Circulation of gut-preactivated naïve CD8⁺ T cells enhances antitumor immunity in B cell-defective mice

Significance

Accumulating evidence supports important roles for the microbiota in health and disease. The absence of IgA induces microbial dysbiosis, leading to inflammation in the gut environment. Here, we found that the strong antitumor immunity of B cell-deficient mice is due to their microbial dysbiosis, leading to activation of type I interferon (IFN) signaling in peripheral CD8⁺ T cells. The constant circulation of CD8⁺ T cells, particularly the naïve subset between the periphery and the gut, leads to the induction of a gut-educated naïve subpopulation in the periphery. Exposure to type I IFN in the gut endows this naïve subpopulation with superior effector potential. These data provide important insights into how the gut environment can shape peripheral immunity.

Abstract

The gut microbiome has garnered attention as an effective target to boost immunity and improve cancer immunotherapy. We found that B cell-defective (BCD) mice, such as µ-membrane targeted deletion (µMT) and activation-induced cytidine deaminase (AID) knockouts (KOs), have elevated antitumor immunity under specific pathogen-free but not germ-free conditions. Microbial dysbiosis in these BCD mice enriched the type I IFN (IFN) signature in mucosal CD8⁺ T cells, resulting in up-regulation of the type I IFN-inducible protein stem cell antigen-1 (Sca-1). Among CD8⁺ T cells, naïve cells predominantly circulate from the gut to the periphery, and those that had migrated from the mesenteric lymph nodes (mLNs) to the periphery had significantly higher expression of Sca-1. The gut-educated Sca-1⁺ naïve subset is endowed with enhanced mitochondrial activity and antitumor effector potential. The heterogeneity and functional versatility of the systemic naïve CD8⁺ T cell compartment was revealed by single-cell analysis and functional assays of CD8⁺ T cell subpopulations. These results indicate one of the potential mechanisms through which microbial dysbiosis regulates antitumor immunity.

It is widely accepted that the gut microbiome has a profound influence on host immunity (1–4). Furthermore, the efficacy of cancer immunotherapy and antitumor CD8⁺ T cell activities were shown to be modulated by the microbiota (5–8). The translocation of commensal bacteria to internal organs, which is augmented by microbial dysbiosis and mucosal barrier dysfunction, can induce inflammatory responses (9–11). Alternatively, microbial products may induce activation of lymphocytes with cross-reactive properties against tumor antigens, thereby boosting immune reactivity against tumors (12–14). However, the detailed mechanisms by which microbiota stimulation modulates antitumor immunity remain elusive.

B cells are central to adaptive immune responses as they secrete immunoglobulins (Igs) and cytokines critical for humoral immunity and also serve as antigen presenting cells (APCs). The role of B cells in antitumor immunity is controversial, with reports of both anti- and protumor activities. Similarly, clinical studies have identified tumor-infiltrating B cells as a negative or positive prognostic indicator in different cancers (15–19). In mice, studies have focused on the immunosuppressive regulation of peripheral T cells by B cells (20, 21). For example, CD40⁺ B cells were reported to compete with professional APCs for their interaction with T cells and deprive antitumor T cells of activation signals and IL-12 (22). Also, regulatory B cell subpopulations contribute to the suppression of antitumor immunity through the production of immunosuppressive cytokines, such as IL-10 and TGF-β, and facilitating the induction of FoxP3⁺ CD4⁺ T cells (23–26). However, all these studies have overlooked the indispensable role of B cells in IgA secretion for regulation of the gut microbiome, which tightly interacts with host immunity.

Secretory IgA plays important symbiotic roles by regulating the distribution, composition, and function of microbial communities along the gastrointestinal tract (1, 27, 28). Functional impairment or the complete absence of IgA leads to microbial dysbiosis and activation of host immunity, not only in mucosal tissues but also peripheral lymphoid tissues (1, 27, 29–31). Indeed, the gut microbiota was shown to regulate the cell compositions of the peripheral lymph nodes (pLNs) and spleen, as well as activation and differentiation of peripheral innate and adaptive immune cells (32–35).

To elucidate the role of B cells in modulation of antitumor responses, we studied two B cell-defective (BCD) mice models: μ-membrane targeted deletion (μMT) mice, which completely lack mature B cells, and activation-induced cytidine deaminase (AID) knockout (KO) mice, which exhibit hyperplasia of mature and activated B cells yet are deficient in class switch recombination and somatic hypermutation, and therefore IgA deficient. Here we show that elevated antitumor immunity in BCD mice depends on microbial dysbiosis and preferential migration of type I interferon (IFN)-preactivated naïve CD44^lowCD62L^high CD8⁺ T cells from the gut mucosal sites to peripheral lymphoid tissues. We demonstrate that naïve CD8⁺ T cells expressing the IFN-inducible protein stem cell antigen-1 (Sca-1) are educated in the gut, endowing them with enhanced mitochondrial activity and antitumor potential, resembling a memory-like phenotype. These results provide insights into the mechanisms by which the microbiota regulates the systemic naïve CD8⁺ T cell compartment and antitumor immunity.

Results

Gut Microbiota Requirement for Enhanced Antitumor Immunity of BCD Mice.

We examined antitumor immunity using murine MC38 colon adenocarcinoma tumor cells in BCD mice in specific pathogen-free (SPF) and germ-free (GF) conditions. In SPF conditions, µMT and AID KO mice showed antitumor immunity significantly superior to wild-type (WT) mice (Fig. 1A). There were no differences in antitumor responses between in-house and outsourced (company) WT mice, illustrating that microbiota differences in BCD mice, rather than facility variations of WT mice microbiota, is responsible for enhanced tumor immunity of BCD mice (SI Appendix, Fig. S1A). In agreement with previous studies (1–3), both µMT and AID KO mice showed microbial dysbiosis, evidenced by the changes in phylum abundance and reduced species diversity in BCD mice (SI Appendix, Fig. S1 B and C). Following antibiotics treatment or in GF conditions, the enhancement of antitumor activity was reduced in BCD mice (Fig. 1 B and C). Together, these data strongly support critical involvement of the microbiota in elevated antitumor activity of BCD mice.

Fig. 1.

