Pregnancy programs epigenetic and transcriptional exhaustion in memory CD8+ T cells

Alloreactive memory T cells, unlike naive T cells, fail to be restrained by transplantation tolerance protocols or regulatory T cells, and therefore represent a critical barrier to long-term graft acceptance. Using female mice sensitized by rejection of fully-mismatched paternal skin allografts, we show that subsequent semi-allogeneic pregnancy successfully reprograms memory fetus/graft-specific CD8+ T cells (TFGS) towards hypofunction in a manner that is mechanistically distinct from naive TFGS. Post-partum memory TFGS were durably hypofunctional, exhibiting enhanced susceptibility to transplantation tolerance induction. Furthermore, multi-omics studies revealed that pregnancy induced extensive phenotypic and transcriptional modifications in memory TFGS reminiscent of T cell exhaustion. Strikingly, at loci transcriptionally modified in both naive and memory TFGS during pregnancy, chromatin remodeling was observed exclusively in memory and not naive TFGS. These data reveal a novel link between T cell memory and hypofunction via exhaustion circuits and pregnancy-mediated epigenetic imprinting. This conceptual advance has immediate clinical relevance to pregnancy and transplantation tolerance.


Introduction 1
Alloreactive memory T cells are key mediators of acute and chronic graft rejection and represent a 2 potent barrier to transplantation tolerance in the clinic 1-4 . Alloreactive memory T cells can be generated by 3 direct sensitization to foreign MHC (via prior transplantation or blood transfusion), or via heterologous 4 immunity, whereby T cells primed during infections or exposure to environmental antigens are cross-5 reactive to donor MHC 5-8 . While numerous therapies can successfully induce transplantation tolerance and 6 cell-intrinsic hypofunction in naive T cells, memory T cells are resistant to these mechanisms [9][10][11] . Indeed, 7 we recently reported that the presence of memory T cells sensitized to only a single donor antigen is 8 sufficient to destabilize co-stimulation blockade-induced transplantation tolerance, underscoring the critical 9 need to identify mechanisms for controlling immunological memory responses long-term 12 .
10 Mammalian pregnancy has long been recognized as a model of spontaneous alloantigen-specific 11 tolerance, in which the maternal adaptive immune system must rapidly regulate responses towards the 12 semi-allogeneic fetus to preserve fetal viability [13][14][15] . We previously reported that pregnancy can efficiently 13 tolerize naive fetus-specific T cells, leading to the spontaneous acceptance of offspring-matched heart 14 grafts in B cell-deficient post-partum mice 16 . This maternal T cell tolerance is characterized by the 15 upregulation of coinhibitory markers, inhibition of pro-inflammatory cytokine production, and expansion of

19
and more specifically whether semi-allogeneic pregnancy would be able to successfully restrain memory 20 T cell responses, thus eliminating the barrier they pose to transplantation tolerance, remains largely 21 unknown.

22
In this study, we define the tolerizing mechanisms of pregnancy on memory fetus/graft-specific 23 CD8 + T cells (TFGS) that were generated by prior rejection of fully-mismatched paternal skin grafts. We 24 report that sensitized female mice consistently achieve spontaneous tolerance towards the semi-25 allogeneic fetus, resulting in pregnancy success rates comparable to those of naive mice. Remarkably, 26 pregnancy reprogrammed memory TFGS to a hypofunctional state that persisted post-partum, manifesting 27 as enhanced susceptibility to co-stimulation blockade-mediated transplantation tolerance. Using high-28 dimensional multi-omics, we demonstrate that pregnancy induced distinct phenotypic and transcriptional 29 profiles in naive versus memory TFGS, as well as an overlapping profile of exhaustion. Strikingly, we report 30 divergent epigenetic fates of memory versus naive TFGS during pregnancy: while the programming of T cell 31 exhaustion was correlated with extensive chromatin remodeling in memory TFGS, these epigenetic 32 modifications were absent in naive TFGS, even at loci of shared transcriptional modification. Collectively, 33 this study highlights the evolutionary robustness of mammalian pregnancy in constraining not only naive 34 1 reprograms memory CD8 + T cells towards hypofunction.

