/Transcription Factor Staining Buffer Set (Thermo Fischer) and BD Cytofix Fixation Buffer followed by BD Perm/Wash Buffer (BD Biosciences) were used for the intracellular staining of transcription factors and cytokines, respectively. Intracellular MAZR and Runx3 expression was detected by anti-MAZR/PATZ1 (D-5: Santa Cruz Biotechnology) and anti-Runx3 (R3-5G4: BD Biosciences) antibodies, followed by Alexa Fluor 647 anti-mouse IgG 1 (RMG1-1: Biolegend) antibody. Flow cytometric data were collected with LSRII or Fortessa (BD), and analyzed with Flowjo software (BD Biosciences).

In Vitro Cytotoxicity Assay
For the in vitro cytotoxicity assay, naïve CD8 + T cells were differentiated into CTLs as described above. Fc receptor-positive P815-GFP + target cells (ATCC, TIB-64) were co-cultured with effector CD8 + T cells at 0.5:1, 1:1, 1:3 and 1:10 ratios in the presence of soluble anti-CD3 antibody at a total density of 2 x 10 5 cells/200 mL/well in 96-well U-bottom shaped tissue culture plates. Cells were incubated for 4h at 37°C. After the incubation period, cells were harvested and cell suspension was stained for Annexin V and CD8a, and apoptosis of target cells (GFP + ) was quantified by flow cytometry. As a negative control, effector and target cells were co-cultured without anti-CD3 antibody.

RNA Isolation and Next Generation Sequencing
Three biological replicates from each genotype were prepared for RNA-sequencing. Total RNA of in vitro generated CTLs (6 days after activation) was isolated with RNeasy kit (Qiagen) combined with DNase I digestion in the extraction columns. RNA concentration was measured using Qubit 2.0 Fluorometric Quantitation (Life Technologies), and the RNA integrity number was determined using Experion Automated Electrophoresis System (Bio-Rad). RNA-seq libraries were prepared using a Sciclone NGS Workstation (PerkinElmer) and a Zepyhr NGS Workstation (PerkinElmer) with the TruSeq Stranded mRNA LT sample preparation kit (Illumina). Library amount and quality were determined using Qubit 2.0 Fluorometric Quantitation (Life Technologies) and Experion Automated Electrophoresis System (Bio-Rad). The libraries were sequenced by the Biomedical Sequencing Facility at CeMM using the Illumina HiSeq 3000/4000 platform and the 50-bp singleread configuration.

Bioinformatic Analysis of RNA-Sequencing Data
The quality of the sequenced reads was checked using the FastQC tool (Babraham Bioinformatics, Babraham Institute, Cambridge, UK). STAR version 2.5.2b was used to align the reads to the mouse reference genome mm10, available at University of California, Santa Cruz Genome Bioinformatics Group (Illumina iGenomes website, San Diego, CA) (30). The number of uniquely mapped reads associated with each gene, according to RefSeq gene annotation, was counted using the Subread version 1.5.0 (31). The RNA-seq data are available from the Gene Expression Omnibus database (https://www.ncbi.nlm.nih.gov/geo; accession number GSE129772). The downstream analysis of the data was performed using R version 3.2.2 and its corresponding Bioconductor module 3.2. The count data were normalized using the trimmed mean of M-values method implemented in the edgeR R-package. The normalized data were further transformed using the voom approach in the limma R-package. R package limma was used for performing the statistical testing, and p-value < 0.05 and absolute fold change > 1.5 were required for detecting the differentially expressed (DE) genes between the sample groups. The Gh gene was removed from the final DE gene list as the inspection of the aligned reads revealed that the alignments for Gh had almost identical positions on a narrow region and contained many mismatches. The normalized expression values represented as Reads Per Kilobase per Million mapped reads (RPKM) were used as input for visualizations. Volcano plots and heat maps were generated with Prism 7 software (GraphPad). Venn diagrams were drawn with BioVenn web application (32). Enriched canonical pathways of the differentially expressed genes were predicted by using Ingenuity Pathway Analysis (Qiagen). To generate the list of the CTL signature genes, the gene expression pattern of vesicular stomatitis Indiana virus (VSV)-specific CTLs (8 days after infection) was compared with the one of naïve CD8 + T cells by utilizing the ImmGen database (33,34), and top 200 upregulated genes and the additional Cd8a and Cd8b1 genes were used to construct the list. Gene set enrichment analysis (GSEA) was performed with GSEA software from the Broad Institute (35).

Quantitative Real-Time PCR
RNA isolation was performed using a RNeasy kit (Qiagen) according to the manufacturer's instruction. To convert RNA samples into cDNA, samples were treated with SuperScript III reverse transcriptase and oligo(dT) primers according to the manufacturer's protocol (Invitrogen). For quantitative real-time PCR SYBR Green (Bio-Rad) and the 7300 Real-Time PCR system was used. Primer sequences have been describer previously (29).

