Peripheral and systemic antigens elicit an expandable pool of resident memory CD8+ T cells in the bone marrow.

BM has been put forward as a major reservoir for memory CD8+  T cells. In order to fulfill that function, BM should "store" memory CD8+ T cells, which in biological terms would require these "stored" memory cells to be in disequilibrium with the circulatory pool. This issue is a matter of ongoing debate. Here, we unequivocally demonstrate that murine and human BM harbors a population of tissue-resident memory CD8+ T (TRM ) cells. These cells develop against various pathogens, independently of BM infection or local antigen recognition. BM CD8+ TRM cells share a transcriptional program with resident lymphoid cells in other tissues; they are polyfunctional cytokine producers and dependent on IL-15, Blimp-1, and Hobit. CD8+ TRM cells reside in the BM parenchyma, but are in close contact with the circulation. Moreover, this pool of resident T cells is not size-restricted and expands upon peripheral antigenic re-challenge. This works extends the role of the BM in the maintenance of CD8+ T cell memory to include the preservation of an expandable reservoir of functional, non-recirculating memory CD8+ T cells, which develop in response to a large variety of peripheral antigens.

Introduction

Memory CD8+ T cells are classically divided into different subsets based on their circulatory patterns: central memory T cells (TCM) patrol secondary lymphoid organs, governed by expression of homing molecules CD62L and CCR7, whereas effector memory T cells (TEM) lack these and express other homing molecules, which enable them to recirculate through nonlymphoid organs 1. Both CD8+ TCM and TEM cells leave the tissues they patrol through blood or lymph by means of CCR7 and/or the sphingosine‐1‐phosphate receptor (S1PR1), and thereby return to the circulation 1, 2. In the past decade, a subset of memory T cells has been identified in both lymphoid and nonlymphoid organs that does not recirculate (reviewed in ref. [3, 4]). These tissue‐resident memory T (TRM) cells are characterized by the expression of CD69 and the lack of S1PR1 and CCR7, which inhibits their egress through blood or lymph and thereby their ability to recirculate 5. CD69 is a valid and functional marker for TRM cells, because of its ability to directly inhibit expression of S1PR1 6, though tissue residency has also been observed in CD69‐negative CD8+ T cells 7. CD8+ TRM cells are important for local tissue protection, as their reactivation leads them to proliferate and accelerate clearance of local re‐infections, independently of the circulatory pool 8, 9. Moreover, local activation of CD8+ TRM cells can also provide protection against unrelated pathogens, as their production of IFN‐γ elicits a pro‐inflammatory gene expression profile in surrounding tissue cells, which boosts innate immunity and induces the recruitment of circulating CD8+ T cells 10, 11.

In recent years, CD8+ TRM cells have been found in virtually every tissue. They were first identified in tissues exposed to external pathogens (skin, lungs, intestines, and female reproductive tract), then in internal organs (liver, kidney, and brain), and also in lymphoid organs (lymph nodes, spleen, and thymus; reviewed in ref. [3]). When comparing resident lymphoid cells from different tissues, they were found to share a core transcriptional expression program, which keeps the cells from recirculating and thereby enables them to mediate local immune‐surveillance 12, 13. We found that the transcription factors Blimp‐1 and Hobit play an important role in establishing this expression program and thereby in the differentiation of CD8+ TRM cells 13.

The BM has been considered an important reservoir for long‐term T cell memory 14, 15, 16. In both human and mice, there is an increased frequency of memory T cells compared to secondary lymphoid organs and blood 16, 17, 18. Moreover, when taking its total volume into account, BM contains one of the largest pools of memory CD8+ T cells in the body; we recently showed that the total number of virus‐specific CD8+ memory T cells in the BM is five to six times higher than in the spleen 19. A preference of memory CD8+ T cells for the BM may be driven by the presence of IL‐7‐ and IL‐15‐expressing stromal cells, which can contribute to their survival and long‐term maintenance 14, 17, 20. But despite its well‐established role as a memory organ, the BM has not been included in numerous studies that addressed the presence and function of CD8+ TRM cells 3, 7, 13, 21, 22.

Although CD69 is expressed on a subset of BM CD8+ T cells 14, 16, 17, 23, 24, 25, the total BM memory pool does not appear to be recently activated and CD69+ CD8+ T cells show some features of residency, such as low S1PR1 expression 17. In fact, the migratory behavior of BM memory CD8+ T cells is still under debate: it has been suggested that the majority of memory CD8+ T cells in the BM are recirculating 26, whereas it has also been proposed that a substantial subset of these cells, even the majority, is resident to the BM 16, 17, 19, 27.

To thoroughly investigate whether the BM contains CD8+ TRM cells and what their requirements for BM residency would be, we used a variety of different infection, immunization, and adoptive transfer models, including parabiosis experiments. Based on our experiments, we can conclude that the BM harbors a subset of bona fide CD8+ TRM cells, next to a pool of conventional circulating memory CD8+ T cells. We show that CD69+ CD8+ T cells in the BM express the transcriptional core program of resident lymphoid cells, on top of a BM‐specific gene signature, and we reveal both intrinsic and extrinsic factors required for the generation and/or maintenance of CD8+ BM TRM cells. These findings demonstrate that the important role of the BM in maintenance of immunological memory is not only governed by supporting recirculating memory T cells, but also by providing residency to a substantial pool of noncirculating, fully functional memory CD8+ T cells that is directed against a large variety of systemic and peripheral antigens.

Results

Memory CD8+ T cells with a resident phenotype are present in BM after acute systemic infections

To examine whether the BM develops a pool of resident memory CD8+ T cells in response to a systemic viral infection, we infected mice with the Armstrong strain of lymphocytic choriomeningitis virus (LCMV), which is efficiently cleared within 2 weeks. After >150 days, memory CD8+ T cells directed against different immunodominant viral epitopes could still be detected in the BM, with similar frequencies as in the spleen (Supporting Information Fig. 1A and Fig. 1A). Yet, we observed that BM from LCMV‐infected mice was significantly enriched for virus‐specific CD69+ CD62L− CD8+ T cells compared to the spleen, whereas the fractions of virus‐specific CD62L+ CD8+ T cells (i.e., TCM cells) or CD69− CD62L− (i.e., TEM) CD8+ T cells were comparable in both organs (Fig. 1B and C and Supporting Information Fig. 2A and B). Similar results were obtained after systemic acute infection with OVA‐expressing Listeria monocytogenes (Listeria‐OVA, Fig. 1D–F). We observed that both total and LCMV‐specific CD69+ CD62L− CD8+ T cells in the BM expressed significantly higher levels of CXCR3, CXCR6, and lower levels of CX3CR1 (Fig. 1G–I and Supporting Information Fig. 2C–G), which corresponds to the phenotype of CD8+ TRM cells in other murine tissues 28, 29. These results suggest that systemic infections with an intracellular pathogen induce the development of specific memory CD8+ T cells in the BM with a tissue‐resident phenotype.

Figure 1

Memory CD8+ T cells with a resident phenotype are present in murine BM after acute systemic infection. (A–C) Spleen and BM were analyzed for virus‐specific CD8 T cells, 172 days after systemic infection with LCMV Armstrong. Data are shown for one representative experiment with n = 4 mice, out of three independent experiments; (A) Frequency of LCMV GP33‐41‐specific and NP396‐404‐specific CD8+ T cells of all CD44+ CD8+ T cells (average + SD); (B) Representative FACS staining for CD69 and CD62L on LCMV GP33‐41‐specific CD44+ CD8+ T cells from spleen and BM; (C) Paired analysis for the percentage of CD62L− CD69+ cells of all LCMV GP33‐41‐specific or NP396‐404‐specific memory CD8+ T cells in spleen versus BM. (D–F) Analysis of OVA‐specific CD8 T cells in spleen and BM of WT mice, 49 days after systemic infection with Listeria‐OVA. Data are shown for one representative experiment with n = 9 mice, out of two independent experiments; (D) Frequency of OVA257‐264‐specific CD8+ T cells of all CD44+ CD8+ T cells (average + SD); (E) Representative FACS staining for CD69 and CD62L on OVA257‐264‐specific CD44+CD8+ T cells from spleen and BM; (F) Paired analysis for the percentage of CD62L+ CD69+ cells of all OVA257‐264‐specific memory CD8+ T cells in spleen vs BM. (G–I) Expression levels of (G) CXCR3, (H) CXCR6, and (I) CX3CR1 in Db‐GP33‐41‐specific memory CD8+ T cell subsets in BM, 60 days after systemic infection with LCMV Armstrong. Data are shown for one representative experiment with n = 10 mice, out of two independent experiments. Data were analyzed by two‐tailed t‐test, with matching (D, F), one‐way ANOVA followed by Tukey's multiple comparisons test (G–I) and two‐way ANOVA, with matching, followed by Bonferroni's multiple comparison test (C). Significance is indicated by *p < 0.05; **p < 0.01; and ****p<0.0001.