Dysbiosis of gut microbiota enhanced antitumor immunity in B cell-defective mice. (A) Tumor growth and survival after inoculation with 5 × 10⁵ MC38 cells in SPF conditions. (WT n = 8, AID KO n = 6, µMT n = 6). Data are representative of three independent experiments. ANOVA analysis was performed on day 18. (B) MC38 growth in WT and BCD mice treated with antibiotics (ABX). Mice were treated with antibiotics for 2 wk prior to tumor inoculation and for 1 mo during tumor measurement (WT n = 7, AID KO n = 6, µMT n = 7). Data are representative of two independent experiments. ANOVA analysis was performed on day 18. (C) MC38 growth in GF mice (WT n = 9, AID KO n = 6, μMT n = 12). Data are pooled from two independent experiments. ANOVA analysis was performed on day 15. (D) The absolute number of CD8⁺ T cells per milligram of tumor tissue in SPF conditions (Left: WT n = 13, AID KO n = 13, and Right: WT n = 9, μMT n = 9). Mann–Whitney test was used. (E) The absolute number of CD8⁺ T cells in dLNs in SPF conditions (Left: WT n = 13, AID KO n = 13, and Right: WT n = 9, μMT n = 9). Mann–Whitney test was used. (F) Depletion of CD8⁺ T cells by anti-CD8 monoclonal antibodies (mAb) 2 d prior to MC38 inoculation in WT, AID KO, and μMT mice (n = 5). Data are representative for two independent experiments. ANOVA analysis was performed on day 18. (G) Depletion of CD4⁺ T cells by anti-CD4 mAb 2 d prior to MC38 inoculation in WT, AID KO, and μMT mice (n = 5). Data are representative for two independent experiments. ANOVA analysis was performed on day 18. (H) The absolute number of CD8⁺ T cells per milligram of tumor tissue in GF conditions (Left: WT n = 8, AID KO n = 10, and Right: WT n = 9, μMT n = 9). Mann–Whitney test was used. (I) The absolute number of CD8⁺ T cells in dLNs (Bottom) in GF conditions (Left: WT n = 8, AID KO n = 10, and Right: WT n = 9, μMT n = 9). Mann–Whitney test was used. Data represent the means ± SEM *P < 0.05, ***P < 0.001, ****P < 0.0001. ns, nonsignificant.

We next analyzed the CD8⁺ T cell compartment both at tumor sites and draining lymph nodes (dLNs) by flow cytometry. Under SPF conditions, the numbers of CD8⁺ T cells in tumors and dLNs was significantly increased in BCD mice compared to WT mice (Fig. 1 D and E and SI Appendix, Fig. S1D). Furthermore, the difference in antitumor immunity between WT and BCD mice was lost by depletion of CD8⁺ T cells (Fig. 1F), demonstrating that the antitumor immunity of BCD mice is primarily dependent on CD8⁺ T cells. On the other hand, BCD mice retained stronger antitumor immunity, relative to WT mice, following depletion of CD4⁺ T cells (Fig. 1G). In sharp contrast to SPF conditions, no increase in CD8⁺ T cells in tumors or dLNs was observed in GF BCD mice compared to GF WT mice (Fig. 1 H and I and SI Appendix, Fig. S1D). These results indicate that microbial dysbiosis in BCD mice confers superior antitumor immunity through activation and expansion of the CD8⁺ T cell compartment in dLNs and tumor sites.

Type I IFN Signature of CD8⁺ T Cells of BCD Mice Contributes to Their Enhanced Antitumor Immunity.

To investigate the functional differences of CD8⁺ T cells between WT and BCD mice, we carried out transcriptomic analysis of CD8⁺ T cells from the gut-draining mesenteric LNs (mLNs) or pLNs of BCD and WT mice in SPF conditions. Gene Ontology (GO) enrichment analysis revealed the major differences between CD8⁺ T cells from mLNs of WT and AID KO or µMT were related to virus response and innate immunity, in agreement with a previous report (Fig. 2A and SI Appendix, Table S1) (29). Among the “response to virus” GO category (the most significantly enriched pathway in both AID KO and µMT), CD8⁺ T cells from the mLNs of BCD mice showed robust up-regulation of type I IFN-inducible genes relative to WT (Fig. 2B and SI Appendix, Table S2). Further, we found enrichment of Kyoto Encyclopedia of Genes and Genomes (KEGG) virus response pathways for both mLN and pLN CD8⁺ T cells from AID KO and µMT mice, relative to WT, albeit the significance (P value) of enrichment was lower for pLNs than mLNs (SI Appendix, Fig. S2 A and B and Table S3). The up-regulation of type I IFN-inducible genes in peripheral CD8⁺ T cells of BCD mice compared with WT mice was confirmed by quantitative PCR (qPCR) analysis (Fig. 2C). Among these genes, we focused on the Ly6a gene as it encodes the IFN-inducible surface protein Sca-1 and therefore serves as a good type I IFN signaling signature (36–38). Indeed, the surface expression of Sca-1 was significantly increased in CD8⁺ T cells from BCD mice compared to WT controls in SPF conditions (SI Appendix, Fig. S2C and Fig. 2D). Furthermore, Sca-1 expression by gut and peripheral CD8⁺ T cells was lost or diminished in BCD mice crossed with IFN receptor α (Ifnra) KO mice as well as GF BCD mice (SI Appendix, Fig. S2D and Fig. 2E). Even in WT mice, the absence of IFN signaling (Ifnra) or microbiota (GF) reduced the frequency of Sca-1-expressing CD8⁺ T cells (SI Appendix, Fig. S2E). These results indicate that the microbiome induces a type I IFN signature in not only mucosal, but also peripheral CD8⁺ T cells and that dysbiosis enhances such a signature in BCD mice. The importance of type I IFN-inducible responses for the antitumor immunity of BCD mice was evaluated by tumor growth following tumor inoculation of IFN receptor α (Ifnra) KO and BCD Ifnra KO mice. The enhanced antitumor response of both AID KO and µMT mice was completely abrogated by the absence of the Ifnra (Fig. 2F), demonstrating that type I IFN signaling is indispensable for the strong antitumor response in BCD mice. Together, these data support that microbiota-dependent type I IFN stimulation of CD8⁺ T cells underlies the strong antitumor response of BCD mice.

Fig. 2.