Statistical Analysis
The statistical analyses were performed using Prism Software (GraphPad). For multiple comparisons in the majority of experiments, a one-way ANOVA analysis followed by Tukey's multiple-comparison test was performed. The p-values were defined as following: *, p < 0.05; **, p < 0.01; ***, p < 0.001. For comparisons of mean fluorescence intensity (MFI) values in vitro experiments (where MFI levels of WT cells were set as 1), a onesample t-test was conducted to compare MFI values between WT (i.e. 1) and each mutant group, whereas the values between three mutant groups were compared using a one-way ANOVA analysis as described above. The p-values based on a one-sample t-test (either p < 0.05, p < 0.01 or p < 0.001) were indicated above the respective diagrams. Finally, in the "overexpression" experiments ( Figure 2) a paired two-tailed Student's t-test was performed for comparisons of either MAZR-or Runx3-transduced mutant CTLs with control vector-transduced cells. The p-values were defined as described above. Differences that did not reach a statistically significant level (i.e. p ≥ 0.05) were not indicated.

Combined Activities of MAZR and Runx3
Are Required for Appropriate CTL Differentiation In Vitro To identify the unique role for MAZR as well as its potential synergistic activity with Runx3 during CTL differentiation, we utilized previously generated mice with a T cell lineagespecific deletion of MAZR (Mazr f/f Cd4-Cre), Runx3 (Runx3 f/ f Cd4-Cre) as well as MAZR and Runx3 double-deficient (Mazr f/ f Runx3 f/f Cd4-Cre) mice (hereafter referred to as MAZR-cKO CD4 , Runx3-cKO CD4 and M/R3-cDKO CD4 mice, respectively) (13). As previously reported (13,21,36), Runx3-cKO CD4 and M/R3-cDKO CD4 CD8 + T cells partially derepressed CD4 ( Figure S1A), and the analysis of the CD44/CD62L expression pattern showed a slightly increased proportion of the CD44 hi CD62Lsubset in the absence of Runx3, compared to wild-type control mice (Mazr f/ f Runx3 f/f , hereafter referred to as WT mice) ( Figures S1A, B). We isolated CD44 lo CD62L + naïve CD8 + T cells from the various mutants (including CD4 med-hi CD8 + T cell population in Runx3-cKO CD4 and M/R3-cDKO CD4 mice) and stimulated them in vitro with anti-CD3 and anti-CD28 antibodies. Naïve CD8 + T cells have been shown to differentiate in vitro during a period of 6 days into fully armed effector CTLs as characterized by high expression levels of Granzyme B, T-bet, Eomes and IFN-g (37). As previously reported (37), Runx3-cKO CD4 CTLs displayed a reduction in Granzyme B and a severe loss of Eomes expression during CTL differentiation ( Figures 1A, B). While MAZR-deficient CTLs did not show altered expression of Granzyme B and Eomes, the deletion of both MAZR and Runx3 led to a further downregulation in Granzyme B expression in comparison to Runx3-cKO CD4 CTLs ( Figures 1A, B). Moreover, whereas there was a mild reduction in T-bet expression in MAZR-cKO CD4 and also a tendency in Runx3-cKO CD4 CTLs, M/R3-cDKO CD4 CTLs displayed a greater degree of its downmodulation, indicating an essential role for the combined activity of the two factors ( Figures  1A, B). To study effector function in the absence of MAZR and/or Runx3, we also examined cytokine production and cytotoxic activity of the mutant CTLs. While there was no alteration in the expression level of IFN-g in the absence of MAZR, R3-cKO CD4 CTLs displayed a reduced level of IFN-g expression, and a similar reduction was also observed in M/R3-cDKO CD4 CTLs ( Figures  1A, B). There was also a similar tendency of change in IL-2 expression pattern, while TNF-a expression was not altered in the mutant cells (Figures S1C, D). Redirected cytotoxicity assay revealed that, while MAZR-cKO CD4 CTLs displayed a similar degree of cytotoxic activity to WT CTLs, the deletion of Runx3 led to almost complete abolishment of the activity ( Figure S1E). Finally, as we have previously shown (38), loss of Runx3 led to a down-regulation of CD8a expression in comparison to WT cells ( Figures 1C, D). While CD8a expression levels were also mildly affected by the loss of MAZR, the combined deletion of both MAZR and Runx3 led to a further downregulation of CD8a expression in comparison to the single mutant CTLs ( Figures 1C,  D). Of note, unlike Runx3-cKO CD4 CTLs, the deletion of MAZR (both on WT and Runx3-deficient backgrounds) had no impact on the proliferation and survival of activated CD8 + T cells ( Figures S2A-E), indicating that the phenotypic alterations observed in MAZR-cKO CD4 and M/R3-cDKO CD4 CTLs (compared to WT and Runx3-cKO CD4 CTLs, respectively) were not due to defects in the expansion of CD8 + T cells. In addition, the impact of MAZR deletion was not yet evident at the early activation phase (i.e. day 3), suggesting that MAZR regulates CD8a, Granzyme B and T-bet expression at the later stage of CTL differentiation ( Figure S2F, G). Together, these results indicate that, in contrast to an essential role of Runx3 for the expression of key CTL proteins and cytotoxic activity, the deletion of MAZR has a mild effect on CD8a and T-bet expression. However, on a Runx3-deficient background, additional loss of MAZR leads to a further reduction of Granzyme B expression, suggesting that MAZR partially compensates for loss of Runx3 in CTLs, albeit this appears to be not linked with cytotoxic activity. Moreover, the degree of reduction in CD8a and T-bet expression was greater in M/R3-cDKO CD4 CTLs compared to the single mutant cells, indicating synergistic activities of MAZR and Runx3 in CTLs in the regulation of the two molecules.

