BALB/c (haplotype K^dD^dL^d) and BALB/c-H-2^dm2 (haplotype K^dD^dØ (42)) mice were bred and housed under specified-pathogen-free conditions by the Translational Animal Research Center (TARC) at the University Medical Center of the Johannes Gutenberg-University Mainz, Germany.

Virus derived from BAC plasmid pSM3fr (43) was used as “wild-type” virus, mCMV-WT. BAC-derived recombinant viruses mCMV-Δm152 (44) and mCMV-m152.IE+E (28) have been described previously.

Syngeneic HCT with BALB/c mice as hematopoietic cell (HC) donors and recipients or allogeneic HCT with BALB/c mice as donors and BALB/c-H-2^dm2 mice as recipients were performed as described in greater detail previously (45). Briefly, hematoablative conditioning of 8 to 10-week-old female mice was achieved by total-body γ-irradiation with a single dose of 6.5 Gy. HCT was performed ~2 hours later by intravenous infusion of 5x10⁶ femoral and tibial donor bone marrow cells. At ~2 hours after HCT, intra-plantar infection of the recipients was performed with 1x10⁵ plaque-forming units (PFU) of the respective viruses.

To determine viral genome load in lung tissue, DNA of infected mice was isolated from the postcaval lobe with the DNeasy tissue kit (catalog no. 69504; QIAGEN, Hilden, Germany) according to the manufacturer’s instructions. Viral and cellular genomes were quantitated in absolute numbers by M55-specific and pthrp-specific qPCRs normalized to a log₁₀-titration of standard plasmid pDrive_gB_PTHrP_Tdy (46).

Virus titers, quantitating productive infection in organs of interest, were performed with organ homogenates by a virus plaque assay performed under conditions of “centrifugal enhancement of infectivity” [(45), and references therein].

Single-cell suspensions were prepared from spleen and lungs as described (21, 45). In the case of splenocytes, mice were tested individually. In the case of lung infiltrate cells, cohort analyses were performed with cell pools due to limited cell yield.

Unspecific staining was blocked with unconjugated anti-FcγRII/III antibody (anti-CD16/CD32, clone 93; catalog no. 14-0161; eBioscience, San Diego, CA, USA), and cells were specifically stained with the following antibodies for multi-color cytofluorometric (CFM) analyses: FITC-conjugated anti-CD8a (clone 53-6.7, catalog no. 553031; BD Biosciences, Franklin Lakes, NJ, USA), PE-conjugated anti-KLRG1 (clone 2F1, catalog no. 12-5893; eBioscience), and PE-Cy7-conjugated anti-CD62L (clone MEL-14, catalog no. 731715; Beckman Coulter, Brea, CA, USA). IE1-epitope-specific CD8 T cells were identified by staining with APC-conjugated peptide-folded MHC-I dextramer H-2Ld/YPHFMPTNL (m123/IE1) (Immudex, Copenhagen, Denmark).

A lymphocyte live gate was routinely set in the forward vs. sideward scatter (FSC vs. SSC) plot. All CFM analyses were performed with flow cytometer FC500 and CXP analysis software (Beckman Coulter).

An interferon gamma (IFNγ) enzyme-linked immunospot (ELISpot) assay was performed for quantification of IFNγ-secreting CD8 T cells after sensitization by peptide-loaded stimulator cells. Frequencies of mCMV-specific CD8 T cells were determined by incubation of graded numbers of immunomagnetically-purified total CD8 T cells with P815 (H-2^d) stimulator cells that were exogenously loaded with synthetic peptides at a saturating concentration of 10^-7M [(27, 47) and references therein]. Spots were counted automatically based on standardized criteria using Immunospot S4 Pro Analyzer (CTL, Shaker Heights, OH, USA) and CTL-Immunospot software V5.1.36.

To evaluate statistical significance of differences between two independent sets of data, the unpaired t-test (two-sided) with Welch’s correction of unequal variances was used. Differences are considered statistically significant at levels of significance marked by asterisks: (*) P < 0.05, (**) P < 0.01, and (***) P < 0.001.

In ELISpot analyses, frequencies of epitope-specific IFNγ-secreting CD8 T cells and the corresponding 95% confidence intervals were calculated by intercept-free linear regression analysis. Frequencies are considered significantly different if the 95% confidence intervals do not overlap.