Sca-1 is one of the type I IFN signatures and highly expressed on CD8⁺ T cells in SPF BCD mice. (A and B) Gene expression array of CD8⁺ T cells from mLNs of BCD and WT mice was performed. Enrichment of GO biological process categories for the up-regulated and down-regulated genes in CD8⁺ T cells of BCD mice in mLNs (A). The top five biological process categories are shown. For more information, please see SI Appendix, Tables S1. Scatterplot analysis of gene expression of AID KO and μMT CD8⁺ T cells for genes among the response to virus GO pathway (B). The genes that were significantly up-regulated, and down-regulated genes for both AID KO and μMT, relative to WT, are shown. We have included Ly6a in this plot (shown in red). For more information, please see SI Appendix, Tables S2. (C) Expression of representative type I IFN-inducible genes was measured by qPCR in spleen (Sp) and pLN of CD8⁺ T cells. Fold change of AID KO and µMT relative to WT is shown. Data are representative of two independent experiments (CD8⁺ T cells were pooled from Sp and pLNs of five individual mice). (D) Frequency of Sca-1 expression among total CD8⁺ T cells of pLNs and Sp in SPF conditions (WT n = 20, AID KO n = 20, µMT n = 20). (E) Sca-1 expression among total CD8⁺ T cells of AID KO and µMT mice in SPF, absence of Ifnra, and GF conditions (SPF n = 10, Ifnra KO n = 5, GF n = 5). Data are representative of two independent experiments. ANOVA analysis was performed. (F) Tumor growth of Ifnra KO and Ifrna KO BCD mice following inoculation with MC38 (Ifnra KO n = 12, Ifnra KO AID KO n = 5, Ifnra KO µMT n = 10). Data are pooled from two independent experiments. Data represent the means ± SEM *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ns, nonsignificant.

Efficient Circulation of Type I IFN-Activated P1 CD8⁺ T Cells from Gut to Peripheral Organs.

CD8⁺ T cells are usually categorized into naïve (P1: CD44^lowCD62L^high), central memory (P2: CD44^highCD62L^high), and effector memory (P3: CD44^highCD62L^low) subpopulations (SI Appendix, Fig. S3A). Next, we checked which subpopulation of CD8⁺ T cells predominantly up-regulates the microbiota-dependent type I IFN signature. Flow cytometry analysis of mLN cells revealed that all CD8⁺ T cell subpopulations significantly up-regulated Sca-1 in BCD mice compared to WT controls (Fig. 3A). In the spleen and pLNs, however, the most significant difference in Sca-1 expression between WT and BCD mice was seen in P1 CD8⁺ T cells (Fig. 3A).

Fig. 3.

CD44^low CD62L^high subpopulation in CD8⁺ T cells is preferentially affected by type I IFN due to dominant circulation between the gut and peripheral organs. (A) Representative histograms (Left) of Sca-1 expression in P1, P2, P3 CD8⁺ T cell subpopulations in mLNs, pLNs, and Sp. Average frequency (Right) of Sca-1 expression among the subpopulations of CD8⁺ T cells in mLNs, pLNs, and Sp in SPF conditions (WT = 12, AID KO = 12, µMT = 12). Data are pooled from two independent experiments. ANOVA analysis was performed. (B) t-SNE plots for WT and µMT transcriptome analysis of single CD8⁺ T cells, gated on P1, P2, and P3. Clustering and distribution were determined by similarity in gene expression. Each color represents a different cluster. While P1, P2, and P3 were clustered independently, WT and µMT were clustered together to compare similarity between cells in each subpopulation (P1, P2, and P3). (C) Frequency of significantly up-regulated or down-regulated genes in µMT relative to WT mice, in P1, P2, or P3 based on B. (D) Schematic diagram for marking mLN cells in µMT KikGR mice by exposure to violet light. The red arrows represent the migration of photoconverted KikGR-Red to the periphery. (E) Representative flow cytometry plots of KikGR-Green/Red in mLNs, pLNs, and Sp CD8⁺ T cells from µMT KikGR mice 0 h (Top) and 24 h (Bottom) after exposure of mLNs with violet light. (F) KikGR-Green cells in the mLNs are cells replaced since violet light exposure by migration of cells from the periphery. KikGR-Red cells in the pLN/Sp are cells which migrated from the mLNs to the periphery. (G) Representative histogram of Sca-1 expression among KikGR-Red (colored red) or KikGR-Green (colored green) CD8⁺ T cells (Left) and the average frequency of Sca-1⁺ in CD8⁺ T cells among KikGR-Green and KikGR-Red cells in the mLNs, pLNs, and Sp (Right). ANOVA analysis was performed. (H) The frequency of CD8⁺ T cell subpopulations among KikGR-Green (for mLN) and KikGR-Red cells (for pLN and Sp). ANOVA analysis was performed. (I) Absolute number of Sca-1⁺ KikGR-Red⁺ for each CD8⁺ T cell subpopulation in pLN (Left) and Sp (Right). ANOVA analysis. Data are representative of two independent experiments (mice, n = 3). Data represent the means ± SEM *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. ns, nonsignificant.

To further explore whether the up-regulation of Sca-1 expression in peripheral (pLN) P1 CD8⁺ T cells of BCD mice associates with a unique gene profile, we performed single-cell transcriptomic analysis. Clustering of P1, P2, and P3 CD8⁺ T cells (SI Appendix, Fig. S3B) by t-distributed stochastic neighbor embedding (t-SNE) revealed that the P1 subpopulation showed the least overlap between WT and µMT mice, in agreement with the Sca-1 expression profile (Fig. 3B). In further analysis of the same single-cell transcriptomic data, comparison of significantly up-regulated and down-regulated genes between WT and µMT for each CD8⁺ T cell subpopulation confirmed the greatest differences in gene expression were observed in P1 (Fig. 3C and SI Appendix, Table S4). In addition, KEGG virus response pathways were most significantly enriched for P1, compared to P2, and no significant changes were found for P3 (SI Appendix, Fig. S3C). Together, these results indicate that among peripheral CD8⁺ T cells, P1 is the most distinct subpopulation between WT and BCD mice.