Ectopic Expression of MAZR Partially Reverses the Defects in CTL Differentiation by Combined Loss of MAZR and Runx3
To further study unique and/or combined activities of MAZR and Runx3 in CD8 + T cells, we next examined whether enforced MAZR or Runx3 expression reverts the impaired CTL differentiation phenotype observed in MAZR/Runx3-deficient CTLs (Figure 2A). While retroviral-mediated overexpression of Runx3 restored the expression of all factors analyzed to almost the levels observed in control vector ("empty")-transduced WT CD8 + T cell, enforced MAZR expression partially compensated for the loss of Runx3 with respect to restoring the expression of CD8a, Granzyme B and T-bet ( Figures 2B, C). Thus, these results underline a modulatory role of MAZR in the expression of a specific set of genes during CTL differentiation.

Deletion of MAZR and Runx3 Causes Impaired CTL Differentiation in a ThPOK-Independent Manner
ThPOK is a central transcription factor for helper lineage differentiation and required for the maintenance of the lineage integrity (39). We previously demonstrated that MAZR and Runx3 synergistically repress ThPOK during CD8 + T cell development and that the combined deletion of MAZR and Runx3 led to the derepression of ThPOK in approximately 30% of peripheral CD8 + T cells (13). We therefore examined whether the cooperative activity of those two factors is also required for ThPOK repression in CTLs. To easily monitor Thpok expression in mutant CTLs, we took advantage of previously generated MAZR-cKO CD4 , Runx3-cKO CD4 and M/R3-cDKO CD4 mice that were crossed with a ThPOK-GFP knock-in reporter strain (referred to as MAZR-cKO CD4/GFP , Runx3-cKO CD4/GFP and M/R3-cDKO CD4/GFP , respectively) (13). In line with previous reports (13, 40) a fraction of WT GFP CTLs expressed GFP (i.e. Thpok), while MAZR-cKO CD4/GFP CTLs displayed an increased proportion of GFP + CTLs ( Figures 3A, B). Notably, whereas Runx3-cKO CD4/ GFP CTLs express GFP at a similar level as WT cells, the combined loss of MAZR and Runx3 resulted in the derepression of ThPOK in a majority (approx. 80%) of CTLs, although their GFP expression level was lower compared to the one in the helper lineage ( Figures  3A, B). These data demonstrate that combined activity of MAZR and Runx3 is also required for ThPOK repression in CTLs.
A previous study showed that ectopic expression of ThPOK in CTLs leads to a reduction in the expression of CD8 and cytotoxic effector genes including Granzyme B (41). It is therefore possible that the impaired expression of effector genes in MAZR/Runx3-deficient CTLs is a consequence of increased ThPOK expression. In order to test this possibility, we crossed M/R3-cDKO CD4 mice with mice having a conditional Thpok allele (22) to generate MAZR/Runx3/ThPOK tripledeficient mice (referred to as M/R3/T-cTKO CD4 ). M/R3/T-cTKO CD4 CTLs displayed similar degrees of reduction in CD8a, Granzyme B and T-bet expression as observed in M/ R3-cDKO CD4 cells ( Figures 3C, D), indicating that the impaired expression of these molecules in MAZR/Runx3-deficient CTLs is not due to the enhanced expression of ThPOK.