16
To investigate the effects of pregnancy on naive vs. memory TFGS, we expanded our experimental 17 model by including Naive and Sensitized mice without mating or pregnancy (Fig. 1a). We designed a 19-18 color spectral flow cytometry panel to profile TFGS phenotypes across multiple activation, memory, and 19 inhibitory markers (Supplementary Table 1). Using this panel, we analyzed OVA-specific CD8 + T cells at 20 day 30+ post-skinTx for Sensitized mice, or at post-partum day 0-3 for N+P and S+P mice. We observed 21 a significant increase in TFGS recovery from S+P mice compared to N+P mice . Despite this 22 expansion, S+P TFGS displayed elevated expression of multiple coinhibitory markers compared to 23 Sensitized TFGS, including PD-1, LAG3, TIGIT,and FR4 (Fig. 1e,. N+P TFGS also 24 upregulated FR4, CD73, and LAG3 compared to Naive TFGS.

25
To visualize TFGS phenotypes at single-cell resolution, we conducted uniform manifold 26 approximation and projection (UMAP) dimensionality reduction along with FlowSOM clustering, which 27 resulted in the identification of 4 major and 3 minor clusters (Fig. 1f). As anticipated, Naive and Sensitized 28 TFGS were largely homogenous, with >75% of these cells mapping to Cluster 1 or Cluster 4, respectively 29 ( Fig. 1g). In contrast, the effect of pregnancy on TFGS was heterogeneous, with ~50% of N+P and ~25% of 30 S+P TFGS remaining phenotypically similar to Naive or Sensitized TFGS, respectively (Fig. 1g). Notably,

31
Cluster 5 was identified as a shared cluster of pregnancy, comprising ~25% of both N+P and S+P TFGS 32 and defined by elevated expression of multiple coinhibitory markers, along with reduced expression of the 33 proliferation marker Ki67 (Fig. 1f-h, Extended Data Fig. 1c-d). Cluster 7 was unique to S+P TFGS; this cluster 34 1 d). Collectively, the phenotypic similarities observed between the Cluster 5 and Cluster 7 raise the 2 possibility that pregnancy programs hypofunction into naive and memory TFGS through the induction of a 3 shared set of modifications associated with anergy and exhaustion.

5
Pregnancy elicits a shared transcriptional signature in memory and naive TFGS 6 We next tested the hypothesis that the shared set of phenotypic markers induced by pregnancy in 7 both naive and memory TFGS was indicative of a broader set of shared transcriptional modifications. 8 Therefore, we performed genome-wide transcriptional profiling of TFGS by designing a cell sorting panel 9 that allowed us to sort TFGS into the 4 predominant phenotypic subsets observed in post-partum TFGS:

10
Cluster 1 (C1, naive-like phenotype), Cluster 4 (C4, sensitized-like phenotype), Cluster 5 (C5, shared by 11 N+P and S+P), and Cluster 7 (C7, unique to S+P) (Fig. 2a). The proportions of each cluster in this panel 12 were consistent with our original phenotypic data (Extended Data Fig. 2b,Extended Data Fig. 2a). This 13 approach allowed us to address the heterogeneity among pregnancy-modified TFGS while also retaining 14 the sequencing depth of bulk RNA-sequencing.

15
With this RNA-seq dataset, we first constructed a heatmap to visualize the transcriptional 16 expression of the markers used in our flow cytometry panel in Figure 1 and found that the expression 17 patterns in our transcriptional dataset were consistent with phenotypic data (Extended Data Fig. 2e). We 18 then performed differential expression analysis across 17,659 genes and visualized the global 19 transcriptional differences via UMAP and principal component analysis (PCA) dimensionality reduction 20 (Fig. 2c and Extended Data Fig. 2b). As expected, Sensitized and Naive TFGS displayed distinct 21 transcriptional signatures, with many genes and pathways for T cell activation and effector function 22 upregulated in Sensitized TFGS (Fig. 2c,. Furthermore, the N+P C1 subset was 23 nearly identical to Naive TFGS, and the S+P C4 subset was similar to Sensitized TFGS (Fig. 2c), corroborating 24 our above conclusion that a subset of TFGS remains unmodified by pregnancy in both Naive and Sensitized 25 mice. Remarkably, the global transcriptional profile between S+P C5 and N+P C5 appeared to be similar 26 via UMAP and heatmap, indicative of considerable transcriptional overlap between post-partum naive and 27 memory TFGS (Fig. 2c-e). We next analyzed the differentially expressed genes (DEGs) between S+P C7 28 vs. C5 TFGS and observed that these populations were more similar at the genome-wide level than initially 29 anticipated based on our phenotypic data (Extended Data Fig. 3b). However, modest transcriptional 30 differences were indeed present between these subsets, including enrichment of Inhibitor of DNA binding 31 (Id) signaling and NK cell activation pathways in S+P C7 (Extended Data Fig. 3a-d).