Based on the microarray profile of type I IFN-related genes in mucosal and peripheral lymphoid CD8⁺ T cells (Fig. 2 A–C and SI Appendix, Fig. S2 A and B), we hypothesized that Sca-1⁺ P1 CD8⁺ T cells are induced by exposure to type I IFN at the mucosal sites and then relocate to the periphery. Thus, we performed tracking cell experiments by flow cytometry analysis of µMT Kikume Green-Red (KikGR) mice, which systemically express a fluorescent protein that converts from green to red upon exposure to violet light (Fig. 3D) (39). Exposure of mLNs to violet light converted about 99% of local CD8⁺ T cells from green to red (Fig. 3E). Twenty-four hours (h) postexposure, about 10% of the total CD8⁺ T cells in pLNs and spleen had migrated from the mLNs in µMT KikGR mice and 80% of total CD8⁺ T cells were replaced in mLNs (Fig. 3 E and F). Analysis of mLNs after 24 h revealed that cells which circulated from the periphery into the gut (KikGR-Green in mLNs) have lower Sca-1 expression than the “resident” (KikGR-Red in mLN) cells (Fig. 3G). Further, CD8⁺ T cells that had migrated from the mLNs to the periphery (KikGR-Red in periphery) had significantly higher expression of Sca-1 compared to CD8⁺ T cells that hadn’t migrated (KikGR-Green in periphery) (Fig. 3G). These results indicate that a large fraction of peripheral Sca-1-expressing CD8⁺ T cells were educated in the type I IFN-rich gut environment of BCD mice. The vast majority of circulating cells among CD8⁺ T cells were P1 (Fig. 3H), and P1 cells represented by far the largest Sca-1⁺ CD8⁺ T cells population migrating from the mLNs to the pLNs and spleen (Fig. 3I).

Flow cytometry analysis of Sca-1 expression in CD8⁺ T cell subpopulations from the gut (mLN) and periphery (pLN and spleen), revealed higher expression in the mLNs. These data support that the gut is the source of type I IFN signaling and subsequent up-regulation of Sca-1 (SI Appendix, Fig. S3D). Further, the differences were more pronounced for P2 and P3, compared to P1 (SI Appendix, Fig. S3D), which is likely due to the high circulation of P1 leading to the equilibration of Sca-1 expression across the peripheral and gut immune organs. Further, P2 and P3 may have resident cells in the gut which highly express Sca-1. Together these data indicate that P1 CD8⁺ T cells are educated to express Sca-1 in the type I IFN-enriched gut environment, and that these activated cells establish the gut–systemic connection by quickly migrating from the gut to the peripheral lymphoid tissues.

Phenotype of Gut-Educated Sca-1⁺ P1 CD8⁺ T Cells.

To characterize the gut-educated peripheral P1 CD8⁺ T cells, we sorted Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells from pLNs of BCD mice and analyzed their transcriptome profiles by microarray analysis. As expected, Sca-1⁺ P1 CD8⁺ T cells from both AID KO and µMT mice showed robust enrichment of type I IFN signaling compared to the Sca-1⁻ control from the same BCD mice (Fig. 4A). GO analysis revealed significant enrichment of mitochondrial cellular component categories in Sca-1⁺ P1 CD8⁺ T cells relative to Sca-1⁻ in both AID KO and µMT mice (Fig. 4B), in agreement with previous reports that up-regulation of type 1 IFN signaling promotes mitochondrial activity (40–42). Next, we used transmission electron microscopy (TEM) to further investigate the mitochondrial phenotype of Sca-1⁺ P1 CD8⁺ T cells and found a significant increase in the number of mitochondria and total mitochondrial area per cell compared with the Sca-1⁻ control (Fig. 4C). Further, measurement of oxygen consumption rate (OCR) revealed that Sca-1⁺ P1 CD8⁺ T cells have higher spare respiratory capacity (SRC), the mitochondrial capacity available under stressful conditions to produce ATP (SI Appendix, Fig. S4A and Fig. 4D) (41, 43, 44). Fatty acid oxidation (FAO) was also increased in Sca-1⁺ P1 CD8⁺ T cells relative to Sca-1⁻ P1 CD8⁺ T cells (SI Appendix, Fig. S4B and Fig. 4E). Higher SRC and FAO activities are known to be associated with quick activation and high-proliferation capacity upon antigen recognition (43–46). Therefore, we next compared the activation and proliferation potential of Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells. Sca-1⁺ P1 CD8⁺ T cells showed greater up-regulation of the early activation marker CD69 within 6 h after in vitro stimulation, clearly faster than Sca-1⁻ P1, albeit slower than memory/effector (P2/P3) CD8⁺ T cells (Fig. 4F). Thymidine incorporation after in vitro stimulation was also increased in Sca-1⁺ P1 CD8⁺ T cells relative to Sca-1⁻ controls (Fig. 4G), demonstrating increased proliferation potential. Sca-1⁺ P1 CD8⁺ T cells also have higher IFN-γ production relative to Sca-1⁻ controls after in vitro stimulation (Fig. 4H), suggesting enhanced cytotoxic potential. Finally, we directly compared the antitumor activities of Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells after their injection into tumor-inoculated RAG-2 KO mice. We found that the transfer of Sca-1⁺ P1 CD8⁺ T cells endowed RAG-2 KO mice with significantly higher antitumor immunity compared to Sca-1⁻ P1 CD8⁺ T cells (Fig. 4I). These results indicate that Sca-1⁺ P1 CD8⁺ T cells are preconditioned at mucosal sites in BCD mice and endowed with a preactivated phenotype which is closer to memory/effector T cells, contributing to a superior antitumor response.

Fig. 4.