MAZR Plays a Compensatory Role in the Runx3-Dependent Transcriptional Program of CTL Differentiation
In order to gain insight into the transcriptional program controlled by MAZR and Runx3 in CTLs, we performed RNA sequencing (RNA-seq) experiments and compared genome-wide transcriptional profiles of in vitro-activated WT, MAZR-cKO CD4 , Runx3-cKO CD4 and M/R3-cDKO CD4 CTLs. Overall, the deletion of Runx3 had a greater impact on gene expression in CTLs (2431 genes dysregulated) in comparison to MAZRdeficient CTLs (410 genes dysregulated) ( Figure 4A). However, the combined deletion of MAZR and Runx3 resulted in a more severe alteration in the transcriptome (2849 genes dysregulated) than Runx3-deficient CTLs ( Figure 4A). In line with the results obtained from the flow cytometric analysis (Figure 1), the RNAseq of M/R3-cDKO CD4 CTLs confirmed the downregulation of Cd8a, Cd8b1, Eomes, Gzmb (encoding Granzyme B) and Tbx21 (encoding T-bet) as well as the upregulation of Zbtb7b (encoding ThPOK), and some of the gene expression patterns were further validated by quantitative real-time PCR ( Figure S3A). This indicates that the expression of these factors is controlled at the transcriptional level by MAZR and Runx3. The comparison of differentially expressed genes in the individual mutant CTLs showed that there was a substantial overlap between Runx3-and MAZR/Runx3-depenedent genes ( Figures S3B, C). Nonetheless, a set of genes were uniquely dysregulated in MAZR-deficient CTLs, suggesting that MAZR regulates distinct pathways/ biological processes in CTLs. To further address the impact of MAZR and Runx3 on transcriptional programs during CTL differentiation, we investigated the expression pattern of CTL signature genes in WT, MAZR-cKO CD4 , Runx3-cKO CD4 and M/ R3-cDKO CD4 CTLs. Using the Immunological Genome Project (ImmGen) database (33), we determined the top 200 genes that are highly expressed in CTLs 8 days after vesicular stomatitis Indiana virus (VSV) infection in comparison to naïve CD8 + T cells (34). By adding Cd8a and Cd8b1 genes we defined a CTL signature gene set containing 202 genes. While loss of MAZR led to minor changes in the expression pattern of these CTL signature genes (one gene was up-and downregulated, respectively), Runx3-cKO CD4 CTLs up-and downregulated 12 and 87 CTL signature genes, respectively, in comparison to WT CTLs ( Figure 4B), thus confirming the key role of Runx3 for CTL differentiation (17,18,37,42). However, the combined deletion of MAZR and Runx3 led to a further increase in the number of downregulated CTL signature genes (i.e. from 87 to 108 genes) ( Figure 4B) due to reduced expression of an additional set of CTL signature genes ( Figure 4C). Moreover, among 87 signature genes downregulated in Runx3-cKO CD4 CTLs, five genes showed further downmodulation in M/R3-cDKO CD4 cells (based on a threshold of fold change > 1.5 and p-value < 0.05, Figure 4D upper panel), and 72 additional genes displayed a greater degree of reduction in their expression levels upon the combined deletion, albeit the differences did not reach the aforementioned threshold ( Figure 4D lower panel). These results indicate that M/R3-cDKO CD4 CTLs display a more severe impairment of CTL signature gene expression in comparison to Runx3-cKO CD4 CTLs. Of note, the analysis of upregulated CTL signature genes showed that the additional deletion of MAZR on a Runx3-deficient background had a lesser impact on a change in gene expression (Figures S3D, E). Finally, consistent with the further downregulation of CTL signature genes by the combined deletion of MAZR and Runx3, gene set enrichment analysis ( G S EA ) r e v e a l e d t h a t C T L s i g n a t u r e g e n e s w e r e underrepresented in M/R3-cDKO CD4 CTLs in comparison to Runx3-cKO CD4 cells ( Figure 4E). Together, the RNA-seq analysis underscores a compensatory role of MAZR for Runx3dependent transcriptome changes during CTL differentiation.