Generation of CD8+ TRM cells does not require local infection or antigen presentation

In nonlymphoid tissues, TRM cell development and maintenance is strongly enhanced by the local recognition of antigen and/or inflammatory mediators 30, 31. To test whether development of pathogen‐specific, resident‐like BM CD8+ T cells is dependent on infection of the BM itself, we infected mice with influenza virus, which is restricted to the airways and does not infect BM 32. Interestingly, in the memory phase (>30 days post infection) we found a significant enrichment of CD69+ influenza‐specific CD8+ TEM cells in BM compared to the spleen (Fig. 2A). Similar observations were made in a local skin‐infection model using HSV‐1: we found that adoptive transfer of naïve CD45.1+ gBT‐I CD8+ T cells (which are transgenic T cells specific for an immunodominant H2‐Kb restricted epitope of HSV‐1) to CD45.2+ recipient mice, followed by skin infection with HSV‐1, led to a significant enrichment of CD69+ gBT‐I cells in the BM compared to spleen in the memory phase of infection (Supporting Information Fig. 1B and Fig. 2B).

Figure 2

Generation of BM CD8+ TRM cells does not require local infection or antigen presentation. (A) Percentage of CD69+ CD62L− cells within Influenza NP366‐374‐specific CD44+ CD8+ T cells in spleen and BM 33 days after intranasal infection with Influenza A/HKx31. Data are shown from one representative experiment with n = 5 mice, out of two independent experiments; (B) Naïve WT mice were epicutaneously infected with HSV‐1 one day after i.v. transfer of 5 × 104 gBT‐I CD8+ T cells, which recognize the HSV‐1 Kb‐gB498‐505 epitope and are identified as Vα2+ Thy1.1+. Expression of CD69 was analyzed on donor cells in spleen and BM 40 days after infection; (C) Percentage of CD69+ within gBT‐I CD8+ T cells in spleen and BM, 11 months after in vitro activation and transfer; for (B) and (C), experiments were performed twice. (D) Frequency of gBT‐I cells within total CD8α + cells in the spleen (left) or CD69+ within gBT‐I cells in the BM (right) of conjoined mice that received gBT‐I and were subsequently infected with HSV, as shown in (B); (E) Frequency of gBT‐I cells within total CD8α + cells in the spleen (left) or CD69+ within gBT‐I cells in the BM (right) of conjoined mice that received in vitro activated gBT‐I cells, as shown in (C). For (D) and (E), organs were analyzed 3 weeks after conjoining. For (D) and (E), data are shown from one experiment with n = 3 mice. Data were analyzed by two‐tailed t‐test, with matching. Significance is indicated by *p < 0.05 and **p < 0.01.

Although the development of pathogen‐specific, resident‐like BM CD8+ T cells is independent of local infection, it is possible that pathogen‐derived antigens are transported to the BM following infection, where they could be presented to CD8+ T cells by local DCs 33. To determine whether local antigen‐recognition by CD8+ T cells in the BM is required for resident‐like memory CD8+ T cell development, we transferred in vitro activated gBT‐I CD8+ T cells into naïve recipient mice. Eleven months after transfer, a stable pool of memory T cells had formed in both spleen and BM, and also in this setting, we observed that the BM was enriched for CD69+ gBT‐I T cells compared to the spleen (Supporting Information Fig. 1C and Fig. 2C). As the cognate antigen for gBT‐I T cells is not present in these mice, these data demonstrate that local antigen recognition is not required for the development of CD8+ TRM‐like cells in the BM.

Finally, to formally demonstrate the permanent residence of these CD69+ memory CD8+ T cells in the BM, we performed parabiosis experiments to induce a shared blood circulation 34. For this purpose, naïve mice (naïve parabionts) were surgically conjoined with mice that had, as shown in Fig. 2B, received naïve gBT‐I CD8+ T cells and were subsequently skin infected with HSV‐1 35 days before surgery (infected parabionts). Three weeks after the conjoining surgery, we found equal numbers of gBT‐I memory T cells in spleens of both groups of mice, indicating that their blood circulation was in equilibrium (Fig. 2D, left). However, CD69+ gBT‐I memory cells remained significantly enriched in the BM of the infected parabionts, compared to their conjoined naive parabiont (Fig. 2D, right). Moreover, when we performed parabiosis with mice that received in vitro activated gBT‐I T cells (as per Fig. 2C) and nontransferred control animals, we observed a similar effect (Fig. 2E). Given that circulatory memory equilibrates in the BM as soon as 2 weeks after conjoining mice 35, we conclude that the BM also harbors a population of bona fide pathogen‐specific, tissue‐resident memory CD8+ T cells. This development is not restricted to a particular route of infection, nor dependent on local antigen‐presentation.

Virus‐specific resident‐like CD8+ T cells are found within human BM

Following previous observations that human BM also contains CD69+ CD8+ T cells with a resting memory phenotype 14, 16, 25, we examined the phenotype of these cells and addressed whether they also contain virus‐specific cells. We found that, unlike CD8+ TRM cells from epithelial barrier tissues, BM CD69+ CD8+ T cells were mostly CD103− (Supporting Information Fig. 1D and Fig. 3A), which is also the case in mice (data not shown). Human BM CD69+ CD8+ T cells also expressed significantly higher levels of CXCR6 and CCR5 and lower levels of CX3CR1, compared to the CD69− CD8+ T cells (Fig. 3B and C), which is in concordance with the CD8+ TRM cell phenotype in other human tissues 12, and murine BM (Fig. 1G–I and Supporting Information Fig. 2C–G).

Figure 3

Virus‐specific resident‐like CD8+ T cells are found within human BM. (A) Representative staining for CD69 and CD103 on CD8+ CD3+ T cells in human BM; (B and C) Expression of chemokine receptors CXCR6, CCR5, and CX3CR1 on CD69− and CD69+ CD8+ T cells in human BM; (B) Representative histograms for CD69− and CD69+ CD8+ T cells; (C) Expression levels depicted as geoMFI for the same populations. For (A–C), data are shown for n = 4 non‐HLA typed BM samples from four different donors; (D) Combinatorial coding analysis of human CD8+ BM T cells specific for CMV or EBV. Dually labelled MHC‐I tetramers loaded with the same immunodominant peptide from either CMV‐VTE, CMV‐TPR, or EBV‐RAK identify virus‐specific CD8+ T cells from BM of HLA‐typed donor #1 (percentages of specific cells is indicated in the gate); (E–G) Quantification of CMV‐ or EBV‐specific CD8 T cells and the percentage of CD69+ cells therein, in human BM. Data are shown for n = 5 HLA‐typed BM samples from five different donors; (E) Table describes detailed information on each specific T cell population detected; (F) The percentage of Tet+ cells detected within total CD8+ cells. For (F) and (G), results are shown as average ± SD. In panel (G), only those samples are indicated for which >100 CD69+ tetramer+ CD8+ T cells were acquired. Data was analyzed by one‐way ANOVA followed by Tukey's multiple comparisons test (G) and two‐way ANOVA, with matching, followed by Bonferroni's multiple comparison test (C). Significance is indicated by *p < 0.05 and **p < 0.01.

As human BM has a higher frequency of CD4+ and CD8+ T cells against certain viral epitopes 16, 36, we examined the presence of virus‐specific memory CD8+ T cells with a resident phenotype. Therefore, we probed human HLA‐typed BM samples for the presence of human cytomegalovirus (CMV)‐ and EBV‐specific CD8+ T cells. To ensure that the measurements would be sufficiently sensitive to detect low frequencies of antigen‐specific T cells, we used a combinatorial coding approach with MHC‐I tetramers 37, loaded with immunodominant epitopes from either CMV or EBV (Table 1). Using this method, we were able to identify a total of 15 virus‐specific CD8+ T cell populations in five BM samples (Supporting Information Fig. 1E and Fig. 3D and E), with an average of 1.10% CMV‐specific CD8+ T cells and 0.49% EBV‐specific CD8+ T cells of total CD8+ T cells (Fig. 3F). CD8+ T cells specific for either virus contained a subset of CD69+ cells (Fig. 3G). These results, therefore, support the hypothesis that human BM also contains virus‐specific CD8+ TRM cells.

