Peptide Processing Is Critical for T-Cell Memory Inflation and May Be Optimized to Improve Immune Protection by CMV-Based Vaccine Vectors

Cytomegalovirus (CMV) elicits long-term T-cell immunity of unparalleled strength, which has allowed the development of highly protective CMV-based vaccine vectors. Counterintuitively, experimental vaccines encoding a single MHC-I restricted epitope offered better immune protection than those expressing entire proteins, including the same epitope. To clarify this conundrum, we generated recombinant murine CMVs (MCMVs) encoding well-characterized MHC-I epitopes at different positions within viral genes and observed strong immune responses and protection against viruses and tumor growth when the epitopes were expressed at the protein C-terminus. We used the M45-encoded conventional epitope HGIRNASFI to dissect this phenomenon at the molecular level. A recombinant MCMV expressing HGIRNASFI on the C-terminus of M45, in contrast to wild-type MCMV, enabled peptide processing by the constitutive proteasome, direct antigen presentation, and an inflation of antigen-specific effector memory cells. Consequently, our results indicate that constitutive proteasome processing of antigenic epitopes in latently infected cells is required for robust inflationary responses. This insight allows utilizing the epitope positioning in the design of CMV-based vectors as a novel strategy for enhancing their efficacy.


Introduction
Cytomegalovirus (CMV) infection maintains the strongest immune response known in clinical medicine, dominating the T-cell memory compartment of seropositive hosts [1]. CMV is a herpesvirus that is never fully eliminated from the host, which may explain why these responses can be detected even at late time points upon initial infection [2,3]. The T-cell responses to immunodominant CMV antigens appear only to increase with age [4] but at the same time stay functional even in very old, otherwise immunosenescent hosts [5]. Therefore, it has been proposed that CMV recombinants expressing heterologous antigenic determinants [6] may be used as superior vaccine vectors.
Cytomegalovirus (CMV) based vaccine vectors have attracted broad attention over the past years as promising vectors for the induction of protective cellular T-cell responses against a variety of viral, bacterial and tumor targets [6][7][8][9][10][11]. Rhesus Cytomegalovirus (RhCMV) vectors encoding antigens from the simian immunodeficiency virus (SIV), a virus used in rhesus monkeys as a model of HIV-AIDS disease, sustain a remarkable SIV-specific T-cell response even in CMV-positive animals [11] and clear highly virulent SIV from more than 50% of the monkeys, thus preventing the development of AIDS-like disease [12,13]. It has been proposed that the ability of CMV-based vaccines to provide this unprecedented level of protection against SIV depends on effector-memory (EM) T-cell responses intercepting viral dissemination at sites of virus entry into the host [14]. How and why CMV sustains this unique immune response is still unresolved and we need to clarify these mechanisms to optimize CMV based vaccines. The cellular and molecular mechanisms of T-cell priming and maintenance by CMV vectors can be addressed in minute detail by infection of inbred and transgenic mice with genetically modified mouse CMV (MCMV).
Experimental MCMV infection was shown to induce persistent infiltrates of CD62L -EM CD8 T-cells in solid organs [15] directed against immunodominant MCMV epitopes [16]. Furthermore, MCMV vectors encoding single antigenic epitopes induced inflationary [17] CD8 T-cell responses against the heterologous epitope, and provided immune control upon challenge with a recombinant vaccinia virus carrying the same epitope [6]. On the other hand, only some epitopes encoded by CMV induce inflationary EM responses [16,18,19] and the a scholarship by the Melanie and Eduard zur Hausen foundation. ABR received funding for this study from a Marie-Curie Career Integration Grant, and by the Lautenschläger foundation. This work was also in part supported by the NIH grant R43CA180177 to Tomegavax. NAWL received intramural funding from the young investigator program MAIFOR of the University Medical Center Mainz. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors declare the following competing financial interest: Oregon Health and Science University (OHSU) and KF have a significant financial interest in TomegaVax Inc., a company that may have a commercial interest in the results of this research and technology. LCS was a paid consultant of Tomegavax in 2015. This potential individual and institutional conflict of interest has been reviewed and managed by OHSU. RBR and CM have been employed by TomegaVax, Inc., a commercial company. The authors declare that there are no further financial interests. mechanisms driving this selection are incompletely understood. A plethora of potential mechanisms contributing to inflationary EM responses has been proposed [20], including the efficacy of antigen processing, avidity of peptide binding to MHC molecules and avidity of the Tcell receptor (TCR) binding to peptide-MHC-I (pMHC-I) complex. We showed previously that inflationary responses depend on the context of epitope expression, rather than peptideintrinsic properties [8]. On the other hand, the viral gene M102 was shown to simultaneously induce inflationary and non-inflationary CD8 T-cell responses [18]. Therefore, promoter activity alone could not explain the entire selection process.