9
Despite the substantial transcriptional overlap, the majority of pregnancy-induced DEGs were 10 actually unique to either N+P C5 or S+P C5 TFGS (Fig. 3a). Pathway analysis of these DEGs indicated that 11 many of the uniquely expressed genes also belonged to pathways of T cell activation and cytokine

28
Notably, both Lag3 and Tigit were identified as part of the leading edge in our Gene Set Enrichment 29 Analysis (Fig. 4b). This finding was corroborated with multiple additional Tex gene sets from cancer, chronic 30 infection, and pregnancy ( Fig. 4c-e). As a control, we ran GSEA on the DEGs of Sensitized vs. Naive TFGS; 31 as expected, there was no enrichment for any exhaustion gene set in these DEGs (Extended Data Fig.   32 4c). 33 1 and S+P C5, and observed an enrichment of the upregulated Tex signature within the set of 196 DEGs 2 shared by N+P C5 and S+P C5 (Fig. 4c). Interestingly, the Tex signature was significantly enriched for only 3 the upregulated genes unique to N+P C5, whereas both upregulated and downregulated genes unique to 4 S+P C5 were enriched for the Tex signature ( Fig. 4d-e, Extended Data Fig. 4a). Taken together, these data 5 confirm that pregnancy induces a partially overlapping signature of transcriptional T cell exhaustion in both 6 naive and memory TFGS, but also provide new evidence suggesting that pregnancy utilizes additional 7 distinct mechanisms to program naive vs. memory TFGS.

8
We next sought to validate the transcriptional findings with phenotypic evidence. We designed a 9 larger 23-color spectral flow cytometry panel to assess the phenotypic expression of additional markers 10 identified in our transcriptional analysis (Supplementary Table 1). First, we confirmed the presence of an

21
Pregnancy induces distinct states of hypofunction in naive vs. memory TFGS

22
The broad phenotypic and transcriptional modifications consistent with T cell exhaustion prompted 23 us to test if pregnancy induced cell-intrinsic hypofunction in memory T cells 16,17,20,21 . We first quantified the 24 in-vitro cytokine production capability of CD8 + T cells following stimulation with allogeneic APCs. As 25 expected, ~12% and 30% of Sensitized TFGS produced TNF-α and IFN-γ, respectively (Fig. 6a). In contrast,

4
The distinct phenotypic and transcriptional signatures induced by pregnancy in N+P vs. S+P TFGS 5 also prompted us to investigate whether this resulted in differential susceptibility to NFAT inhibition, which 6 has been reported to hinder the induction of T cell exhaustion signatures 17 . Using FK506 as a 7 pharmacological inhibitor of NFAT, we showed that FK506 treatment hindered the expression of

13
Pregnancy programs extensive chromatin remodeling in memory TFGS, but not naive TFGS

14
It is well-established that CD8 + T cells undergo epigenetic modifications during both conventional 15 effector responses and exhausted/dysfunctional responses that promote the stability of effector and 16 dysfunctional phenotypes [25][26][27][28][29] . Therefore, we tested the hypothesis that pregnancy epigenetically 17 reprograms naive and memory TFGS to maintain distinct states of exhaustion and hypofunction. To test this, 18 we used the same sorting strategy as described above for RNA-Seq to perform the Assay for Transposase-

19
Accessible Chromatin with high-throughput sequencing (ATAC-Seq) and assess the chromatin   1 not N+P C5 actually occurred during skin sensitization rather than pregnancy, and that the chromatin 2 accessibility profile of S+P C5 TFGS may be more indicative of memory than exhaustion. Due to the level 3 of variability within Sensitized samples in our ATAC-Seq dataset, a simple pairwise comparison of these 4 two groups was not sufficient to rule out this possibility. Therefore, we leveraged insights obtained from 5 our RNA-Seq dataset to assess chromatin remodeling associated with pregnancy-induced DEGs unique 6 to either N+P C5 or S+P C5 TFGS ( Fig. 7c-d). Notably, both of these gene sets were enriched for T cell 7 exhaustion ( Fig. 4). At the loci of DEGs (n=831) induced uniquely in naive TFGS by pregnancy, we observed 8 no significant change in chromatin accessibility between N+P C5 vs. naive, or S+P C5 vs. Sensitized TFGS 9 ( Fig. 7c). Furthermore, at the loci of DEGs (n=817) induced in memory TFGS by pregnancy, we again 10 detected no significant changes in chromatin accessibility in N+P C5 vs. Naive TFGS. Importantly, significant 11 increases and decreases in chromatin accessibility were observed in S+P C5 vs. Sensitized TFGS at these 12 loci, corresponding to transcriptional up-and down-regulation (Fig. 7d). Similar chromatin remodeling was 13 also observed for S+P C7 TFGS (Extended Data Fig. 8f). These observations suggest that the chromatin 14 accessibility profiles of S+P-C5 and C7 are indeed associated with pregnancy, thus raising the possibility 15 that pregnancy-mediated chromatin remodeling is a mechanism unique to memory TFGS, while the 16 induction of exhaustion in naive TFGS does not require chromatin remodeling.