Sca-1⁺ P1 CD8⁺ T cells are functionally and metabolically preconditioned, endowing with superior antitumor activity. (A and B) Gene expression array of Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells in pLNs from BCD was performed. Scatterplot analysis of gene expression profiling for Sca-1⁻ and Sca-1⁺ P1 CD8⁺ T cells. Highly up-regulated type I IFN-inducible genes are marked (A). Significant GO mitochondrial-related cellular component pathways for structure in Sca-1⁺ relative to Sca-1⁻ P1 CD8⁺ T cells of BCD mice. ns, nonsignificant (statistically) (B). (C) Representative image from TEM of mitochondria in Sca-1⁻ and Sca-1⁺ P1 CD8⁺ T cells of µMT mice (Left). The number of mitochondria and total mitochondrial area per cell are shown for Sca-1⁻ and Sca-1⁺ P1 CD8⁺ T cells of AID KO and µMT mice (Right). Mann–Whitney test was performed (pool of three independent experiments, n = 45 cells). (D) OCR curve (Left) and SRC (Right) (maximal − basal OCR) of Sca-1⁻ and Sca-1⁺ P1 CD8⁺ T cells for AID KO and µMT mice. Details of analysis are described in SI Appendix, Fig. S4A. Data are representative of three independent experiments (pool of more than 3 mice, n = 3 to 4 wells). Mann–Whitney test was performed. (E) Enhancement of basal and maximal OCR due to the utilization of exogenous palmitate/bovine serum albumin (BSA) (P/B) or control BSA (B) was measured to evaluate FAO activity. Details of analysis are described in SI Appendix, Fig. S4B. The difference of OCR between P/B and B condition was calculated for Sca-1⁻ and Sca-1⁺ P1 CD8⁺ T cells in AID KO and μMT mice for basal and maximal respiration. Data are representative of three independent experiments (pool of more than 10 mice, n = 3 to 5 wells). Multiple t tests with Holm–Šidák correction were performed. (F) CD69 expression among Sca-1⁻ and Sca-1⁺ P1, P2+P3 (pooled) CD8⁺ T cells after 6 h of stimulation with anti-CD3/CD28 mAb. Data are representative of two independent experiments (pool of more than three mice, n = 3 wells). ANOVA analysis was performed. (G) The proliferation of Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells 36 h after anti-CD3/CD28 stimulation were assessed by thymidine incorporation assay. Mann–Whitney test was performed. (H) The IFN-γ level 24 h after anti-CD3/CD28 stimulation was assessed by ELISA. Mann–Whitney test was performed. (I) 2 × 106 Sca-1⁺ or Sca-1⁻ P1 CD8⁺ T cells were injected into RAG-2 KO mice on days 2, 7, and 10 after MC38 inoculation. Mann–Whitney test on day 20 was performed. Data represent the means ± SEM *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Together, our study provides insight into the mechanisms by which microbial dysbiosis enhances antitumor immunity and shapes the phenotype of peripheral CD8⁺ T cells in BCD mice (summarized in Fig. 5).

Fig. 5.

Schematic diagram of the mechanism of preconditioning of naïve CD8⁺ T cells in the gut and subsequent migration to the periphery to join the antitumor response. 1) Microbiota dysbiosis in BCD mice leads to the 2) enrichment of type I IFN and subsequent 3) exposure of CD8⁺ T cells in the gut environment. 4) Preferential migration of P1 CD8⁺ T cells from the gut to the periphery leads to the 5) induction of preactivated Sca-1⁺ P1 CD8 T cells in the periphery and 6) promotes the antitumor CD8⁺ T cell response.

Discussion

Growing evidence indicates that microbial dysbiosis can stimulate the host immune system and shape the phenotype of peripheral CD8⁺ T cells, particularly effector and memory CD8⁺ T cells (1, 47–49). In this study, we found that microbial dysbiosis in the gut mucosal environment engendered a type I IFN signature in CD8⁺ T cells of BCD mice. The predominant circulation of naïve (P1) CD8⁺ T cells between the periphery and gut leads to the expression of IFN-inducible surface protein Sca-1 on a peripheral P1 subpopulation. Gut-educated Sca-1⁺ P1 CD8⁺ T cells are preactivated, but have not differentiated into memory (P2) or effector (P3), with enhanced activation and proliferation capacity, elevated mitochondrial function, and antitumor activities relative to Sca-1⁻ P1 cells. The Sca-1⁺ P1 population was significantly reduced or absent in BCD mice rendered deficient in type I IFN signaling (Ifnra KO) or lacking commensal bacteria (antibiotics [ABX] and GF).

In our analysis of antitumor immunity and type I IFN signaling in the two models of B cell deficiency, the phenotype of μMT was consistently stronger than AID KO mice. These mouse strains are fundamentally different; AID KO have B cells and can produce IgM and as a result microbial dysbiosis may be less extreme compared to μMT. This idea was supported by species richness, for which AID KO was intermediate between WT and μMT (SI Appendix, Fig. S1C). Greater dysbiosis could explain the more striking up-regulation of type I IFN signaling in the gut and further enhancement of antitumor immunity of μMT. In addition, CD4 depletion, which eliminated the difference in tumor growth between the two BCD strains, suggests that CD4⁺ T cells contribute to the superior antitumor immunity of μMT mice. Consistent with this hypothesis, μMT mice had significantly reduced numbers of FoxP3⁺ CD4⁺ T cells, relative to WT and AID KO mice (SI Appendix, Fig. S5A). It is of note that μMT retained this difference under GF conditions (SI Appendix, Fig. S5B). These data may explain the residual differences between WT/AID KO and μMT mice after antibiotics treatment and in GF conditions. There are likely other inhibitory factors produced by B cells which may contribute to differences between these two BCD strains of mice, given AID KOs still have semifunctional B cells.

Our analysis of KikGR mice revealed that P1 has the greatest propensity to migrate from the gut to the periphery, relative to P2 and P3 differentiated CD8⁺ T cell populations, providing important insight into the migration dynamics of CD8⁺ T cells following exposure to commensal bacteria. Although previous studies on the effect of the microbiota on CD8⁺ T cells have focused on effector and memory CD8⁺ T cells (47–49), our results demonstrate that the naïve subpopulation may be integral to the potential of the microbiota to modulate systemic immunity, including antitumor immunity. Further analysis of GF and Ifnra KO BCD KikGR mice will be required to understand exactly how microbial dysbiosis and type I IFN in the gut microenvironment affects the circulation and functional properties of CD8⁺ T cells.

CD8⁺ T cell subpopulations, including naïve, effector memory, and central memory, are commonly defined according to the expression of surface markers, such as CD44 and CD62L in mice or CD45RA/CD45RO and CCR7 in humans (50–52). We used single-cell t-SNE analysis to compare the gene expression profile similarity between WT and µMT mice in each CD8⁺ T cell subpopulation (Fig. 3C). We further examined the homogeneity among each of the P1, P2, and P3 subpopulations from both genotypes and unexpectedly found that in both WT and µMT mice (SI Appendix, Fig. S6 A and B), “naïve” (P1) and “differentiated” (P2+P3) CD8⁺ T cells were intermixed. These data suggest that although the conventional cell surface markers roughly distinguish subpopulations, CD8⁺ T cells consist of a spectrum of cells spreading from true naïve to fully differentiated effector and memory cells. Even in the naïve population, our single-cell analysis data showed striking heterogeneity, as revealed by the presence of multiple distinct cell clusters in the t-SNE plot (Fig. 3C). Sca-1⁺ P1 and Sca-1⁻ P1 CD8⁺ T cells, with significant differences in their gene expression profiles and functional phenotypes, exemplify the heterogeneity of naïve CD8⁺ T cells in BCD mice. Together, these data suggest that a metabolic axis could help to distinguish between distinct CD8⁺ T cell subpopulations.