We show here that C-terminal localization of a peptide results in drastically improved immune protection by CMV-based vaccine vectors. We also show that an MCMV peptide that induces conventional CD8 T-cell responses from its native site becomes inflationary when transferred on the C-terminus of the same viral protein. Finally, we show that these striking differences in size and type of response correspond to the presence of the pMHC-I complexes on the surface of in vitro virus-infected endothelial cells and that the effect critically depends on the availability of the peptide for constitutive proteasomal processing.

C-terminal localization provides optimal immune protection
To test immune protection by CMV-based vectors in a model of human papilloma virus (HPV) induced cancer, we generated a recombinant MCMV (MCMV E6+E7 ) expressing the full-length E6 and E7 proteins of the HPV strain 16 (HPV16). We placed E6 and E7 under the control of the HCMV immediate-early (IE) promoter, because we showed previously that an IE promoter induces stronger CD8 T-cell responses than an early one [8]. Moreover, we utilized an MCMV backbone lacking the viral genes m1 to m16 [21] and thus providing ample cloning capacity. At 10 weeks post infection (p.i.) with MCMV E6+E7 , the response to the immunodominant D b -restricted E7 49-57 peptide RAHYNIVTF was detectable, although weaker than the response to the endogenous inflationary IE3 epitope (S1 Fig). We tested next the immune protection by MCMV E6+E7 against a challenge with E6+E7 transformed TC-1 tumor cells [22]. 25,000 TC-1 cells were administered at 27 weeks post immunization and MCMV E6+E7 immunized mice showed reduced tumor growth in comparison to mockvaccinated mice, but not a complete block of tumor growth ( Fig 1A). To define if protection was mediated by CD8 T-cell responses, we generated another MCMV recombinant expressing only the MHC class I epitope E7 49-57 on the C-terminus of the ie2 protein (MCMV ie2E7 ). Control mice were immunized with the recombinant MCMV ie2SL , expressing the SSIEFARL epitope in the same location [8], or mock-immunized with PBS. Tumor growth upon challenge with TC-1 cells was completely prevented in MCMV ie2E7 (Fig 1B), and the mice remained tumor-free for the duration of the experiment. The improved immune protection by MCMV ie2E7 over MCMV E6+E7 was unexpected, because MCMV E6+E7 expressed numerous antigenic epitopes and MCMV ie2E7 only one, which was also present in MCMV E6+E7 . We reasoned that improved protection by MCMV ie2E7 could have been due to differences in promoter activity between HCMV IE and MCMV ie2, or by the better priming due to the Cterminal peptide localization. We analyzed this by generating a third MCMV mutant, where we fused the full-length E6 and E7 proteins to the C-terminus of the ie2 protein of a fulllength MCMV (MCMV ie2E6-7full ). Hence, MCMV ie2E6-7full used the MCMV ie2 promoter to drive the full-length E6+E7 transcriptional unit, but RAHYNIVTF was located in its native site, and not at the C-terminus. Six out of 9 mice vaccinated with MCMV ie2E6-E7full and challenged with 25,000 TC-1 cells displayed tumor growth (Fig 1C), arguing that epitope localization, rather than promoter activity, resulted in absolute immune protection by weeks intervals with recombinant MCMVs or control virus and challenged with 2.5x10 4 TC-1 cells/mouse at least 10 weeks after priming. Tumor size measured by caliper (mean +/-SEM is shown). (A) Immunization was performed with 10 6 PFU of MCMV E6+E7 (n = 9) and tumor growth compared to unvaccinated (naïve) controls (n = 10) (B) Mice were immunized with 10 5 PFU of MCMV ie2E7 , and compared to MCMV ie2SL or PBS control (n = 10 in each group) (C) Mice were prime/boosted with 10 5 PFU of MCMV ie2E7 . In that case, CD8 T-cell responses against the E7 49-57 epitope RAHYNIVTF should be stronger in MCMV ie2E7 infection than in MCMV recombinants expressing the full length protein. We compared immune response by pMHC-I dextramer staining (representative dot blots in Fig 1D), and the response was undetectable in MCMV E6+E7 infected mice, stronger in the MCMV ie2E6-7full group and strongest upon MCMV ie2E7 infection ( Fig 1E). In sum, our data argued that C-terminal epitope localization improved CD8 T-cell responses and thus, immune protection.