Altered In Vitro CTL Differentiation by Loss of MAZR and/or Runx3 Is Largely Due to CD8 + T Cell Lineage-Intrinsic Defects We have previously demonstrated that the combined deletion of MAZR and Runx3 during thymocyte development (using the Cd4-Cre strain) leads to the derepression of helper lineage genes in peripheral CD8 + T cells (13). Therefore, the phenotypic alterations observed in the mutant CTLs on a Cd4-Cre background might be, at least in part, due to defects in CD8 lineage differentiation during T cell development. In order to address this issue, we utilized an E8I-Cre strain to delete MAZR and/or Runx3, where the Cre recombinase is specifically active in   Figures 5A, B) as observed in the mutant cells on a Cd4-Cre background ( Figure 1). However, unlike MAZR-cKO CD4 CTLs, MAZR-cKO E8I CTLs displayed no alteration in CD8a expression and a tendency of reduction in Eomes and IFN-g expression levels. In addition, there was a greater degree of impairment in T-bet expression upon E8I-Cremediated MAZR deletion. Of note, a fraction of cells "escaped" Cre-mediated deletion of Runx3 in Runx3-cKO E8I and M/R3-cDKO E8I CTLs, and the expression pattern of key CTL proteins was therefore assessed within the Runx3-negative population (as determined by intracellular Runx3 staining). These data indicate that altered expression of CTL-characteristic proteins by loss of MAZR and/or Runx3 is largely due to CD8 lineageintrinsic defects.  (26,43), virtually all YFP + peripheral T cells in WT E8I/YFP mice were CD8 + T cells ( Figure S4A) and the efficient deletion of Mazr and/or Runx3 in the mutant YFP + T cells was confirmed by intracellular staining ( Figure S4B). E8I-Cre-mediated deletion of MAZR and/or Runx3 did not lead to alterations in the number as well as the homeostatic status (based on CD44/CD62L expression) of YFP + cells ( Figure S4A, C) and, unlike in Runx3-deficient mice on a Cd4-Cre background, no apparent CD4 derepression was observed in Runx3-cKO E8I/YFP and M/R3-  cDKO E8I/YFP YFP + T cells ( Figure S4A). We infected the mutant mice on a E8I-Cre/Rosa26-EYFP background with the LCMV Armstrong strain and performed immunophenotyping of virusspecific CD8 + T cells eight days after infection (44). In MAZR-cKO E8I/YFP mice, there was no alteration in the numbers of total YFP + T cells as well as glycoprotein 33-41 (GP33)-specific CTLs, compared to WT E8I/YFP mice ( Figures 6A, B). In contrast, Runx3-cKO E8I/YFP mice displayed reduced numbers of total YFP + T cells and GP33-specific CTLs, similar to the observations made in previous studies (17,18) (Figures 6A,  B). However, M/R3-cDKO E8I/YFP mice showed a tendency of a further reduction in their cell numbers in comparison to Runx3-cKO E8I/YFP mice ( Figures 6A, B), and a similar tendency was also observed for the number of nucleoprotein 396-404 (NP396)specific CTLs (Figures S5A, B). These results suggest the modulatory role of MAZR for clonal expansion of CTLs, at least in the absence of Runx3. We then characterized CD8a, Granzyme B, perforin, T-bet and Eomes expression in the mutant CTLs. This analysis revealed that the combined deletion of MAZR and Runx3 led to downmodulation of CD8a expression in comparison to WT and the individual deletions, in both GP33- (Figures 6C, D) and NP396- (Figures S5C, D) specific CTLs. In addition, Runx3-cKO E8I/YFP CTLs expressed Granzyme B at a lower level than WT E8I/YFP CTLs, and the combined loss of MAZR and Runx3 led to a further reduction of Granzyme B expression. Moreover, the expression level of perforin was lower in M/R3-cDKO E8I/YFP CTLs in comparison to WT E8I/YFP cells, despite its comparison to single mutant CTLs did not reach a statistically significant level. These results suggest that MAZR and Runx3 cooperatively regulate CD8a and perforin expression, and that their combined activity is required for appropriate expression of Granzyme B. In contrast, there was no alteration observed in the regulation of T-bet and Eomes in M/R3-cDKO E8I/YFP CTLs, whereas Runx3-cKO E8I/YFP CTLs showed an increase in Eomes expression compared to WT cells ( Figures 6C, D). In order to assess effector function of the mutant CTLs we examined the surface expression of CD107 protein, which is a marker for cytotoxic degranulation (45), as well as cytokine production upon GP33 peptide restimulation. This analysis revealed that, while Runx3-cKO E8I/YFP CTLs displayed an alteration in cytokine expression pattern as well as a tendency of reduced CD107 expression, the deletion of MAZR alone as well as the additional deletion of MAZR on a Runx3-deficient background had no impact on their expression (Figures S5E, F). However, there was a tendency that viral loads in the mutant mice were increased in the liver of M/R3-cDKO E8I/YFP mice, compared to WT E8I/YFP and single mutant mice ( Figure 6E). This suggests that MAZR and Runx3 cooperatively contribute to viral control, possibly in part via regulating the expansion of CTLs and/or the expression of cytolytic proteins.