Mitochondrial metabolism is an important determinant of quiescence or activation and distinguishes naïve, memory, and effector CD8⁺ T cells (53, 54). Quiescent naïve CD8⁺ T cells have low metabolic activity and are dependent on oxidative phosphorylation (OXPHOS) to generate ATP (55–58). Effector CD8⁺ T cells are characterized by high OXPHOS and significant enhancement of glycolysis (59–62), whereas memory CD8⁺ T cells are characterized by high FAO but low glycolysis (41, 63). In our study, baseline OXPHOS and glycolysis levels (indicated by basal OCR and extracellular acidification rates (ECAR), respectively) were similar between Sca-1⁺ and Sca-1⁻ P1 CD8⁺ T cells (SI Appendix, Fig. S7 A and B). In contrast, Sca-1⁺ P1 CD8⁺ T cells had increased FAO, as well as higher mitochondrial number and mitochondrial area per cell, relative to the metabolically naïve Sca-1⁻ P1 population. In addition, Sca-1⁺ P1 CD8⁺ T cells had increased SRC, a key property of memory CD8⁺ T cells which contributes to their long-term survival and rapid response upon cognate antigen reencounter (41, 62–65). Overall, our data show that the energy metabolism profile of Sca-1⁺ P1 cells resembles that of memory CD8⁺ T cells. Therefore, we propose that enhanced mitochondrial capacity of Sca-1⁺ P1 cells facilitates a preactivated ready-to-respond state, endowing them with superior activation, proliferation, and effector potential. This proposal is supported by previous reports that type I IFN increases SRC and reduces the activation threshold of CD8⁺ T cells (40, 66, 67).

Interestingly, the Sca-1⁺ P1 CD8⁺ T cell population also bears resemblance to T stem cell memory (T_SCM) cells, a rare memory T cell subset, which are characterized as multipotent progenitors with the capacity to self-renew and replenish CD8⁺ T cell effector pools (68–70). T_SCM cells are defined as CD44^low CD62L^high and express CD122 (IL-2/15 receptor β-chain), B cell lymphoma 2 (BCL-2), and CXC-chemokine receptor 3 (CXCR3) in addition to Sca-1 (68, 71–73). Sca-1⁺ P1 CD8⁺ T cells have increased Cxcr3 gene expression and surface protein CXCR3 levels relative to Sca-1⁻ P1 (Fig. 4A and SI Appendix, Fig. S7C). Further, the functional phenotypes of Sca-1⁺ P1 CD8⁺ T cells, particularly their enhanced mitochondrial function, proliferation, and antitumor potential, are shared with T_SCM. However, CD122 and BCL-2 were not coexpressed on Sca-1⁺ P1 CD8⁺ T cells (SI Appendix, Fig. S7D).

Overall, our study provides mechanistic insight into how the microbiota can shape peripheral immunity and has important implications for immunotherapies which target the microbiota or naïve CD8⁺ T cells. Recent studies of chimeric antigen receptor (CAR) T cell therapy have demonstrated that more naïve T cells are better for clinical outcome than fully activated effector T cells because “younger” effector T cells have greater capacity to replenish the effector T cell pool (74–78). Our Sca-1⁺ P1 CD8⁺ T cell population is intermediate between naïve and memory, young but with high effector potential. This phenotype suggests that weak type I IFN stimulation of naïve CD8⁺ T cells could be ideal for CAR T therapy. Further study is necessary to identify human markers of microbiota modulation of naïve CD8⁺ T cells. Sca-1 does not have a human homolog; however, its coexpression with CXCR3 compliments a recent study which used CXCR3 to identify human naïve CD8⁺ T cells with enhanced effector differentiation potential and showed CXCR3⁺ mouse and human naïve CD8⁺ T cells are transcriptionally equivalent (79).

Materials and Methods

Mice.

C57BL/6 WT mice were purchased from CLEA Japan. C57BL/6 AID KO and KikGR were generated as described previously (4, 5). C57BL/6 μMT mice were purchased from The Jackson Laboratory. C57BL/6 Ifrna KO mice were donated from Michel Aguet (80). WT, AID KO, μMT, Ifnra KO, Ifrna KO μMT, Ifrna KO AID KO, Rag2 KO mice, and μMT KikGR were maintained under SPF conditions at the Institute of Laboratory Animals, Graduate School of Medicine, Kyoto University (all strains), RIKEN Center for Integrative Medical Sciences (WT and μMT), and Faculty of Pharmaceutical Sciences, Osaka Ohtani University (μMT KikGR). GF WT, AID KO, and μMT mice were bred and maintained at RIKEN Center for Integrative Medical Sciences. μMT KikGR mice were bred and maintained at Kyoto University and the Faculty of Pharmaceutical Sciences, Osaka Ohtani University. Mice used in the experiments were of both genders and 2 to 4 mo old. The study was approved by the respective institutional review boards.

Cell Lines.

Murine colon adenocarcinoma (MC38) cells were kindly provided by James P. Allison, Memorial Sloan Kettering Cancer Center, New York, NY. Cell lines were cultured in RPMI medium (Gibco; 11875-093) with 10% (vol/vol) heat-inactivated fetal bovine serum and 1% (vol/vol) penicillin-streptomycin mixed solution (Nacalai Tesque). Cell lines were free of mycoplasma contamination and were used within the fifth passage.

Tumor Models.

MC38 cells (5 × 10⁵) were intradermally injected on the right flank (day 0). Tumor measurement was performed on each alternate day, and tumor volume was calculated using the formula for a typical ellipsoid: π (length × breadth× height)/6. In antibiotic experiments, mice were given the following antibiotics in drinking water: ampicillin (Nacalai Tesque) 0.5 g/L; neomycin sulfate (Nacalai Tesque) 1 g/L; imipenem monohydrate (Chem-Impex International, Inc.) 0.25 g/L. The antibiotics were administered 2 wk prior to tumor inoculation and for a total of 6 wk (until day 30). Sterilized milliQ was used for control mice and bottles were changed every 3 d. In the anti-CD8⁺ and anti-CD4⁺ T cell-depletion experiment, 50 µg of anti-CD8 mAb (clone 53-6.7) (Bio X Cell) and 100 µg of anti-CD4 mAb (clone GK1.5) (Bio X Cell) was injected intraperitoneally 1 d before tumor inoculation. Therapy was repeated four times, every 5 d.