C-terminal epitope localization drives protective effector memory CD8 Tcell responses
To test if immune protection by a C-terminally expressed epitope would be generally applicable to another epitope and another MCMV gene, we used our previously described MCMVs expressing the K b -restricted SSIEFARL epitope as a C-terminal tag on the ie2 (MCMV ie2SL ) or on the M45 (MCMV M45SL ) protein [8]. Mice were immunized with either of the mutants or wild-type MCMV as control and challenged with a recombinant vaccinia virus expressing the same epitope (rVACV SL ). Both MCMVs expressing the SSIEFARL epitope significantly controlled rVACV SL replication (Fig 2A), providing further evidence that C-terminal localization of the antigenic epitope will result in immune protection. It has been proposed that the immune protection induced by CMV-based vaccine vectors rests on the induction of antigen-specific EM T-cells [12]. Hence, we analyzed the phenotype of SSIEFARL-specific CD8 T cells upon MCMV ie2SL , MCMV M45SL or rVACV SL infection. Blood leukocytes were in vitro restimulated with the peptide for 6h and IFNγ responding cells were classified according to CD127 and KLRG1 expression into EM (KLRG1 + CD127 -) or central memory (CM) (KLRG1 -CD127 + ) cells. All infections induced a predominantly EM phenotype in IFNγ + cells at 7 days post infection (dpi), but this response remained EM in both MCMV infections until 180 dpi (Fig 2B and 2C), while it rapidly shifted to a CM phenotype in rVACV infection ( Fig 2C).

The surface phenotype of CD8 T cells recognizing MCMV antigens is not determined by the gene expression pattern
Since the M45 gene encodes a natural D b -restricted epitope (HGIRNASFI), which induces non-inflationary CM responses [18,23,24], we were surprised by the EM response to SSIE-FARL upon MCMV M45SL infection (Fig 2). We showed previously that MCMV ie2SL induces inflationary CD8 T-cell responses to SSIEFARL, whereas the responses contract by 14 dpi in MCMV M45SL infection [8]. We considered that the C-terminal modification of the M45 protein could have destabilized the whole protein and altered the phenotype of the responses against all epitopes encoded by this M45 variant. Therefore, we compared the CD8 T-cell response to SSIEFARL and HGIRNASFI epitopes upon MCMV M45SL infection and noticed a drastic difference in their size and the quality. SSIEFARL induced~30-fold stronger responses than HGIRNASFI at all times p.i. (Fig 3A) and cells responding to HGIRNASFI showed a CM phenotype (CD127 + , KLRG1 -) within weeks p.i., while MCMV ieE6-7Full (n = 9) and compared to unvaccinated controls (n = 12) (D) Representative flow cytometry plots of dextramerstained blood lymphocytes from mice infected with 10 5 PFU of MCMV E6+E7 , MCMV ie2E7 , MCMV ie2E6-7Full or MCMV ie2SL and analyzed by D(b) E7 49-57 dextramer staining for the presence of E7-specific CD8 T cells at 21 weeks post-priming. (E) Group values from dextramer staining as in panel D are shown (each symbol is a mouse; horizontal line shows the median). Significance was assessed by Kruskal-Wallis test followed by Dunn's post hoc analysis for indicated columns. **p < 0.01, ns-not significant.