MAZR and Runx3 Cooperatively Regulate CTL Differentiation in Response to Viral Infection
Finally, since a previous study demonstrated an essential role of Runx3 for promoting memory cell formation (18,46), we investigated the impact of MAZR-and/or Runx3-deletion on the generation of memory precursor (MP) CTLs. For this, we examined KLRG1 and CD127 expression patterns of the mutant CTLs, which allows us to identify KLRG1 -CD127 + MP and KLRG1 + CD127terminal effector (TE) CTLs (5,8,10). Consistent with the previous report (18), GP33-specific Runx3-cKO E8I/YFP CTLs displayed impaired MP cell differentiation, and the additional deletion of MAZR led to a similar degree of the impairment (Figures 6F, G). In contrast, MAZR-cKO E8I/YFP CTLs showed a reduction in the proportion of TE cells, while there was no change in MP cells (Figures 6F, G). NP396-specific mutant CTLs displayed similar alterations in KLRG1 and CD127 expression patterns (Figures S5G, H). These results indicate that MAZR and Runx3 play different roles for TE-versus-MP subset differentiation, albeit impaired TE cell differentiation by loss of MAZR become less evident on a Runx3-deficient background. In summary, the phenotypic analyses of virus-specific CTLs revealed that a combined activity of MAZR and Runx3 is required for the acquisition of certain CTL characteristics in vivo. In addition, they suggest distinct roles of the two molecules for the generation of memory CD8 + T cell precursor.

MAZR and Runx3 Exert Distinct Functions for the Generation of Memory T Cell Subsets
Having demonstrated altered TE/MP subset distribution in MAZR-and/or Runx3-deficient CTLs, we next characterized memory T cells in the mutant mice thirty days after infection. In line with a previous report (18), Runx3-cKO E8I/YFP mice showed reduced numbers of CD90.2 + YFP + T cells, and the deletion of MAZR had no impact on their numbers both in WT and Runx3deficient backgrounds (Figures 7A, B). Similar patterns of changes in cell numbers were also observed for the mutant memory T cells specific for GP33 (Figures 7A, B) and NP396 ( Figures S6A, B). We then assessed the distribution of KLRG1 -CD127 + and KLRG1 + CD127subsets in virus-specific memory T cells [hereafter designated as MP-and TE-like subsets, respectively, due to less clear relationship of the canonical markers to cell fate at memory time points (5,47)]. Notably, this analysis revealed that MAZR-cKO E8I/YFP GP33-specific T cells displayed an increase in the proportion of MP-like cells, along with reduced proportion of TE-like subset ( Figures 7C,  D). Moreover, while loss of Runx3 led to an increased percentage of TE-like subset (accompanied with an appearance of atypical KLRG1 -CD127subset), additional deletion of MAZR slightly reverted the phenotype (Figures 7C, D). The mutant NP396specific memory cells displayed similar alterations in the subset distribution ( Figures S6C, D), and these results therefore indicate opposite function of MAZR and Runx3 for TE-like subset differentiation.
We further characterized the mutant memory T cells based on CD62L/CD127 expression pattern [introduced by a recent study from Goldrath's lab (48)], which allows us to identify three memory T cell subsets: CD62L -CD127terminal effector memory [TEM, also known as long-lived effector CD8 + T cells (LLEC) (49)], CD62L -CD127 + effector-memory (EM) and CD62L + CD127 + central-memory (CM) subsets. While TEM cells display characteristics similar to TE cells (with a certain degree of longevity), EM and CM cells persist for extended periods of time (particularly CM cells) and preferentially localize in the vasculature and lymphoid tissues, respectively (4,6,7,48,49). The analysis of CD62L and CD127 expression revealed that the deletion of MAZR results in the enlargement of EM subset, along with a reduced proportion of CM cells (Figures 7E, F). In addition, while in Runx3-cKO E8I/YFP mice there was almost complete loss of EM and CM subsets, the additional deletion of MAZR led to a tendency of increase in the proportion of EM subset ( Figures 7E, F). The expression pattern of CD8a as well as key transcription factors largely reflected the altered subset distribution in the mutant mice ( Figures 7G, H). The deletion of MAZR led to the downmodulation of Eomes, whose expression is lower in EM cells in comparison to CM cells (48). Moreover, CD8a and TCF1 expression is reduced in Runx3-cKO E8I/YFP cells, which is in line with their lowest expression in TEM subset (48). Together, these results highlight a unique role of MAZR to restrain the generation of EM subset (possibly in part through regulating EM-versus-CM diversification processes), which is distinct from the function of Runx3 to promote long-lived memory T cell differentiation.