Feces and Flora Analysis.

Feces were collected freshly from individual mice, weighed, and then frozen. Following the extraction of genomic DNA, 16s rRNA-specific PCR was performed by Repertoire Genesis, followed by next-generation sequence analysis using Miseq. The sequence data were analyzed using the Flora Genesis software. In brief, the R1 and R2 read pairs were joined and then chimera sequences were removed. After this preprocessing, operational taxonomic unit (OTU) picking was conducted by the open-reference method using the uclust and 97% ID prefiltered Greengenes database. The representative sequences of each OTU were picked up and taxonomy assignment was done by the ribosomal database project (RDP) classifier using its threshold score 0.5 or more. The OTUs were grouped if their annotations were the same regardless of RDP score. The species richness was calculated as described previously (81). The species number of AID KO and μMT was divided by sequence depth, in ratio to WT (sequence depth WT = 188,483, AID KO = 199,026, μMT = 187,776).

Cell Preparation for Analysis.

For dLN analysis, cells from axillary, brachial, and inguinal LNs on the right side of tumor-bearing mice were harvested. These three dLNs per individual mouse were pooled and homogenized. For pLN analysis, cells from axillary, brachial, and inguinal LNs on both sides (total of six pLNs) of mice without tumors were harvested, homogenized, and pooled. For spleen analysis, the spleen was harvested and homogenized and then treated with ACK Lysing Buffer for 2 min to lyse red blood cells (RBCs). For mLN analysis, cells from three to five mLNs per individual mouse were harvested, homogenized, and pooled. Averaged cell numbers per one LN or spleen were used to calculate absolute cell numbers. For tumor-infiltrating lymphocyte (TIL) analysis, tumor samples were minced into 1- to 2-mm pieces with scissors and digested with collagenase type IV (Thermo Fisher Scientific) using a gentleMACS Dissociator (Miltenyi Biotec). The number of TILs per milligram was used to calculate absolute numbers. For in vitro analysis, CD8⁺ T cells were stimulated with anti-CD3 and CD28 mAb beads (Thermo Fisher Scientific) for staining or measurement of proliferation by thymidine incorporation.

Flow Cytometry Analysis.

The following monoclonal antibodies recognizing the indicated antigens were used: CD44 PB (1M7), CXCR3 APC (CXCR3-173), CD8 PE (53-6.7), CD62L PE-Cy7 (MEL14), SCA-1 APC-Cy7 (D7), and BCL2 Alexa Fluor 488 (BCL/10C4) from BioLegend and CD44 BV510 (1M7), CD62L BUV395 (MEL14), CD8 BV605 (53-6.7), CD4 BUV496 (GK1.5), Foxp3 Alexa Fluor 488(FJK-16s), and CD122 PE (TM-B1) from BD Bioscience. TIL cells were seeded but not stimulated for 2 h. Brefeldin A/monesin (1,000×, Thermo Fisher Scientific) was then added for 2 h before staining. All flow cytometry experiments were performed on a FACSCanto II (BD Biosciences) or LSR Fortessa X-20 (BD Biosciences) and analyzed using FlowJo software. For assessment of intracellular proteins or intranuclear transcription factors, cells were stained according to the Foxp3/Transcription Factor Staining Buffer Set (eBioscience).

Cell Isolation and Sorting.

CD8⁺ T cells or naïve (CD44^low) CD8⁺ T cells were purified by autoMACS (Miltenyi Biotec; 130-096-543) and/or the Mojosort System (MojoSort; 480008, 480044) according to the manufacturer’s instructions. We further isolated the interested population (e.g., Sca-1⁺ P1 CD8⁺ T cells) by FACSAria (BD Biosciences). Purified cells were used for RNA-seq microarray, flow cytometry, OCR, and thymidine analyses. The purity of sorted cell populations was confirmed to be >95%.

Measurement of Oxygen Consumption Rates and Fatty Acid Oxidation.

OCR of isolated cells was measured using an XFe96 Extracellular Flux Analyzer (Seahorse Bioscience). Cells (4 × 10⁵ per well) were seeded in an XFe96 plate, as previously described (82). The three pharmaceutical modulators of mitochondrial oxidative phosphorylation, which were included in the XFeCell Mito Stress Test Kit (Seahorse Bioscience), were used for both OCR and FAO assays. For OCR analysis, XF DMEM (Agilent Technologies,103575-100) base medium with 1 mM Hepes and pH 7.4 was supplemented with 1 mM pyruvate (Nacalai), 2 mM glutamine (Nacalai), and 10 mM glucose (Nacalai). Oligomycin (1.5 μM), carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (FCCP) (1 µM), and rotenone/antimycin A (1 μM) were injected sequentially. For FAO, XF RPMI (Agilent Technologies,103576-100) base medium with 1 mM Hepes and pH 7.4 was supplemented with 2.5 mM glucose and 0.75 mM carnitine (Sigma-Aldrich) on the day of the assay (Sigma-Aldrich). Palmitate (200 μM) or bovine serum albumin (BSA) control was added immediately before the assay. For measurement, 2 µM FCCP was used. The concentrations of the other chemicals (oligomycin and rotenone/antimycin) were consistent with OCR.

For analysis, basal OCR was defined as (last rate measurement before the first injection) − (nonmitochondrial respiration rate). Maximal respiration was defined as (maximum rate measurement after FCCP injection) − (nonmitochondrial respiration rate). SRC was defined as (maximal respiration) − (basal respiration). For exogenous FAO, we first calculated oxygen consumption due to uncoupling by free fatty acid (the difference in OCR after oligomycin injection between palmitate/BSA (P/B) and BSA control groups). Basal respiration due to utilization of exogenous fatty acids was defined as the (difference between basal rates in P/B and BSA treated wells) − (oxygen consumption due to uncoupling by free fatty acid). Maximal respiration due to utilization of exogenous fatty acids was defined as the (difference between maximum rates in P/B and BSA treated wells after FCCP injection) − (oxygen consumption due to uncoupling by free fatty acid). See SI Appendix, Fig. S4 A and B for diagrammatic explanation.