Table 1

Tetrameric complexes used to detected virus‐specific human CD8+ T cells by combinatorial coding

Tetramer specificity	HLA‐restriction	Virus	Protein	Sequence	Amino acid positions
A1‐CMV‐VTE	HLA‐A*0101	HCMV	pp50	VTEHDTLLY	245–253
A2‐CMV‐NLV	HLA‐A*0201	HCMV	pp65	NLVPMVATV	495–504
B7‐CMV‐TPR	HLA‐B*0702	HCMV	pp65	TPRVTGGGAM	417–426
A2‐EBV‐GLC	HLA‐A*0201	EBV	BMLF‐1	GLCTLVAML	259–267
B7‐EBV‐RPP	HLA‐B*0702	EBV	EBNA‐3A	RPPIFIRRL	247–255
B8‐EBV‐RAK	HLA‐B*0802	EBV	BZLF‐1	RAKFKQLL	190–197

Steady‐state BM is enriched for CD69+ memory CD8 T cells that share strong homology with TRM cells in other organs

Given that BM CD8+ TRM cells develop against a large variety of both systemic and peripheral antigens, we next focused on the broad pool of memory CD8+ T cells in the BM of steady‐state mice. We found that in naïve animals, the BM contains significantly more CD8+ T cells expressing CD44 as compared to those in the spleen or peripheral blood (Fig. 4A), and we found that a substantial fraction of these CD8+ CD44+ T cells expressed CD69 (Fig. 4B). The expression of CD69 is predominantly found on CD8+ T cells with a TEM, rather than a TCM phenotype in the BM, whereas the corresponding memory subsets in spleen and peripheral blood lack CD69 expression (Fig. 4C). Further analysis of the different memory populations within total CD44+ pool of CD8+ T cells revealed that steady‐state BM is significantly enriched for CD69+ CD62L− and CD69+ CD62L+ cells compared to the spleen (Fig. 4D). These results are in agreement with our findings for antigen‐specific memory CD8+ T cells (Figs. 1 and 2). RT‐qPCR analysis revealed that CD69+ CD62L− memory (CD44+) CD8+ T cells in the BM express low levels of S1pr1, Klf2, and Ccr7, and high levels of Zfp683 (encoding Hobit), which is consistent with their TRM phenotype (Fig. 4E–H). To thoroughly investigate the residency profile of BM TRM cells, we performed genome‐wide transcriptional profiling on TRM‐like CD8+ T cells and compared this to TNV, TCM, and TEM CD8+ T cells from steady‐state BM. This revealed that only the expression profile from TRM‐like BM cells fully mirrors the gene expression program that is associated with tissue residency in various other tissue sites and across immune lineages 13; 22 of the 30 genes from this program are significantly different from BM TEM and/or TCM (Fig. 4I and Table 2).

Figure 4

Murine steady‐state BM lodges CD8+ T cell memory‐like cells with a resident phenotype. (A–D) Analysis of the CD8+ T cell memory‐like compartment in adult mice under SPF breeding. Data are shown for 1 experiment with n = 5 mice; (A) Frequency of memory CD44+ cells within CD8αβ + TCRβ + T cells in peripheral blood (PB), spleen, and BM; (B) Frequency of CD69+ cells of all CD44+ CD8αβ + TCRβ + T cells; (C) Representative FACS plots showing expression of CD62L and CD69 on memory CD8 T cells (defined as CD44+ CD8αβ + TCRβ + cells) in the different organs; (D) Frequency of memory CD8+ T cells subsets in spleen and BM, as gated for in panel C; (E–H) Relative mRNA expression of (E) Ccr7, (F) Klf2, (G) S1pr1, and (H) Hobit relative to the housekeeping gene Cyclophilin A in sort‐purified memory CD8+ T cell subsets from BM, as defined in (C) and (D). Data is shown for four independent sorting experiments, each with pooled BM from n = 4 mice; (I) Heatmap of RNA sequencing data for the 30 genes that compose the universal transcriptional signature of lymphocyte tissue residency 13 in CD8+ TNV (CD44− CD62L+), TCM (CD44+ CD62L+), TEM (CD44+ CD62L− CD69−), and TRM (CD44+ CD62L− CD69+) cells that were sort‐purified from BM of naïve mice. Data are shown from three independent sorting experiments, each with pooled BM from n = 6 mice. Relative expression levels (Z‐scores) of genes are shown, color coded according to legend. Data are shown as mean ± SD (A and B) or mean ± SEM (E–H). Statistical analysis was performed with one‐way ANOVA followed by Tukey's multiple comparisons test (A and B) and two‐way ANOVA followed by Bonferroni's multiple comparison test (D–H). Significance is indicated by *p < 0.05; **p < 0.01; ***p < 0.001; and ****p < 0.0001.

Table 2

Gene enrichment of tissue residency signature in BM TRM‐like cells

#	Gene symbol	FDR‐corrected p‐value		Average RPKM
		TRM versus TEM	TRM versus TCM	TNV	TCM	TEM	TRM
1	Rasa3	ns	3.3 × 10–03	115.0	100.3	91.8	44.1
2	Klf2	9.0 × 10–26	7.0 × 10–41	173.3	123.6	77.9	9.4
3	Sbk1	9.8 × 10–04	4.7 × 10–03	32.7	23.1	26.0	10.4
4	S1pr1	1.2 × 10–09	3.9 × 10–18	160.1	114,4	64.1	9.7
5	Cdc25b	9.2 × 10–08	1.5 × 10–03	23.2	24.0	33.7	10.3
6	AB124611	ns	5.0 × 10–03	58.8	54.4	39.8	20.2
7	Pik3r5	ns	ns	79.0	78.7	65.7	48.3
8	Sidt1	4.1 × 10–04	4.2 × 10–19	87.0	110.2	54.5	20.9
9	Atp1b3	2.4 × 10–02	8.0 × 10–09	247.2	192.0	132.5	64.0
10	Fam65b	ns	ns	66.2	51.5	44.3	27.1
11	Arhgef18	ns	9.5 × 10–03	87.8	70.0	56.3	34.5
12	Rasgrp2	2.7 × 10–03	1.6 × 10–08	66.0	41.2	30.8	13.1
13	Gramd4	1.0 × 10–04	2.2 × 10–26	112.5	84.5	37.9	14.8
14	Cxcr4	ns	ns	64.4	75.9	75.0	56.1
15	Klhl6	ns	6.4 × 10–06	90.0	93.4	47.3	29.7
16	Gm11346	ns	1.3 × 10–05	38.4	20.8	11.9	6.2
17	Tcf7	ns	3.3 × 10–03	562.4	459.9	265.9	195.8
18	A430078G23Rik	ns	1.5 × 10–02	73.7	51.2	31.7	22.6
19	Bcl9l	ns	3.7 × 10–03	80.7	50.6	34.3	23.8
20	S1pr4	ns	ns	100.8	61.6	58.1	41.7
21	S1pr5	7.9 × 10–03	ns	0.1	2.1	19.7	1.1
22	Pogk	ns	1.9 × 10–04	2.7	18.1	14.0	7.2
23	Inpp4b	ns	ns	41.6	21.5	19.9	33.0
24	Xcl1	ns	ns	0.7	56.8	69.6	158.2
25	Abi3	ns	ns	1.0	1.9	2.0	2.4
26	Dusp6	ns	4.5 × 10–11	4.3	18.1	38.6	79.5
27	Osgin1	ns	ns	0.3	0.2	1.2	1.7
28	Glrx	ns	2.9 × 10–02	9.2	7.6	16.4	20.9
29	Cxcr6	ns	3.0 × 10–20	39.6	61.8	185.7	360.2
30	Zfp683	2.8 × 10–03	2.8 × 10–34	2.7	3.7	45.8	164.9

Steady‐state BM TRM cells are efficient, polyfunctional cytokine producers

The BM has been proposed to act as a “reservoir” for memory T cells 15, 25, 36. We showed thus far that the BM contains both a circulating (TCM and TEM cells) and a resident memory CD8+ T cell compartment (TRM cells). Next, we used our gene expression data to determine to what extent these BM T cell subsets are transcriptionally different from each other. We found the highest similarity between TRM and TEM cells (only 93 differentially expressed (DE) genes, FC > 1.5, RPKM > 8, p < 0.05), whereas TRM cells were much more distinct from TCM and TNV cells (TRM vs TCM cells: 493 DE genes; TRM vs TNV cells: 927 DE genes) (Fig. 5A). From the 93 and 493 genes that are differentially expressed between TRM versus TEM and TRM versus TCM, respectively, we found 39 genes that are significantly different between resident and circulating memory CD8+ T cells in the BM (Fig. 5B and Supporting Information Table 1). From these 39 genes, nine belong to the aforementioned tissue residency signature that we recently described 13, such as Klf2, S1pr1, and Zfp683 (Fig. 5C). Additionally, BM CD8+ TRM cells express higher levels of Itga1 (encoding CD49a) and Cd101, both also described in CD8+ TRM cells from murine skin and human lung 9, 12, as well as other immune‐related genes, such as Tnfsf10 (encoding TRAIL) and Icosl (Fig. 5C).