DISCUSSION
The transcription factors MAZR and Runx3 play an essential role for CD8 + T cell development, and synergistically repress ThPOK expression during the process (13). In the present study we examined the unique role of MAZR as well as its potential synergistic activity with Runx3 during CTL and memory T cell differentiation. Consistent with previous reports (17,18,37), our analysis of key CTL protein expression as well as genome-wide profiling of gene expression showed an essential role of Runx3 for CTL effector function. In contrast, the deletion of MAZR had a milder effect on expression patterns, indicating its less contribution to CTL differentiation. However, the additional deletion of MAZR on Runx3-deficient background results in much more widespread downregulation of CTL signature genes compared to single Runx3 deletion, indicating a compensatory role of MAZR for Runx3-mediated transcriptional program in CTLs. Interestingly, we observed that not all the Runx3dependedent genes/proteins were further downmodulated upon the combined loss of MAZR and Runx3, suggesting that MAZR only partially compensates for loss of Runx3 in the regulation of a specific set of genes. This was further supported by our observation that ectopic expression of MAZR in M/R-cDKO CD4 CTLs restored the expression levels of some factors (i.e. CD8a, Granzyme B and T-bet) but not others (i.e. Eomes and IFN-g). Our results also revealed that MAZR regulates a set of key CTL proteins in synergy with Runx3 [e.g. CD8a expression in vitro CTLs (Figure 1)], where the combined deletion of the two molecules resulted in a greater than additive reduction in their expression levels, compared to the individual deletions (see also Figure S7A for the definition of synergistic regulation). In line with this observation, our RNAseq analysis showed that 21 out of 104 genes downregulated in all the three mutant CTLs ( Figure S3C) are synergistically regulated by MAZR and Runx3 ( Figure S7B). Moreover, since part of the 21 genes displayed subtle "additive" changes upon the combined deletion, MAZR and Runx3 might regulate some common target genes in a parallel manner ( Figure S7A). Together, our data suggest that MAZR contributes to CTL differentiation via exerting a compensatory function for Runx3-mediated CTL programs. In addition, it regulates a smaller number of genes in a cooperative manner with Runx3, highlighting a complex interplay between the two molecules during the generation of CTLs. Mechanisms by which MAZR compensates for loss of Runx3 in the regulation of a set of genes in CTLs remain to be elucidated. However, considering that MAZR represses ThPOK expression in immature thymocytes via interacting with Runx1 (13), a similar interaction might mediate Runx3-independent CTL gene regulation. Interestingly, our transcriptome data as well as immunoblotting analysis showed that the deletion of Runx3 led to the elevated expression of Runx1 (but not MAZR) ( Figures S7C, D and data not shown). Hence, the enhanced activity of the MAZR/Runx1 complex (incl. their increased recruitment to the CTL gene loci) might be responsible for the compensatory function of MAZR. In order to test these hypotheses, it is essential to first identify genomic regions bound by MAZR (and Runx1) in CTLs on Runx3-sufficient and deficient backgrounds. Unfortunately, currently available anti-MAZR/PATZ1 antibodies are not suitable for ChIP-qPCR or ChIP-seq approaches in our hands (data not shown), and alternative approaches such as generating ChIP-grade anti-MAZR antibodies or in vivo tagging of MAZR have to be considered for future studies.
We demonstrated that the mutant CTLs on an E8I-Cre background displayed a similar expression pattern of key CTL proteins as observed upon Cd4-Cre-mediated deletion. However, there was also a slight difference in the pattern, such as no reduction in CD8a expression as well as a greater reduction in Tbet expression by loss of MAZR on the E8I-Cre background. Therefore, while the phenotypic changes upon deletion by Cd4-Cre are largely due to CD8 + T cell-intrinsic defects, to a certain degree they might result from altered thymic development in the mutant mice. Since we also used the Cd4-Cre deleter line to perform our transcriptome analysis, a fraction of differentially expressed genes might be regulated by MAZR and/or Runx3 during earlier developmental processes. Moreover, while MAZR has been shown to repress ThPOK expression in a post-thymic manner (13), for the rigorous assessment of MAZR/Runx3mediated ThPOK repression and its impact on CTL function, compound mutant mice on an E8I-Cre (or an inducible Cre) background remains to be elucidated.