doi:10.1371/journal.ppat.1006072.g001 SSIEFARL-specific cells retained an EM (CD127 -, KLRG1 + ) phenotype for up to 180 dpi ( Fig 3B). Similarly, CD62L expression remained low on SSIEFARL specific cells, regardless of expression context (S1B Fig). Since the C-terminal modification did not alter the phenotype of responses against HGIRNASFI, the EM phenotype of responding T cells was specific for SSIEFARL.

C-terminal peptide localization results in effector memory CD8 T-cell responses
The test if the difference in responses to SSIEFARL and HGIRNASFI depended on peptide sequence or localization within the protein, we generated a new recombinant MCMV, where the HGIRNASFI epitope was moved from its original location to the M45 C-terminus. First we generated a negative-control virus lacking the HGIRNASFI epitope (MCMV M45I->A ), by replacing the D b -anchoring isoleucine at the HGIRNASFI C-terminus with an alanine (S2A Fig), which precluded efficient peptide processing and anchoring to the D b molecule (the approach is described in [25]). On this background we introduced the HGIRNASFI peptide at the C-terminal end of the M45 protein ( Mice were infected with either of the new mutant viruses or with MCMV WT and HGIR-NASFI-specific CD8 T-cell frequencies were monitored over 180 days by peptide restimulation of blood leukocytes and intracellular staining for IFNγ. As expected, HGIRNASFI-specific CD8 T cells were undetectable in mice infected with the MCMV M45I->A recombinant, but clearly responded to MCMV WT infection. Remarkably, MCMV M45Cterm elicited an~8-fold stronger response to HGIRNASFI at 7 dpi than MCMV WT (Fig 4A, upper row). While HGIR-NASFI-specific CD8 T cells declined from this peak in both groups by 180 dpi, their percentage was about 70-fold higher in MCMV M45Cterm than in MCMV WT infected mice (Fig 4A,  bottom row). Detailed analysis of the response kinetics showed that responses upon MCMV WT infection contracted by 14 dpi and remained low thereafter, whereas the contraction of responses in MCMV M45Cterm infection was followed by a slight, but clearly noticeable inflation starting by 28 dpi (Fig 4B).
The long-term phenotype of the peptide-specific cells was CM in MCMV WT infection, but EM upon MCMV M45Cterm infection (Fig 4C-4E). Thus, C-terminal localization of the HGIR-NASFI peptide within the M45 protein resulted not only in a stronger and inflating CD8 T-cell response, but also in a high percentage of peptide-specific cells with an EM phenotype at late time points after infection.