For analyzing the dynamics of epitope-specific CD8 T-cell populations, a trend analysis was performed by linear regression. Rising and declining trends are reflected by positive and negative slopes of regression lines, respectively. Trends are considered statistically significant for P-values of < 0.05, confirming linearity, and 95% confidence intervals for the slope that do not include a slope of zero. Calculations were performed with Graph Pad Prism 10 (Graph Pad Software, San Diego, CA, USA).

It was the aim of our study to identify the predominant route of mCMV antigen presentation in the specific context of hematopoietic reconstitution after experimental HCT. We took the approach of modulating the level of presented antigenic peptide by presence or absence of the key immune evasion protein m152, which traps pMHC-I complexes in a cis-Golgi compartment [(48–50), reviewed in (40)]. For this, we infected HCT recipients either with WT virus mCMV-WT or with the m152 gene deletion mutant mCMV-Δm152, resulting in low and high cell surface expression of pMHC-I complexes, respectively ( Figure 1 ).

The time course of CD8 T-cell reconstitution in the spleen after syngeneic HCT revealed a comparable reconstitution of total CD8 T cells in mice infected with the immune evasion gene deletion mutant mCMV-Δm152 compared to WT virus ( Figure 2A , left panel). This makes sense, because total CD8 T cells represent the broad and random TCR-specificity repertoire, whereas modulation of mCMV immune evasion primarily affects the priming and clonal expansion of CD8 T cells specific for viral peptides. In accordance with this reasoning, a more efficient response of antiviral CD8 T cells specific for a viral peptide, here shown for the immunodominant IE1 peptide presented by the MHC-I molecule L^d (51–53), was seen as a trend at 6 weeks after infection with the mutant virus. This trend reached statistical significance at later times, until the frequencies of IE1-specific cells converged again at a late stage ( Figure 2A , right panel).

If “direct antigen presentation” applies to the priming of naïve CD8 T cells in our system, the magnitude of the CD8 T-cell response should reflect the cell surface level of pMHC-I complexes on infected cells determined by immune evasion gene expression, and correlate inversely with viral load. In contrast, if “antigen cross-presentation” applies, the magnitude of the CD8 T-cell response should be independent of immune evasion gene expression in infected cells and should rather reflect the viral load that determines the amount of antigenic material available for uptake by cross-presenting DCs. As we have shown in a previous report on the HCT model (54), CD11c⁺ DCs of donor-genotype are successfully reconstituted, including the CD8⁺ cDC1 subset that is capable of antigen cross-presentation (55–57).

The time course of clearance of productive infection in the spleen after experimental syngeneic HCT has already been published and showed that virus production ceases between weeks 8 and 12 after HCT and infection [ Supplementary Figure 1 , modified from (53)]. On this basis, CD8 T-cell response and viral load in the spleen of HCT recipients were determined at 8 weeks, shortly before the end of the productive phase of infection, that is, at a time when viral antigenic material was still available from current and preceding viral replication for a potential cross-presentation.

After infection with WT virus, a low IE1-specific CD8 T-cell response corresponded to a high viral load, whereas after infection with mutant virus mCMV-Δm152, a high response corresponded to a low viral load ( Figures 2B, C ). Differentiated by CD8 T-cell activation subsets (58) (recall Figure 1 ), cells of inflationary T effector-memory cell (iTEM) phenotype KLRG1⁺CD62L^-, which reflects more recent sensitization by antigen (59), benefited most from deletion of m152 ( Figures 2B, D ). Notably, recent work has shown that KLRG1^-CD62L⁺ T central memory cells (TCM) contribute most to the control of infection upon adoptive transfer due to their high proliferation potential (60). Although TCM did not profit from deletion of m152 relative to the other subsets ( Figure 2D ), their absolute number was increased due to the overall increase in the number of IE1-specific CD8 T cells in absence of immune evasion ( Figure 2C , center panel).

In essence, the magnitude of the antiviral CD8 T-cell response positively correlated with antigen presentation on infected cells and negatively correlated with the amount of antigenic material available for a potential cross-presentation.

Up to this point, we have shown that abrogation of immune evasion by deletion of m152 leads to an enhanced CD8 T-cell response due to improved direct antigen presentation. Following this logic, one must postulate that in the reverse case of enhanced immune evasion by overexpression of m152, a reduced CD8 T-cell response should result, because direct antigen presentation is further inhibited compared to infection with WT virus.