Our data indicate that a combined activity of MAZR and Runx3 is required for appropriate CD8 and Granzyme B expression during in vitro and in vivo CTL differentiation. In contrast, altered T-bet expression in the absence of both MAZR and Runx3 was only observed in in vitro but not in in vivo CTLs. Such phenotypic differences between in vitro and in vivo CTLs was also detected upon Runx3 deletion (e.g. Eomes expression), which is in line with the observation made in a previous study (17). Given that in vivo CTL differentiation is initiated by multiple and distinct factors such as TCR signal strength, inflammatory cytokines and tissue microenvironment (3,50), it is conceivable that in vitro CTL generation does not fully follow the CTL differentiation process upon viral infection. Indeed, recent ATAC-seq analysis revealed a substantial difference in the accessibility profile between in vitro CTLs and LCMV-specific in vivo CTLs (51). Therefore, despite the similarity in expression pattern of key CTL molecules (52), in vitro and in vivo CTLs might be generated by differential transcriptional programs, and the contribution of MAZR and Runx3 to these programs might be different between in vitro and in vivo CTLs. In this regard, while a study from Xue's group has shown that Runx3-deficient CD8 + T cells are more prone to apoptosis upon LCMV infection (17), it remains unclear whether their impaired expansion in vivo ( Figures 6A, B) is due to defects in both proliferation and cell survival as observed in vitro (Figures S2A-E). CTL subsets are highly heterogeneous and their differentiation depends on the types of pathogens and tumors including differential cytokine milieu (e.g. predominant production of IL-12 and IFN-a, upon Toxoplasma gondii and LCMV infections, respectively) (53-57). One might even speculate that a differential interplay between MAZR and Runx3 is part of the mechanisms underlying the generation of the CTL diversity. Indeed, in "inflammatory" CTLs generated in vitro in the presence of IL-12 (58) MAZR and Runx3 regulate the expression of CTL proteins in a different manner ( Figure S7E, F), compared to CTLs stimulated with anti-CD3/28 "alone" (Figure 1). Moreover, it has been postulated that the combinatorial activity of two other transcription factors, T-bet and Blimp1, is important for robust CTL responses against various types of infections and tumor development (59). It might be therefore interesting to test the role of MAZR and Runx3 in CTL induction in other infection models or for cancer immunity.
Finally, our study revealed distinct roles of MAZR and Runx3 for memory T cell subset differentiation. Consistent with a previous study (18), Runx3 is required for the generation of MP subset at the effector phase, and contributes to the establishment of CD8 + memory T cells (in particular EM and CM cells). In contrast, MAZR appears to negatively regulate memory T cell differentiation, and its deletion impairs the differentiation of CTLs into TE subset and eventually leads to the enlargement of the EM population. Given that CTLs in vivo mainly consist of TE cells, the acquisition of CTL effector function might be tightly linked with naïve-to-TE cell differentiation. Our data suggest that this differentiation process is in part mediated by MAZR. Moreover, since Runx3deficient CTLs (contain approx. 80% of TE cells) displayed impaired effector function (e.g. impaired Granzyme B expression and reduced proportion of IFNg + TNFa + double producers), this indicates functional alteration of TE cells by loss of Runx3. Therefore, unlike their distinct roles in memory T cells, MAZR and Runx3 might cooperatively promote TE cell differentiation, through regulating different aspects of its differentiation process (i.e. MAZR initiates TE differentiation program, whereas Runx3 mediates their functional maturation).
Loss of both MAZR and Runx3 might result in combined defects in the generation of TE cells, leading to further impairment of CTL function in vivo. Since our transcriptome analysis identified unique sets of genes regulated by either MAZR or Runx3 ( Figures S3B, C), this might be linked with their differential regulation of TE cell differentiation (and also their distinct roles for the generation of memory T cells). During the last two decades, sets of transcription factors essential for memory subset differentiation (e.g. TE-versus-MP cell fate decision) have been identified (8)(9)(10). Among those, the roles of four transcription factors (T-bet, Blimp1, BCL6, FOXO1) have been recently reassessed with regard to TEM, EM and CM subset differentiation (using the "new" CD127/CD62L-based scheme) (48). This analysis revealed that T-bet and Blimp1 suppress development of EM and CM cells, whereas FOXO1 is required for the generation of both subsets (48). Therefore, these transcription factors exert partially overlapping function of either MAZR or Runx3, and the investigation of their molecular relationship/interaction with MAZR and Runx3 might provide further insight into a transcriptional network underlying memory T cell heterogenicity. In addition, given that EM cells consist of, at least, three transcriptionally distinct subsets (which include recently identified CX3CR1 int peripheral memory (PM) cells) (49,60) and that loss of Runx3 leads to the appearance of atypical KLRG1 -CD127or CD62L + CD127memory subset ( Figures 7C-F), it is of great interest to further characterize the subset composition of the memory T cell pool by loss of MAZR and/or Runx3, utilizing single cell-RNA sequencing combined with high-throughput flow cytometry.
In summary, our study demonstrated that MAZR compensates for loss of Runx3 during CTL differentiation, whereas the two molecules have distinct functions for the generation of memory T cell subset. This highlights a complex interplay between MAZR and Runx3 during CTL/memory T cell differentiation.

DATA AVAILABILITY STATEMENT
The datasets generated for this study can be found in the Gene Expression Omnibus database, accession number: GSE129772.

ETHICS STATEMENT
The animal study was reviewed and approved by Federal