Figure 5

Specific features of BM CD8+ TRM cells. (A–D) Analysis of RNA sequencing data obtained for CD8+ TNV (CD44− CD62L+), TCM (CD44+ CD62L+), TEM (CD44+ CD62L− CD69−), and TRM (CD44+ CD62L− CD69+) cells that were sort‐purified from BM of naïve mice. Data are shown from three independent sorting experiments, each with pooled BM from n = 6 mice; (A) Number of differentially expressed (DE) genes (downregulated in white, upregulated in black) between TRM versus TEM, TRM versus TCM, and TRM versus TNV. DE expressed genes were specifically up‐ or downregulated in BM TRM in comparison to their circulating counterparts, as determined by RNA sequencing (fold change [FC] > 1.5; and reads per kilobase per million mapped reads (RPKM) > 8]; (B) Venn diagram showing the overlap between genes DE by TRM and TEM, TRM, and TCM cells; (C) Heatmap displays relative amounts of transcripts of the 39 DE genes that were shared in the comparison shown in (B). Relative expression levels (Z‐scores) of genes are shown, color coded according to legend; (D) Correlation of gene ontology (GO)‐terms enriched in the sum of DE expressed genes between TRM versus TEM and TRM versus TCM (547 DE genes, FC > 1.5; RPKM = 8). GO‐terms were filtered to contain between 20 and 100 genes. Enrichment analysis revealed enrichment of the DE genes in 104 GO‐terms, which clustered into nine groups by comparison of their overlap index. Results are shown as a heatmap of the overlap indices, with the clustering and a color representation of each group on the left and a representative group name on the right; (E and F) Transcript levels of IFN‐γ in TCM, TEM, and TRM sorted from steady‐state mice, analyzed via RNAseq (E, obtained as in A–D) or RT‐qPCR (F, data are from four different sorting experiments, each with pooled BM from n = 4 mice); (G–I) Production of IFN‐γ by TCM, TEM, or TRM CD8+ T cells from LCMV‐infected mice. BM cells were incubated for 5 h with Brefeldin A in the absence (no stim.) or presence of GP33‐41 peptide and IFN‐γ production was evaluated by ICS in the different memory populations. Data are shown for one representative experiment with n = 4 mice, out of two independent experiments; (G) Representative plots and quantification of the production of IFN‐γ in the absence (H) or presence of LCMV peptide (I); (J and K) Polyfunctionality of TEM and TRM cells was analyzed as in panels (G–I). Data are from one representative experiment with n = 5 mice, out of two independent experiments; (J) Co‐production of IFN‐γ and CCL3 by TEM or TRM CD8+ T cells from LCMV‐infected mice. Representative plots are shown; (K and L) Representative plots (K) and quantification (L) of TEM and TRM capable of simultaneously producing two, three, or four out of four cytokines (IFN‐γ, CCL3, IL‐2, and TNF‐α). Results in (H), (I), and (L) show mean ± SD. Statistical analysis was performed by one‐way ANOVA followed by Tukey's multiple comparisons test. Significance is indicated by **p < 0.01 and ***p < 0.001.

Gene ontology analysis on the DE genes between resident and circulating T cells (TRM vs TEM = 93 DE genes + TRM vs TCM = 493 DE genes, Total = 586 DE genes, RPKM > 8, FC > 1.5) yielded 104 Gene Ontology (GO) terms, which clustered into nine groups (Fig. 5D). According to their GO‐term content (Supporting Information Table 2), BM CD8+ TRM cells differed from their circulatory counterparts in categories related to T cell activation and differentiation, cytotoxicity, protein synthesis, leukocyte migration, and response to IFN‐γ. Interestingly, the first cluster, with 23 GO‐terms, is associated with the “Regulation of cytokine production and proliferation.” Because the most significant GO‐term within this cluster is “Positive regulation of IFN‐γ production” (Supporting Information Table 2), we determined the expression levels of IFN‐γ in the different memory populations. We detected high levels of IFN‐γ mRNA in TEM and TRM cells, both in the RNAseq dataset (Fig. 5E) and by RT‐qPCR (Fig. 5F). To study the regulation of IFN‐γ production, we analyzed IFN‐γ protein expression by flow cytometry in LCMV‐specific memory CD8+ T cells from BM without or with GP33‐41 peptide stimulation (Fig. 5G). Importantly, the expression of CD69 and CD62L did not change during the stimulation due to the addition of Brefeldin A, which allows reliable identification of the different T cell subsets following short antigenic stimulation (Fig. 5G and Supporting Information Fig. 3A). Interestingly, we found that BM CD8+ TRM cells from LCMV‐immune mice already produced IFN‐γ without antigen stimulation, which was not observed in TCM or TEM cells (Fig. 5H). Upon the addition of cognate antigen, IFN‐γ production rapidly increased in both TRM and TEM, but not TCM cells (Fig. 5I). The percentage of IFN‐γ‐producing T cells upon activation was slightly lower for TRM compared to TEM cells (Fig. 5I), even though the frequency of Db GP33‐41‐specific cells was equal between these two groups (Supporting Information Fig. 3B). Db NP396‐404‐specific cells were more abundant within the TEM than in the TRM compartment and this was also seen for IFN‐γ‐producing T cells upon activation with the cognate peptide (Supporting Information Fig. 3C and D). In both cases, the amount of IFN‐γ per cell was lower in TRM than in TEM (Supporting Information Fig. 3E and F). We also observed that all IFN‐γ+ T cells co‐expressed CCL3 (Fig. 5J), a pro‐inflammatory chemokine that can attract many different leukocytes through CCR1, CCR4, or CCR5, and which is expressed in CD8+ TRM cells (Supporting Information Table 2 and ref. [11]). Like their circulating counterparts, IFN‐γ‐producing CD8+ TRM cells in the BM are highly polyfunctional, as they can coproduce significant amounts of IL‐2 and/or TNF‐α (Fig. 5K and L).

Gene expression analysis also indicated a distinct “Cytotoxity” cluster for BM TRM cells (Fig. 5D and Supporting Information Fig. 4A). As one of the genes in this cluster is Granzyme B (Gzmb), we analyzed protein levels by flow cytometry in LCMV‐specific T cells; but despite higher mRNA levels, Granzyme B protein was hardly expressed in TRM and TEM cells, in particular when compared to LCMV‐specific TEM cells in spleen and peripheral blood (Supporting Information Fig. 4B and C). Yet, upon stimulation with LCMV peptide, BM CD8+ TRM cells degranulated as indicated by expression of CD107a, demonstrating their cytotoxic ability (Supporting Information Fig. 4D and E). The percentage of CD107a+ TRM cells was similar as TEM cells in the BM, whereas the TRM fraction had even higher expression of CD107a (Supporting Information Fig. 4F). These data indicate that although BM TRM cells may not be loaded with cytotoxic granules in the steady state and capable of immediate cytotoxicity, they do have high transcript levels for genes involved in cytotoxicity and are able to degranulate upon activation, at least in a comparable fashion as BM TEM cells.

Overall, these results indicate that BM CD8+ TRM cells share important functional features with circulating CD8+ TEM cells, such as rapid responsiveness to cognate antigen with poly‐functional cytokine production and cytotoxic potential, but they also have a gene expression pattern that is clearly distinctive from their circulating counterparts.

BM TRM cells reside in the parenchyma in close contact with the circulation

We next investigated how CD8+ TRM cells in BM related to CD8+ TRM cells in other tissues. Therefore, we compared their transcriptional profile with our previously published dataset on CD8+ TRM cells in liver, skin, and small intestines 13. To compare both datasets, the gene expression in each subset was related to naïve splenic CD8+ T cells, as this population was a common denominator in both experimental designs. This analysis revealed that 136 of the 458 DE genes in BM CD8+ TRM cells (FC > 2, RPKM > 8, p < 0.05) are shared between all organs, whereas 82 are unique to the BM (Fig. 6A). Comparison of these 82 genes to the genes differentially expressed between resident and recirculating memory BM T cells, revealed that the expression of 28 genes is unique to BM TRM cells and is not expressed in the circulating memory CD8+ T cell subsets in the BM. They include genes such as Sulf2 (regulates binding of chemokines/cytokines to the ECM), Mmp9 (involved in tissue remodeling), Chst2 (involved in biosynthesis of chondroitin sulfate proteoglycans), Vav3 (involved in adhesion via β‐integrins), three transcriptional activators/repressors (Zbtb42, Maf, and Trerf1), and Icosl.