17
We next used HOMER de-novo motif analysis to search for enrichment of conserved transcription 18 factor DNA binding motifs associated with T cell function and differentiation among DAPs in S+P C5 vs.

19
Sensitized TFGS. This analysis identified key motifs associated with master regulators influencing both

29
Finally, to confirm that pregnancy-mediated epigenetic reprogramming occurs uniquely in memory 30 TFGS and is not detectable in naive TFGS, we focused our analysis on the 196 shared DEGs induced by 31 pregnancy in both N+P C5 and S+P C5 TFGS (Fig. 3). Visualizing the chromatin accessibility of these loci 32 via UMAP and box plots clearly demonstrated the significant epigenetic remodeling present in S+P C5 vs.

33
Sensitized TFGS, but not in N+P C5 vs. Naive TFGS (Fig. 8a+c). These differences are also readily apparent  Fig. 8f). Notably, this pregnancy-mediated chromatin remodeling remained detectable at 3 distances of up to 100kb from the transcription start sites of these loci, supporting the possibility of both 4 proximal remodeling of the locus itself along with distal enhancer remodeling (Fig. 8d). Taken together, we 5 conclude that pregnancy induces significant chromatin remodeling to reprogram memory CD8 + T cells 6 towards transcriptional and phenotypic exhaustion and increased susceptibility to transplantation tolerance 7 induction, while naive CD8 + T cells undergo minimal epigenetic modification to achieve this hypofunctional 8 state (Extended Data Fig. 9). 9 10 1 Pregnancy is an immunological paradox, where the conflict between robust immunity towards 2 foreign pathogens and tolerance to the semi-allogeneic fetus must be simultaneously resolved in order to 3 preserve the survival of the species. The imperative to preserve fetal viability underscores the necessity of 4 multiple redundant mechanisms to achieve fetal tolerance. In this study, we demonstrate the remarkable 5 ability of pregnancy to restrain not only naive, but also memory T cell responses towards the semi-6 allogeneic fetus to achieve full-term delivery of viable offspring, even after sensitization via prior rejection 7 of a fully mismatched paternal allograft. We identified a core transcriptional signature of 196 DEGs induced 8 by pregnancy in both naive and memory TFGS which was corroborated with phenotypic evidence, and 9 showed that this signature was highly enriched for CD8 + T cell exhaustion.

10
We revealed that the majority of the phenotypic and transcriptional signatures induced in naive and 11 memory TFGS following encounter with the semi-allogeneic fetus were unique. As a result, the overall state 12 of hypofunction achieved differed between naive and memory TFGS. Whereas post-partum naive TFGS 13 displayed a complete inhibition of proinflammatory cytokine production, post-partum memory TFGS were 14 only partially inhibited for TNF-α, and not for IFN-γ. Conversely, post-partum memory TFGS were more 15 robustly protected than naive TFGS against the loss of exhaustion markers in the presence of NFAT 16 inhibition (FK506). Importantly, pregnancy relieved the barrier memory TFGS normally pose to 17 transplantation tolerance, as evidenced by the enhanced survival of subsequent offspring-matched heart 18 grafts in S+P recipients under co-stimulation blockade. The potential mechanistic significance of these 19 nuanced differences between post-partum memory vs. naive TFGS may be explained by the observation 20 that pregnancy induced profound chromatin remodeling at exhaustion-associated loci exclusively in 21 memory CD8 + TFGS, whereas naive TFGS remained epigenetically unmodified during pregnancy. This 22 difference in epigenetic signature may provide an explanation for the enhanced resistance to FK506 23 treatment seen in S+P TFGS. We speculate that transcriptional and epigenetic differences could be 24 explained by differential responses to identical pregnancy-induced cues in naive vs. memory CD8 + T cells, 25 differences in migratory behavior between naive and memory CD8 + T cells that dictate divergent access 26 to pregnancy-induced cues, or by a combination of both factors.