Microarray Analysis.

Total RNA was extracted from cells pooled from n = 5 mice (per group) using the RNeasy Micro kit (Qiagen) according to the manufacturer’s protocols. The data were deposited at the GEO repository (GSE155632). Microarray analysis was performed by Macrogen using GeneChip Mouse Gene 2.0 ST Array. Type I IFN-inducible genes are defined according to the INTERFEROME database (http://interferome.its.monash.edu.au/interferome/). Gene lists were analyzed in DAVID for enrichment analysis of biological process ontology (GO) and KEGG pathways (83, 84). For analysis of CD8⁺ T cells in mLNs and pLNs, the genes with fold change more or equal to twofold were selected. Data analysis of differentially expressed genes was conducted using R3.5.1 (https://www.r-project.org/).

Real-Time RT-PCR.

We isolated RNA from purified CD8⁺ T cells with the RNeasy Mini Kit (Qiagen) and generated cDNA by SuperScript III reverse transcriptase (Invitrogen) according to the manufacturer’s instructions. The list of genes and primers can be found in SI Appendix, Table S5. Expression in WT, AID KO, and μMT CD8⁺ T cells was normalized to the geometric mean of the housekeeping gene HPRT to control the variability in expression levels. Fold expression between WT and BCD was analyzed using the 2-ΔΔCT method described by Livak and Schmittgen (85).

Cell Circulation Analysis Using the KikGR Model.

Photoconversion of the mLN KikGR mice was performed as described previously (39, 86–88). Briefly, KikGR mice were anesthetized, and mLNs were exposed to violet light for 3 min (33 mW/cm2 from USHIOSP500 spot UV curing equipment with a 405-nm bandpass filter). Mice were killed 24 h after photoconversion and organs were harvested for flow cytometry analysis.

Single-Cell Analysis.

Generation of single-cell gel beads in emulsion and sequencing libraries for single cell.

Samples for single-cell analysis were performed in one batch. The data were deposited at the GEO repository (GSE156338). Single-cell suspensions were labeled with oligo DNA-tagged antibodies against mouse CD44 and CD62L (TotalSeq-A, BioLegend) to detect cell surface protein (also known as CITE-Seq) according to the 10× Genomics cell labeling procedure. Single-cell gel beads in emulsion (GEM) generation and library construction were performed by GENEWIZ, Japan. Briefly, labeled single-cell suspensions were loaded onto 10× Genomics Single Cell Chip B along with the reverse transcription (RT) master mix, cell partitioning oil and gel beads as per the manufacturer’s protocol to generate GEMs. The Chromium Single Cell 3′ Library Construction Kit v3 (10× Genomics) was used to establish single-cell 3′ Gene Expression libraries.

The resulting libraries were assessed by gel electrophoresis (Agilent D1000 BioAnalyzer) and Qubit dsDNA assay and further quantified with qPCR (Illumina KAPA Library Quantification Kit). Following normalization to 2 nM, libraries were denatured and diluted to 17 pM of cluster generation using the Illumina cBot (HiSeq PE Cluster Kit v4) and then sequenced on Illumina HiSeq platform with 150-bp PE configuration.

Data sequencing and analysis for single cell.

The sequencing data were processed with the Cell Ranger Single Cell Software Suite (version. 3.0.2) provided by 10× Genomics. Raw base call files from the HiSeq sequencer were demultiplexed with the cellranger mkfastq pipeline into library-specific FASTQ files. Next, cellranger count performed alignment against mouse reference mm10, filtering, barcode counting, and unique molecular identifiers (UMI) counting, yielding the cell/gene expression matrix files and Loupe Cell Browser files. Two datasets were further aggregated by cellranger aggr, enabling the comparison between two groups.

The Loupe Cell Browser files for data analysis generated by Cell Ranger in three sets. For the t-SNE plot comparing WT and μMT CD8⁺ T cell subpopulations, we imported the cells in P1, P2, or P3 separately. For comparison of gene expression for P1, P2, and P3 CD8⁺ T cell subpopulations, we imported all CD8⁺ T cells (WT and μMT together). For comparison of P1, P2, and P3 clustering overlap, we imported all CD8⁺ T cells (WT and μMT separately). To analyze the significant differences in gene expression between WT and μMT for P1, P2, and P3 CD8⁺ T cells, we performed locally distinguishing significant feature comparison in Loupe Cell Browser. P values were adjusted using the Benjamini–Hochberg correction for multiple t tests. The −log10(P value) of the significance of the change in gene expression was calculated for each gene differentially expressed (P value >0.05). Significant gene lists for each subpopulation were analyzed in DAVID for KEGG pathways analysis (83, 84)

TEM.

For TEM observation, we isolated Sca-1⁺ or Sca-1⁻ P1 CD8⁺ T cells and embedded them in iPGell (GenoStaff), according to the manufacturer’s instructions. The cell blocks were fixed with 4% formaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (PB) (pH 7.4) overnight at 4 °C, then postfixed with 1% O_SO₄ in 0.1M PB for 2 h. Following dehydration in a series of graded concentrations of ethanol, the fixed cell blocks were embedded in epoxy resin (Luveak 812; Nacalai Tesque). Ultrathin sections (70-nm thickness) were prepared on an ultramicrotome (EM UC6; Leica). The sections were then stained with uranyl acetate and lead citrate and finally examined with an electron microscope (H-7650 Hitachi Tokyo). TEM was performed at the Division of Electron Microscopic Study, Center for Anatomical Studies, Graduate School of Medicine, Kyoto University.

Statistical Analysis.

Statistical analysis was performed using Prism 6 (GraphPad Software). One-way ANOVA analysis followed by Dunnett’s multiple comparisons test was used to analyze three or more variables. To compare two groups, the Mann–Whitney test or multiple t tests with Holm–Šidák correction were performed. All statistical tests were two-tailed assuming nonparametric data, and a P value of <0.05 was considered significant. The variations of data were evaluated as the means and SEM. Five or more samples were thought to be appropriate for the sample size estimate in this study. Samples and animals were randomly chosen from the pool and treated. No blinding method was used for the treatment of samples and animals.

Data Availability.

Data have been deposited in the Gene Expression Omnibus repository (accession IDs GSE155632 and GSE156338).