Figure 6

BM CD8+ TRM cells reside in the parenchyma in close contact with the circulation. (A and B) Comparison of DE genes between TRM from different organs. RNA sequencing data for BM TRM was acquired as in Fig. 4I, while data for liver, SI, and skin TRM were obtained from ref. [13]; (A) Venn diagram showing the overlap between DE genes by TRM from BM, Liver, SI, and skin, compared to splenic TNV (Sp) (FC > 1.5; RPKM = 8); (B) Correlation plot depicting the similarity between TRM in different organs. The correlation is pairwise, and was calculated using log2 RPKM values of the 136 DE genes that were significant in all four contrasts; (C–E) Intravascular staining to probe the localization of TRM within the BM. Mice were i.v. injected with 3 μg of an antibody against CD8a. Organs were obtained 2 min after injection and processed immediately. Data are shown for one representative experiment with n = 3 mice, out of two independent experiments; (C) A representative histogram comparing the staining with the anti‐CD8α antibody in CD8β + CD44+ cells from spleen, peripheral blood (PB), liver, and BM; geoMFI of CD8α staining in (D) CD8β + CD44+ cells from the different organs or (E) CD8β + memory subsets within the BM; (F and G) Immunofluorescence analysis of (F) CD8α + cells or (G) CD69+ CD8α + cells within steady‐state BM. Scale bar = 50 μm. Vasculature was visualized with antibodies against CD31 and CD144 and nuclei with Helix NP green staining. One representative region from a tile scan of 4 × 7 images is shown. The full tile scan is shown in Supporting Information Fig. 2; 40× magnification. Each CD8a + cell is encircled in white to identify these cells across the different stainings. CD69+ CD8α + cells are marked with an arrow.

A pairwise correlation of the 136 genes that were shared between CD8+ TRM cells in all four tissues revealed that BM TRM are more similar to TRM cells in the liver than in the small intestine (SI) or skin (Fig. 6B). A striking difference between these three organs is their localization of CD8+ TRM cells, as skin and SI CD8+ TRM cells are located within the epithelium and detect antigen coming from their local environment, while liver TRM cells are found inside the liver sinusoids and can respond to blood‐derived antigens 4, 28, 38. To assess the localization of BM TRM cells and their exposure to the blood stream, we examined their ability to rapidly bind an intravenously injected anti‐CD8α antibody 39. As expected, both blood and liver CD8β + CD44+ cells showed the highest degree of staining after intravascular injection of anti‐CD8α antibody, while the spleen had two populations, one protected from and one accessible to the anti‐CD8α antibody (Fig. 6C), which relates to their location in the white and red pulp, respectively 39. Interestingly, BM CD8β + CD44+ T cells showed some degree of intravascular staining (Fig. 6D), which was comparable between all three memory subsets (Fig. 6E). These results suggest a parenchymal, rather than sinusoidal localization, for all CD8+ memory T cell subsets in the BM, which was confirmed by immunofluorescent staining of CD8+ T cells on BM sections (Fig. 6F). Moreover, staining for CD69 further corroborated the parenchymal distribution of both CD69+ and CD69− CD8+ T cells (Fig. 6G and Supporting Information Fig. 5).

Maintenance of BM TRM cells depends on IL‐15, Blimp‐1, and Hobit

The BM is rich in IL‐15‐producing stromal cells 40 that provide a niche for long‐term maintenance of memory T cells 14, 41. As the dependence of TRM cells on IL‐15 varies per tissue 22, we investigated whether BM CD8+ TRM cells depend on IL‐15 for their long‐term maintenance. Therefore, we transferred in vitro‐activated gBT‐I CD8+ T cells into WT or IL‐15‐deficient animals (IL‐15−/−), and we analyzed the BM for total gBT‐I and CD69+ gBT‐I T cells after 30 days (Fig. 7A). We detected gBT‐I T cells in both WT and IL‐15−/− recipients (Fig. 7B), although we observed a significant decrease of CD69+ gBT‐I cells in the IL‐15−/− recipients (Fig. 7C). This finding indicates that CD8+ TRM cells in the BM are dependent on IL‐15 signaling.

Figure 7

Maintenance of BM CD8+ TRM cells depends on IL‐15 and Hobit. (A–C) In vitro‐activated gBT‐I cells were transferred into WT or IL‐15−/− mice and, after 30 days, the presence of total gBT‐I and CD69+ gBT‐I was analyzed in the BM. Results are shown as (A) representative plots, (B) quantification of the % of gBT‐I within total CD8α + cells and (C) quantification of TRM (CD69+) cells within gBT‐I cells in the BM, from one experiment with n = 5 WT and 4 IL‐15−/− mice. Results show mean ± SD; experiment was performed twice. (D and E) WT (Blimp+/+ Hobit+/+) or mice deficient for Hobit (Blimp+/+ Hobit−/−), Blimp1 (Blimp−/‐ Hobit+/+), or both (Blimp−/− Hobit−/−) were infected i.n. with Influenza A/HKx31. BM was harvested and analyzed 63 dpi. Results are shown from one representative experiment with n = 5 mice, out of three independent experiments, as (D) percentage of Db‐NP366‐374 + cells within total CD8αβ + T cells and (E) percentage of TRM (CD69+ CD62L−) cells within Db‐NP366‐374 + CD8αβ + T cells. Data were analyzed by two‐tailed t‐test (B and C) or one‐way ANOVA followed by Tukey's multiple comparisons test (D and E). Significance is indicated by *p < 0.05 and **p < 0.01.

It has been shown that CD8+ TRM cells in a variety of tissue sites are dependent on the expression of the transcription factors Blimp‐1 and Hobit 13. Therefore, we next analyzed how the genetic deletion of these transcription factors affected the development of CD8+ TRM cells in the BM. Following infection with influenza virus, we found that the percentage of total virus‐specific CD8+ T cells was not different in WT, Blimp1−/−, Hobit−/−, or Blimp1−/−xHobit−/− mice >60 days post‐intranasal infection (Fig. 7D). However, the percentage of CD69+ cells within the Db‐NP366‐374 + cells was significantly reduced in the Blimp1−/−xHobit−/− mice (Fig. 7E). Altogether, these results indicate that BM parenchyma hosts CD8+ TRM cells that depend on Blimp‐1 and Hobit for their generation and/or maintenance.

The pool of BM CD8+ TRM cells is expandable

Human CMV infection is associated with increased levels of IL‐15 mRNA in the BM, together with an accumulation of CD8+ CD45RA+ CD27− memory T cells, that are highly responsive to IL‐15 42. We questioned whether the inflationary accumulation of memory T cells seen during murine CMV infection (MCMV) 43 would also lead to inflation of the BM TRM pool. To address this, we analyzed the frequency of CD8+ T cells in the BM that were specific for acute (M45, M57) and latent inflationary (M38, m139) epitopes in MCMV >60 days after infection. CD69+ CD62L− CD44+ CD8αβ + T cells were detected for both acute and inflationary epitopes at this time point (Fig. 8A and B). However, whereas BM TEM cells were significantly enriched for inflationary epitopes, this was not the case for BM TRM cells, which showed a similar size for all tested epitopes (Fig. 8C). This would indicate that MCMV‐induced memory inflation favors CD8+ TEM rather than TRM cells, but it could also be that there is a restriction in the pool size or occupancy of the niches that maintain the CD8+ TRM cells in the BM. To investigate the latter, we transferred OVA‐specific OT‐I cells to naïve WT mice and challenged them either once or thrice with OVA in different ways (Fig. 8D). More than 30 days after the last challenge, we observed a significant increase in the percentage of OT‐I cells in mice that were challenged thrice, compared to mice that were challenged only once (Fig. 8E). This strong expansion was observed for both the TRM and TEM OT‐I cells (Fig. 8F), indicating that the BM TRM pool can expand upon antigen re‐encounter. Yet, it was possible that after three immunizations the CD8+ TRM niche was still not fully occupied. We reasoned that full occupancy of the niche might be achieved by consecutive infections with more complex pathogens. To address this, we challenged mice with Listeria‐OVA on day 0, after which the mice were infected on day 30 with LCMV Armstrong or not, followed by analysis on day 60 for CD8+ memory T cells in the BM specific for either pathogen (Fig. 8G and Supporting Information Fig. 6A and B). Importantly, these experiments revealed that the generation of new CD8+ TRM cells against LCMV did not affect the size of the preexisting pool of CD8+ TRM cells against Listeria‐OVA (Fig. 8H). In fact, the total population of CD8+ TRM cells in the BM expanded 2.6‐fold after the second infection (Fig. 8I). Comparable results were obtained when we performed the inverse experiment, i.e., first LCMV, followed by Listeria‐OVA (Supporting Information Fig. 6C–F). We, therefore, conclude that the pool of CD8+ TRM cells in the BM is readily expandable and can increase in both size and diversity when the host is exposed to new antigenic challenges.