We have recently described the new recombinant virus mCMV-m152.IE+E (28), with which m152 is expressed from its authentic genomic position as an Early (E) phase protein and, in addition, expressed ectopically as an Immediate-Early (IE) phase protein. This leads to an overexpression of m152 combined with an earlier onset of immune evasion in infected cells (28).

Consistent with enhanced immune evasion, infection with the “super-evasion” virus mCMV-m152.IE+E resulted in an increased viral replication associated with a reduced CD8 T-cell response compared to WT virus and reciprocal to the data with mCMV-Δm152 at the end of the productive phase in the spleen at 8 weeks after syngeneic HCT ( Figure 3A ). In the latent phase at 24 weeks after syngeneic HCT, the latent viral genome loads of WT virus and mCMV-m152.IE+E in the spleen had almost equalized and were significantly higher by a factor of ~10 compared to the immune evasion gene deletion mutant mCMV-Δm152 ( Figure 3B , left panel). Imprinted by the CD8 T-cell response during productive infection, the frequencies of viral epitope-specific CD8 T cells during latent infection remained in the rank order of mCMV-Δm152 >> mCMV-WT > mCMV-m152.IE+E, which is most pronounced for the known immunodominant epitopes IE1 and m164 in the H-2^d haplotype (53, 61, 62) ( Figure 3B , right panel).

It should be noted that differences between epitopes do not necessarily indicate differences in the mode of priming, but merely reflect differences in clonal expansion. In particular, minor differences between experimental groups do not reach statistical significance after a few proliferation cycles of CD8 T cells specific for subdominant epitopes such as M105 and m145, but can reach statistical significance after several proliferation cycles of CD8 T cells specific for immunodominant epitopes such as IE1 and m164. The same principle has also been shown for the two immunodominant epitopes, where the difference between the WT virus and an immune evasion gene deletion mutant only became statistically significant over time of clonal expansion (27).

Altogether, the approach to up- or down-modulate immune evasion in infected cells confirmed direct antigen presentation as the predominant pathway of antigen presentation.

The lungs represent the most relevant organ site of CMV pathogenesis, specifically in the phase of hematopoietic reconstitution, both after clinical HCT (17–20) as well as after experimental HCT in the mouse model (21, 52). We therefore turned to the analysis of immune evasion-regulated viral infection of the lungs and the CD8 T-cell response in pulmonary infiltrates in the phase of productive infection and during latent infection with the immune evasion gene deletion mutant mCMV-Δm152 compared to WT virus ( Figure 4 ).

Levels of infectious virus ( Figure 4 , upper panel) were compared to viral genome load ( Figure 4 , lower panel) to define the time when productive infection was cleared and latent infection established in the lungs. Already at the beginning of the time-course analysis at 4 weeks after HCT, mCMV-Δm152 was more efficiently controlled than the WT virus, both in terms of reduction of productive infection as well as of viral DNA load. To be on the safe side, we defined the time after which latent infection was established as 14 weeks after HCT. In accordance with the definition of viral latency (2), infectious virus was absent beyond that time, whereas viral genome was maintained until the end of the observation period. Of note, the load of latent viral DNA was lower for the mutant virus throughout, indicating more efficient control by antiviral CD8 T cells during the resolution of acute infection based on enhanced direct antigen presentation by infected cells in the absence of immune evasion.

Depending on how complete the hematoablative treatment has eradicated cells of the bone marrow and the immune system, recipients of clinical HCT establish complete or mixed chimerism, in which all or only a fraction of hematopoietic cells are of donor-genotype, respectively (63, 64).

As we have shown in a previous report on latent infection established after sex-mismatched HCT in the mouse model (65), recipient-genotype CD11c⁺ DCs become largely replaced by donor-genotype CD11c⁺ DCs, whereas donor-genotype CD11b⁺ macrophages account for only half of the population. In the lungs, latent mCMV genomes do not localize to cells expressing the fractalkine receptor CX3CR1 (66), which excludes both CD11b⁺ CX3CR1⁺ macrophages and CD11c⁺CX3CR1⁺ DCs as sites of mCMV latency and direct antigen presentation. Non-hematopoietic parenchymal or connective tissue cells are exclusively of recipient-genotype.