27
Stable epigenetic programming has been reported in CD8 + T cells during the establishment of both 28 memory and exhaustion [25][26][27][28][29] . We report that pregnancy-induced increases in chromatin accessibility in S+P 29 TFGS were enriched for transcription factor motifs implicated in both early-and late-stage T cell exhaustion, 30 whereas motifs enriched in reduced chromatin accessibility included important hallmarks of T cell memory.

31
These data raise the tantalizing possibility that pregnancy utilizes targeted epigenetic manipulation in 32 memory TFGS not only to induce transcriptional exhaustion, but also to silence pre-existing memory/effector 33 programming. Taken together, our studies reveal a novel endogenous mechanism for the reprogramming 34 of antigen-specific memory T cells towards exhaustion and hypofunction. This insight is of critical relevance 35 for the success of transplantation tolerance in the clinic, where the conceptual framework for 1 reprogramming of memory donor-specific T cells heretofore has not been identified 9-11 .

2
Finally, viewing CD8 + T cell exhaustion/hypofunction through the lens of pregnancy potentially 3 solves the seemingly counterintuitive evolutionary puzzle of why exhaustion is so quickly induced when T 4 cells are exposed to chronic infections or tumors, which is often detrimental to the host. We theorize that 5 this timeline is imposed by mammalian pregnancy requiring a rapid restraint of fetus-specific alloreactive 6 T cells to preserve fetal viability. Moreover, while the phenotype and transcriptome of exhaustion was first 7 discovered in the context of chronic infection and tumors, we posit that this phenomenon should be re-8 evaluated from the perspective that exhaustion pathways developed under the stringent need to preserve 9 the semi-allogeneic fetus, and these mechanisms were subsequently hijacked by chronic infections and 10 tumors. Thus, insights into how exhaustion is programmed into naive and memory TFGS during pregnancy 11 are relevant not only to addressing problems related to high-risk pregnancies and transplantation 12 tolerance, but also to broader clinical issues such as autoimmunity, chronic infection, and cancer, where 13 controlling T cell hypofunction is also desirable.

Methods 1
Mice. Eight-to twelve-week-old female C57Bl/6 (B6, H-2 b ) mice were purchased from Harlan Laboratories.            6 Sequencing Data Analysis and Visualization. After completing data preprocessing as described above, 7 the DESeq v1.34.0 package was used to conduct differential expression/accessibility analysis on 8 sequencing datasets 47 . For both RNA-Seq and ATAC-Seq, the threshold for determining differential 9 expression/accessibility was FDR (padj)<0.1 and absolute value of log2 fold-change >0.9. In addition to 10 DESeq2, we used current versions of the following packages for analysis and visualization (with description 11 of purpose in parentheses). Viridis and RColorBrewer (color scale creation). Gplots, ggplot2 and ggrepel

ATAC-Seq motif analysis and locus visualization. Motif analysis was performed by identifying unique
15 and common peak sets between two experimental groups (using the reproducible peaks for each group 16 as described above). These peak sets were then analyzed via HOMER de-novo analysis 48 to search for                 . c-d, Box plots visualizing chromatin accessibility at DEGs unique to N+P C5 vs. N (c), or unique to S+P C5 vs. S (d) as identified in Fig. 3. e-f, HOMER de-novo analysis indicates nucleotide motifs associated with transcription factor binding that were significantly enriched in reproducible ATAC peaks opening (e) or closing (f) in S+P C5 vs. Sensitized T FGS . Data acquired from 2 or more biologically independent experiments. P values were determined by Welch's t-test. ns: not significant; *P<0.05; **P<0.01; ***P<0.001; ****P<0.0001.  Fig 3A) between N+P C5 and S+P C7.

2 3
Extended Data Fig. 9: Pregnancy adaptively utilizes multiple mechanisms to induce dysfunction in memory and naive T FGS . Graphical abstract. Pregnancy induces a shared set of modifications in both naive and memory T FGS , including transcriptional exhaustion, reduced cytokine production, and T cell hypofunction. However, substantial phenotypic and transcriptional differences were observed between N+P and S+P T FGS , indicating that reprogramming of naive vs. memory T FGS during pregnancy is mechanistically distinct. Finally, pregnancy mediates profound chromatin remodeling to induce hypofunction in memory T FGS , but these epigenetic modifications are not observed in naive T cells after pregnancy.

2 3 4
Supplementary Table 1: List of markers used in the two spectral flow cytometry panels used in this study.