Figure 8

The pool of BM CD8+ TRM cells is expandable. (A–C) Distribution of CD8αβ + memory populations specific for mCMV acute and inflationary epitopes analyzed 65 days p.i.; Representative plots are shown for (A) acute epitopes M45 and M57 and (B) inflationary epitopes M38 and m139. Tetramer stainings are shown in the left panels and the gating for TRM (CD62L− CD69+) and TEM (CD62L− CD69−) is shown in the right panels; (C) Percentage of TEM and TRM within live BM cells for each mCMV‐specific population. Data is shown for one representative experiment with n = 4 mice, out of two independent experiments; (D–F) Distribution of OT‐I memory populations analyzed more than 30 days after the last immunization in mice that received one or three homologous boosts to generate OVA‐specific memory. Data are shown for one experiment with n = 5 mice that received one challenge and 4 mice that received three challenges; (D) Schematic depiction of the experimental setup; (E) Percentage of OT‐I cells within total CD8α + cells in the BM; (F) Percentage of TCM, TEM, and TRM within OT‐I cells in the BM after one or three homologous challenges; (G–I) Distribution of OVA‐specific CD8αβ + memory populations analyzed more than 60 days after the first challenge in mice that received only Lm‐OVA or Lm‐OVA followed by LCMV Armstrong infection, more than 30 days after the first challenge. Data are shown for one experiment with n = 5 mice/group. Results from the reverse experiment are shown in Supporting Information Fig. 3; (G) Schematic depiction of the experimental setup; (H) Percentage of Kb‐OVA257‐264 + TRM cells of live BM cells; (I) Percentage of total TRM within live BM cells. Data were analyzed by two‐tailed t‐test (E, H, and I) and two‐way ANOVA, followed by Bonferroni's multiple comparison test (C and F). Significance is indicated by **p < 0.01 and ***p < 0.001.

Discussion

Here, we show that the BM harbors an expandable pool of bona fide resident memory CD8+ T cells that develop in response to peripheral antigen recognition. BM CD8+ TRM cells share transcriptional similarities with other tissue‐resident lymphoid cells found in a variety of organs, while also expressing a unique set of genes related to tissue remodeling and adhesion, which is likely induced by the BM environment. We demonstrate that BM CD8+ TRM cells are polyfunctional cytokine producers and depend on IL‐15, and the transcription factors Blimp‐1 and Hobit for their maintenance.

Both human 14, 16 and murine BM 17, 18, 19 contain a large number of memory T cells. In this work, we demonstrate that the BM, in contrast to the spleen, contains a unique subset of CD62L− CD69+ memory T cells, both in the steady‐state and after a variety of infectious challenges. Our results suggest that the enrichment in memory CD8+ T cells in BM compared to spleen can be largely attributed to the presence of resident cells. These results fit with previous observations that show that, compared to matched PB, the human BM is enriched for CD8+ T cells with a TEM (CD45RA− CCR7−) phenotype 14, 36, and that CD69+ cells are present among the TEM and TEMRA subsets (CD45RA+ CCR7−) in this organ. Additionally, human BM is enriched for CD8+ T cells expressing CXCR6 and CCR5, as well as CD8+ T cells specific for an epitope from EBV lytic protein BZLF‐1, but not for CD8+ T cells specific for epitopes from EBV latent protein EBNA3A or CMV pp65 36. We observed that CXCR6 and CCR5 expression is significantly higher in CD69+ than in CD69− CD8+ BM T cells and the percentage of CD69+ is higher in CD8+ T cells specific for the same EBV lytic epitope than in CMV‐specific CD8+ T cells. Altogether, these results suggest that both human and murine BM are enriched for CD8+ TRM cells and these comprise a true reservoir of systemic memory, in the sense that they are maintained in the organ at disequilibrium with the circulatory pool. The recirculating behavior of the CD69− memory CD8+ subsets present in the BM, i.e., TEM and TCM cells, remains to be addressed. It is possible that a part of these populations also resides in the BM, but our data—from the parabiosis experiments and RNAseq analysis—support the notion that most of them are in equilibrium with the circulatory pool.

Certain tissues, such as brain and lung, require local antigen recognition for CD8+ TRM cell development 44, 45, whereas other organs intrinsically support TRM cell lodgment and survival in the absence of antigen 21, 34, 46. Previous work on CD4+ T cells in human BM showed a preferential recruitment and/or maintenance of long‐term T‐cell memory for systemic pathogens (measles, rubella, and mumps) versus that for skin or mucosal pathogens (Candida albicans and Vaccinia virus) 16, suggesting that local antigen presentation might be required for BM TRM cells. Yet, we could demonstrate that both systemic (LCMV, Listeria) and peripheral (Influenza, HSV) infections led to the generation of antigen‐specific CD8+ TRM in the BM, as did transfer of in vitro activated TCR‐transgenic CD8+ T cells, demonstrating that local antigen recognition is not necessary for the establishment of CD8+ TRM cells in the BM. The difference with what has been suggested for human CD4+ BM T cells could be explained by several factors, such as differences between human versus mouse, or differential requirements for the development of CD4+ versus CD8+ TRM cells. The requirements for human CD8+ TRM generation are to date largely unexplored.

In several organs, CD8+ TRM cells are not in contact with the circulation, but rather localized in close association with the epithelium, since this is where they are most likely to re‐encounter their cognate antigen 39. The liver is an exception to this rule, as liver TRM cells are lining the sinusoids, which enables them to re‐encounter their antigen arriving via the bloodstream. Interestingly, we found for the BM that CD8+ TRM cells reside inside the parenchyma, but they are still in close connection to the vasculature, given their rapid intermediate level of staining with an intravenously injected anti‐CD8α antibody. This is also true for the other CD8+ T cell subsets in the BM, and thus a likely consequence of the high permeability of the BM sinusoids 47, rather than a specific feature of the CD8+ TRM cells. We recently confirmed the perivascular localization of memory CD8+ T cells in the BM, which is most likely driven by their CXCR4 expression and the CXCL12‐expressing stromal cells around the BM sinusoids 48. Both CD4+ and CD8+ T cell populations are evenly distributed throughout the BM parenchyma 17, and we also did not observe a different localization for CD69+ compared to CD69− CD8+ cells. Nevertheless, it is conceivable that the BM does have specialized niches for different CD8+ T cell subsets, as we found that CD8+ TRM cells in BM depend on IL‐15, whereas CD69− memory CD8+ T cells do not. We have previously shown that IL‐15 can induce expression of Hobit, which induces a subset of genes that prevents egress and increases tissue interactions 13. IL‐15‐expressing VCAM‐1+ stromal cells are scattered throughout the entire BM 40, making it possible that local IL‐15 signaling contributes to the generation of CD8+ TRM cells through the induction of Hobit. Yet, immunofluorescent imaging suggests that IL‐15‐expressing stromal cells are more abundant than CD8+ TRM cells in the BM; moreover, these stromal cells are likely to harbor also other memory CD8+ T cell subsets in the BM, as most memory T cells are in direct contact with IL‐7+ VCAM‐1+ stromal cells 17, and these largely overlap with IL‐15‐expressing cells 40. We, therefore, postulate that although IL‐15 triggering is required for the generation of CD8+ TRM cells, it is most likely not sufficient, and their development may depend also on other intrinsic and/or extrinsic signals.

Another important issue regarding the generation of TRM cells is the expandability of the pool, because of its implications in vaccine design and the development of protective tissue memory. Just as the total pool of circulating memory T cells can expand upon repetitive challenges 49, the pool of CD8+ TRM can dramatically grow in size in barrier tissues, like the skin 34, 50, where the size of the pool directly correlates with protection 50. Because the BM is a hematopoietically active compartment with a fixed cellularity and little space to expand due to the encompassing bone, we addressed whether the pool of BM CD8+ TRM cells could expand or if there would competition for niche signals. Remarkably, the number of BM CD8+ TRM cells specific for a single antigenic peptide could indeed expand upon multiple immunizations. In agreement with this, when mice were sequentially challenged with two different pathogens, BM CD8+ TRM cells against the first pathogen were not deleted to make space for newly formed TRM against the second challenge. Instead, the total pool of BM CD8+ TRM cells grew in size to accommodate the new specificities. These observations reinforce the concept of the BM as an expandable reservoir for long‐term memory. It remains to be shown whether this reservoir can contribute to the systemic pool upon re‐challenge, especially after long periods between generation and re‐encounter with the cognate antigen.