By using different genetic approaches, own previous work (54) and work by the group of A. Oxenius (67) have independently shown that viral antigen presentation during latency of WT virus depends on direct antigen presentation by latently infected non-hematopoietic tissue cells of recipient-genotype. At that time, the latently infected cell type for mCMV was still unknown. Meanwhile, endothelial cells (58, 65, 68) and PDGFRα⁺ fibroblasts (69), both non-hematopoietic cell types, were identified as cellular sites of mCMV latency.

To test if direct antigen presentation by latently infected non-hematopoietic tissue cells also applies to latent infection with mCMV-Δm152, we compared CD8 T-cell responses in H-2^dχd syngeneic chimeras, in which donor and recipient cells differ only epigenetically, with H-2^dχdm2 allogeneic chimeras, in which only donor-derived pAPCs express the MHC class-I molecule L^d that presents the antigenic IE1 peptide (for the principle, see Figure 5A ). The result was clear and showed that the pool sizes of IE1 epitope-specific total CD8 T cells and the three activation subsets thereof were largely reduced during latent infection with mCMV-WT as well as with mCMV-Δm152 when cells of recipient-genotype did not express the presenting MHC-I molecule L^d ( Figure 5B ). In conclusion so far, regardless of whether or not direct antigen presentation was enhanced, the CD8 T-cell response during latent infection depended on cells of recipient-genotype and thus not on reconstituted hematopoietic-lineage pAPCs.

It is becoming increasingly clear that latent CMV genomes are not completely silenced at all genomic loci and at all times. Instead, episodes of local epigenetic viral gene desilencing lead to transient events of transcription (70–73) that do not follow the coordinated productive cycle gene expression cascade of immediate-early (IE), early (E), and late (L) phase transcription (74–76), and that therefore do not lead to a recurrence of infectious virus. Linking this insight to the CD8 T-cell response during latent infection, it has been a major contribution of our group to have shown stochastic and transient expression also of viral genes that encode antigenic peptides (58, 77, 78) driving a more or less continuous expansion of the viral epitope-specific CD8 T-cell pool over time. This phenomenon is known as “memory inflation (MI)” [for reviews, see (77, 79–81)], but in both the H-2^d (61, 82) and the H-2^b (83, 84) haplotype, MI applies only to few of the known antigenic viral peptides. MI is primarily based on the expansion of KLRG1⁺CD62L^- iTEM (58), which were originally named “short-lived effector cells” (SLECs) (85), but were found to differ from terminally-differentiated effector cells by their proliferative capacity and dependence of their tissue maintenance on IL15 (86).

When comparing the time course of the CD8 T-cell response to the prototypical MI-inducing epitope IE1 (82) after syngeneic HCT for mCMV-WT and mCMV-Δm152, a fundamental difference became apparent ( Figures 6A, B ). Infection with WT virus led to a low response during productive infection due to low direct antigen presentation, followed by iTEM-based MI aided by high latent viral genome load (recall Figure 4 ) associated with frequent episodes of restimulation during latency. Just opposite to this, infection with the m152 gene deletion mutant led to an initially high response due to high direct antigen presentation, followed by a steady decline in the number of iTEM due to low latent viral genome load (recall Figure 4 ) that limits restimulation during latency. Inflation and deflation of iTEM are statistically confirmed by a linear regression analysis revealing a positive and a negative trend after infection with mCMV-WT and mCMV-Δm152, respectively ( Figure 6C ). Surprisingly, loss of iTEM did not result in a notable gain of cTEM ( Figure 6B ), although conversion of iTEM to cTEM by loss of KLRG1 expression was expected. We did not pursue this finding further and can therefore only speculate that iTEM do not quantitatively convert to cTEM but get lost.

Throughout the time course, IE1 epitope-specific TCM were not notably involved in the composition of the CD8 T-cell pool in pulmonary infiltrates, which is consistent with the fact that TCM, expressing the lymphoid homing receptor CD62L, do not home to non-lymphoid tissues but first need to convert to CD62L^- TEM (87). Consistent with this, an own recent study localized IE1 epitope-specific TEM, but not TCM, to the extravascular compartment of the lungs (88).

Altogether, our data prove that the pool of viral epitope-specific CD8 T cells in pulmonary infiltrates is predominantly generated by direct antigen presentation during both productive and latent infection.