The implications of this resident, long‐term memory storage in the BM are intriguing. It could be that CD8+ TRM cells have a protective function in the BM, comparable to their role in liver, brain, and barrier tissues 28, 46, 51. Yet, in contrast to liver TRM 52, BM TRM cells did not contain preformed Granzyme B in their cytotoxic granules and are thus less likely to exert immediate cytotoxicity upon re‐challenge. Human BM CD8+ T cells with an effector memory phenotype, of which more than 70% expressed CD69, were also shown to contain much lower levels of Perforin and Granzyme B compared to paired samples of peripheral blood 25. Furthermore, BM TRM cells specific for Influenza or HSV are also not likely to be protective to the tissue, as these pathogens do not infect the BM; however, it is conceivable that these TRM cells do get activated when neutrophils bring antigens from the periphery to the BM, thereby driving local T cell activation 33. This may lead to proliferation of BM TRM cells, comparable to skin and female reproductive tract 9, 53, and the formation of effector T cells that leave the BM and provide protection in the infected organ. CD8+ TEM and TCM cells in the BM are known to be major contributors to the organ's memory reservoir function 15, 25, and these subsets outnumber the TRM cells in the BM. It is therefore interesting to speculate that TRM cells in the BM may also fulfill a different function. Memory CD8+ T cells support engraftment of hematopoietic stem and progenitor cells upon transplantation 54 and enhance the maintenance of hematopoietic stem cells (HSCs) 23. This could be important during viral infections, when HSCs are lost due to increased differentiation and impaired self‐renewal (reviewed by Pascutti et al. 55). Influx of memory CD8+ T cells in the BM following viral infection may serve to prevent or restore the HSC pool, and it is conceivable that CD8+ TRM cells also play a role in this process. This is corroborated by our observation that BM CD8+ TRM cells spontaneously produce IFN‐γ, given that the production of IFN‐γ during the steady state is beneficial for the pool of HSCs 56, 57. This implies a role for BM TRM cells during homeostasis, although they also rapidly upregulate IFN‐γ upon activation, by which they may contribute to skewing hematopoiesis towards the type of immune cell that is beneficial for eradicating the invading pathogen (reviewed in ref. [58]. It could thus well be that CD8+ TRM cells in the BM not only have an immune protective role, but also support the local hematopoietic process, though this requires further investigation.

In conclusion, the BM harbors a subset of memory CD8+ T cells that are not part of the recirculating pool, but that have taken up residency in the parenchyma of the BM. Given their broad range of specificities and ability to respond to and expand upon antigenic re‐challenge, we believe that these cells constitute an attractive target for protective vaccination strategies.

Materials and methods

Mice

C57BL/6J mice were purchased from Janvier Labs. For steady state analysis, mice were aged in the NKI animal facility and studied between the ages of 16 and 30 weeks. Zfp683 ‐/− (Hobit KO 59), Prdm1 flox/flox × Lck Cre (Blimp1 KO 60) and IL‐15−/− (IL‐15 KO 61) mice were maintained on a C57BL/6J background. Blimp1 KO mice were crossed onto Hobit KO mice to generate Blimp1 × Hobit double KO (Blimp1 × Hobit DKO) mice. gBT‐I mice are CD8+ TCR transgenic mice that recognize the H‐2Kb‐restricted HSV‐1 gB epitope of amino acids 498–505 (gB498‐505, SSIEFARL) 46. Mice were maintained under SPF conditions and animal experiments were performed according to national and institutional guidelines.

Infections, immunizations, and cell transfers

Mice were infected with either LCMV (intraperitoneal route [i.p.], 2 × 105 PFU Armstrong), Listeria‐OVA (intravenous route, 1 × 104 CFU or oral route, 2 × 109 CFU), Influenza A/HKx31 (x31, H3N2, intranasal route [i.n.], 10 × 50% tissue culture‐infective dose), Influenza A/HKx31‐OVA (x31‐OVA, H3N2, i.n., 1 × 104 PFU), Influenza A/PR8/34‐OVA (PR8, H1N1, i.p., 1.5 × 107 PFU), HSV type 1 KOS (epicutaneous application after scarification, 1 × 106 PFU), or MCMV‐Smith (intraperitoneal route, 1 × 104 PFU). For the HSV‐1 infections, naïve gBT‐I cells (5 × 104 cells) were injected intravenously (i.v.) prior to HSV infection. In vitro‐generated gBT‐I effector splenocytes were activated by peptide‐pulsed splenocytes as described previously 9, of which 5 × 106 cells were injected i.v. in either WT and IL‐15−/− mice. Multiple immunizations were performed by administering OVA‐peptide loaded BM‐derived DCs, followed by infection with PR8‐OVA and later Influenza x31‐OVA. BM‐derived DCs were generated by culture of BM cells for 7 days in the presence of 20 ng/mL GM‐CSF and IL‐4 to allow for DC differentiation. DC were then matured overnight in the presence of 150 ng/mL LPS and pulsed with OVA257‐264 peptide (SIINFEKL, 1 μg/mL, 45 min) prior to transfer. Then 2.5 × 105 DC were transferred i.v. into recipients. Mice were sacrificed at the indicated time points after infection or cell transfer and organs were harvested for analysis.

Murine tissue collection and preparation

Single cell preparations of spleen were obtained by passing organs over a 70 μm cell strainer (BD Biosciences) with MACS buffer (PBS + 1% FCS + 2 mM EDTA). Bones were harvested, cleaned, and crushed in MACS buffer using a mortar and pestle. BM cell suspensions were filtered through a 70 μm cell strainer to remove bone debris. Erythrocytes were lysed with lysis buffer (155 mM NH4Cl, 10 mM KHCO3, 127 mM EDTA). For intravascular stainings, 3 μg of PE‐labeled anti‐CD8α (53‐6.7) antibody was injected i.v. 2 min before sacrificing the mice. All organs were harvested and processed immediately.

Flow cytometry analysis of murine samples

Single cell suspensions were labeled with the indicated fluorescently conjugated antibodies at 4°C for 30 min in PBS containing 1% FCS. The antibodies used were purchased from eBioscience, BD Biosciences, or BioLegend: CD44 (IM7), CD62L (MEL‐14), CD69 (H1.2F3), CD8α (53‐6.7), CD8β (YTS156.7.7), and TCRβ (H57‐597). Virus‐ specific CD8+ T cells were examined with the following MHC class I tetramers: Db‐GP33‐41 (LCMV, KAVYNFATC) and Db‐NP396‐404 (LCMV, FQPQNGQFI), Kb‐OVA257‐264 (Listeria‐OVA, SIINFEKL), Db‐NP366‐374 (LCMV, ASNENMETM), Db‐M45985‐993 (MCMV, HGIRNASFI), Kb‐M57816‐824 (MCMV, SCLEFWQRV), Kb‐M38316‐323 (MCMV, SSPPMFRV), and Kb‐m139419–426 (MCMV, TVYGFCLL). MHC class I peptide tetramers for Kb‐OVA257‐264 and Db‐NP366‐374 were kind gifts from Dr. Derk Amsen (Sanquin, Amsterdam, The Netherlands). Tetramer labeling was performed at room temperature. Adoptively transferred gBT‐I cells were identified by their expression of the TCR α‐chain variable region 2 (Vα2 (B20.1)) and Thy1.1 (HIS51). For ICS, 1 × 106 splenocytes or 5 × 106 BM cells were cultured in the presence or absence of peptide (2 μg/mL) and brefeldin A for 5 h at 37°C. Staining was carried out with the BD Cytofix/Cytoperm kit, using antibodies against CCL3 (DNT3CC), IFN‐γ (XMG1.2), IL‐2 (JES6‐5H4), and TNF‐α (MP6‐XT22). Samples were collected by an LSR Fortessa (BD) and analyzed with FlowJo software (Tree Star). For analysis, cells were fixed in 0.5% paraformaldehyde (PFA). Dead cells were excluded with LIVE/DEAD® Fixable Near‐IR Dead Cell Stain Kit (Thermo Fisher Scientific), following manufacturer's instructions. Cells were acquired on an LSR Fortessa (BD Biosciences) flow cytometer and analyzed with FlowJo software (Tree Star, Inc.).

Parabiosis experiments

Parabiosis experiments were performed as described by Collins et al. 62. In short, 5 × 104 gBT‐I CD45.1+ cells were transferred i.v. into WT CD45.2+ mice and then the mice were infected with HSV. Alternatively, mice received in vitro activated gBT‐I cells. In the memory phase (HSV = 40 days; in vitro‐activated = 11 months), these mice were conjoined to non‐infected CD45.2+ control mice. Each mouse was anesthetized with ketamine (100 mg/kg) and xylazine (15 mg/kg). Skin was shaved and disinfected with alcohol and betadine pads. Matching incisions were made from the olecranon to the knee joint of each mouse and subcutaneous fascia was bluntly dissected to create 0.5 cm of free skin. The olecranon and knee joints were attached by a 5‐0 silk suture, and dorsal and ventral skins were attached by continuous staples or sutures. Betadine was used to disinfect the entire incision following surgery. Both mice of each pair were injected subcutaneously with PBS and buprenorphine (0.1 mg/kg) to treat pain. Three weeks after conjoining, the mice were sacrificed and the presence and phenotype of CD45.1+ gBT‐I donor cells in spleens and BM were analyzed in all mice.

Human BM analysis

BM samples were obtained from allo‐SCT donors after approval of the Leiden University Medical Center institutional review board and informed consent according to the Declaration of Helsinki. Informed consent form all participants involved in this study were written for samples obtained since 2003 and verbal for older samples when guidelines provided no written consent. Bone marrow mononuclear cells (BMMC) were isolated by Ficoll gradient centrifugation, and frozen in liquid nitrogen until further analysis. For T cell staining of approximately 1 × 106 BMMC a final concentration of 2 μg/mL per pMHC tetramer was added and incubated for 15 min at 37°C. Next, antibody‐mix was added and cells were incubated for 30 min at 4°C. Prior to flow cytometry, cells were washed twice. Dual‐encoding pMHC tetramer analysis was performed as previously described 37 and description of the tetramer complexes used is shown in Table 1. The following antibodies from Thermofisher, BD, or Biolegend were used: anti‐CD8 (3B5), anti‐CD3 (UCHT1), anti‐CD69 (FN50), anti‐CD103 (Ber‐ACT8), anti‐CXCR6 (K041E5), anti‐CX3CR1 (2A9‐1), and anti‐CCR5 (J418F1). Near‐IR fixable dye (Invitrogen) was used to exclude dead cells from the analysis. All samples were measured in PBS 0.5% FCS with an LSR Fortessa (BD) and the analysis was performed using FlowJo Version 10 software.

Immunofluorescence staining

Bones were embedded in Tissue‐Tek OCT compound (Sakura Finetek) and cooled with liquid nitrogen. Sections of 8 mm were made with a cryostat (Leica) using the CryoJane Tape‐Transfer System (Leica). Sections were fixed in acetone, air‐dried, and blocked with CAS‐Block (Thermo Fisher Scientific) for 15 min, and subsequently stained with antibodies in PBS/1% BSA. Primary antibodies were stained O/N at 4⁰C, followed by secondary antibodies for 1–2 h at room temperature. The following antibodies were used: CD8α‐Alexa Fluor 647 (53‐6.7, Biolegend), CD144/VE‐cadherin biotin (BV13, Biolegend), CD31/PECAM‐1 biotin (MEC 13.3, BD Pharmingen), CD69 (R&D systems), Donkey‐Anti‐Goat IgG Alexa Fluor 568 (Invitrogen), and Streptavidin eFluor 450 (eBioscience). Nuclei were stained using Helix NP green (eBioscience). Sections were viewed using a SP8 confocal (Leica). Images were further processed in LASX (Leica) and Fiji Imaging Software.

Quantitative PCR

RNA was extracted using TRIzol (Invitrogen) and complementary DNA was made with random hexamers and Superscript II reverse transcriptase (Roche). Quantitative real‐time PCR was performed in duplicate with Express SYBR Green reagents (Invitrogen) on the StepOnePlus RT‐PCR system (Applied Biosystems), and data were normalized using Cyclophilin A as reference gene. Primer sequences are available upon request.

RNA sequencing

For sorting, CD8+ T cells from spleen and BM were pooled and enriched with CD8α microbeads (Miltenyi Biotec) and MACS LS columns (Miltenyi Biotec) (6 mice/sort, three sorts in total). The following subsets of CD8αβ + T cells were sorted to purity on an FACSAria II or III (BD Biosciences): TRM (CD44+CD62L−CD69+), TEM (CD44+CD62L−CD69−), TCM (CD44+CD62L+), and TNV (CD44−CD62L+). RNA was isolated with the NucleoSpin RNA XS kit (Macherey‐Nagel). A DNAse treatment step was performed during the isolation. RNA quality check, polyA RNA selection and sequencing were performed at GenomeScan BV (Leiden, The Netherlands). RNA quality and quantity was determined by a Bioanalyzer Picochip (Agilent). Per sample, 25 ng RNA was used for library preparation and single‐end, 75 nt fragments were analyzed by Illumina NextSeq 500. At least 20 × 106 reads were analyzed and mapped to the mouse genome (UCSC, mm10) with TopHat (v2.1.0). Reads that aligned to (exonic parts of) genes annotated by Ensembl (release 78) were quantified using featureCounts (v1.4.3‐p1). Differential gene expression analysis was done with edgeR (version 3.14.0) in R (v3.3.1). Genes that had less than 1 CPM (counts per million mapped reads) in less than three samples were removed. A negative binomial generalized log‐linear model was fit to the data. Pair‐wise comparisons (contrasts) were made between the different cell populations, and glmTreat was used to find genes that had at least 1.5‐ (comparisons between BM subsets) or 2‐ (comparisons between TRM and TNV in different organs) fold change difference. Genes with FDR‐adjusted p‐values < 0.05 were considered significantly differentially expressed, if they also had at least eight RPKMs (reads per kilobase of transcript per million mapped reads) in one or both of the two cell types that were compared in the contrasts. Data are available from GEO under accession number GSE96839.

GO‐term enrichment analysis was done using the Wallenius method including gene length bias in goseq (v1.24.0), followed by gogadget (v2.0) 63 in R (v3.3.2). Enrichment was done using the genes that were differentially expressed in either the BM TRM versus BM TCM cells comparison, and/or the BM TRM versus BM TEM cells comparison, combined, restricting on only biological processes (BP). In gogagdet, enriched GO‐terms (FDR‐adjusted p‐value < 0.05) were first filtered so that only terms containing more than 20 but less than 100 genes were left, and were subsequently clustered using the overlap index, with Ward.D method and Manhattan distance.

To compare TRM cells from BM to TRM cells from other organs (liver, small intestine (SI), and skin), RNA‐Seq data from ref. [13] were downloaded (GEO accession: GSE70813), and were also mapped to the mouse genome similar as described above. In edgeR the same filtering criteria were used, and batch effect (both experiments) was taken into account in the model. Differentially expressed genes between BM TRM versus liver TRM, BM TRM versus SI TRM, and BM TRM versus skin TRM were again defined as genes that differed at least twofold change, and had at least eight RPKMs in one or both groups.

To find a BM TRM specific signature, gene expression of TRM cells per organ (BM, liver, SI, and skin) were compared with that of naïve T cells from the spleen. Here, differentially expressed genes were defined more stringent, analogous to ref. [13], FDR‐adjusted p‐values < 0.05, at least twofold change difference, and eight RPKMs in one or both groups. The differentially expressed genes uniquely found in the BM TRM versus spleen naïve T cells were further used for GO‐term enrichment analysis. In gogadget, all BP enriched GO‐terms were clustered using the overlap index, with Ward.D method and Euclidean distance.

Statistical analysis

Statistical analyses were performed with Prism (GraphPad Soſtware, Inc.) using t‐test, one‐way or two‐way ANOVA, depending on the experimental design. Significance is indicated by *p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001.

Author contributions

M.F.P., S.G., N.C., M.W., M.H.M.H., P.H., R.A., L.K.M., K.P.J.M.v.G., and M.A.N. designed the research. M.F.P., S.G., N.C., G.B., R.S., F.B., A.O., E.S., E.P., J.E.P., S.H., and P.H. performed the research. M.F.P., S.G., N.C., G.B., B.N., A.O., J.E.P., P.H., and M.A.N. analyzed the data. M.H.M.H. and R.A. provided essential biological material. M.F.P., S.G., M.W., and M.A.N. wrote the manuscript.

Conflict of interest

The authors declare no financial or commercial conflict of interest.

cytomegalovirus

gene ontology

lymphocytic choriomeningitis virus

OVA‐expressing Listeria monocytogenes

small intestine

sphingosine‐1‐phosphate receptor

central memory T cells

effector memory T cells

resident memory T cells

FiguresS1‐S6

Click here for additional data file.

TableS1

Click here for additional data file.

TableS2

Click here for additional data file.