C-terminal localization of the epitope improves processing and direct surface presentation
It was theoretically possible that virus mutagenesis and propagation resulted in unwanted mutations of immune evasion genes, improving peptide presentation and CD8 T-cell priming. To exclude this scenario, we compared MHC-I surface expression upon infection with MCMV M45Cterm , MCMV M45I->A , MCMV WT and a mutant lacking the immune evasion genes m06 and m152 (MCMV Δm06m152 ). All of the MCMV recombinants, except MCMV Δm06m152 , efficiently down-regulated MHC class I D b molecules on infected liver sinusoidal endothelial cells (LSECs) (Fig 5A and S3A Fig). Thus, the inflationary EM response to the HGIRNASFI epitope encoded by MCMV M45Cterm was not due to different surface levels of MHC-I. Since the HGIRNASFI epitope located at its native site is poorly processed and presented on infected fibroblasts [26], we considered that the improved response to MCMV M45Cterm might be due to the availability of the peptide itself on MHC-I molecules. Hence, to measure the presentation of HGIRNASFI on LSECs, we analyzed the IFNγ response of a HGIRNASFI-specific CD8 T-cell line (CTL) upon co-culture with a recently published LSECs line [27]. LSECs infected with the control virus (MCMV M45I->A ) or with MCMV WT did not activate M45-specific CTL, whereas the MCMV M45Cterm virus induced a robust activation ( Fig 5B, upper row). Importantly, no CTL responses were observed upon infection with MCMV Δm06m152 , demonstrating that the HGIRNASFI epitope expressed at its native site cannot activate CTL responses to infected LSECs even if MHC-I molecules are present at high levels on the cell surface. In theory, it was possible that an unknown MCMV gene that impairs the processing and presentation of the HGIRNASFI peptide in MCMV WT infection was accidentally lost in the MCMV M45Cterm mutant. Therefore, we UV-inactivated the viruses to abolish de novo gene expression and we co-cultured the cells exposed to UV-inactivated virus with M45-peptide-specific CTLs. Since M45 is a tegument protein, it is available in cells upon entry of UV-inactivated virus. As shown in Fig 5B bottom  To conclusively show that HGIRNASFI localization within the M45 protein determines antigenic peptide availability on the D b molecules of the infected cells, we performed a targeted nanoflow liquid chromatography mass spectrometry (nanoLC-MS 3 ) analysis. Total D b MHC-I molecules were immunoprecipitated (IP) from lysates of LSECs infected with MCMV WT , MCMV M45I->A , MCMV Δm06m152 or MCMV M45Cterm . Upon epitope elution from the pMHC-I  complexes, IP samples were analyzed by targeted nanoLC-MS 3 for the presence of the HGIR-NASFI peptide.
As expected, the target peptide was not detected in the sample from MCMV M45I->Ainfected cells. In line with previously published functional assays [26,28], the peptide was also not detected in MCMV WT -infected cells, or in those infected with MCMV Δm06m152 , although we detected traces of HGIRNASFI in one out of three repetitions of MCMV WT and MCMV Δm06m152 infection. However, the target peptide was highly abundant in all 3 replicates of the MCMV M45Cterm -infected LSECs ( Fig 5C) and its spectrum matched the one from the synthetic HGIRNASFI peptide (S3C Fig). Thus, the MS data confirmed that the C-terminal localization of the peptide facilitated its processing and its presentation in pMHC-I complexes.
Considering the very poor HGIRNASFI presentation on MCMV WT infected cells, it is counterintuitive that the M45 epitope is immunodominant at 7 dpi [18]. However, this early immunodominance might be explained by peptide cross-presentation on professional antigen-presenting cells (APCs). This would imply that inflationary responses depend on epitope presentation on latently infected non-hematopoietic cells. To validate this idea, we generated chimeric mice with impaired MHC-I antigen presentation on professional APCs (but maintained on non-hematopoietic cells), by hematopoietic reconstitution of gamma-irradiated C57BL/6 recipients with TAP-deficient bone-marrow (BM) cells (TAP -/-!B6). Mice were infected with MCMV M45Cterm or MCMV WT and monitored for HGIRNASFI-specific responses. Additional controls included homochimeric mice where C57BL/6 mice were used both as BM donors and recipients (B6!B6). While B6!B6 mice showed kinetics that essentially matched the one observed in wild-type mice, the TAP -/-!B6 mice revealed a complete loss of HGIRNASFI response upon MCMV WT infection (Fig 5D top panel and S3D Fig). In contrast, MCMV M45Cterm induced strong CD8 T-cell responses against the peptide in TAP -/-!B6 mice (S3D Fig) that were undistinguishable from responses in B6!B6 mice at later times p.i. (Fig 5D bottom panel). Interestingly, the initial peak response, seen in WT mice or in B6!B6 controls was absent from the TAP -/-!B6 mice, which may indicate that this initial response is mainly driven by cross-presentation. Finally, one should note that approximately 90% of BM-derived cells in chimeric mice were donor-derived (S3E Fig). Therefore, the long-term CD8 T-cell response to the C-terminal epitope was maintained in absence of APC-dependent cross-presentation (Fig 5D), implying that it may depend on its direct presentation by virus-infected cells, although we cannot exclude the possibility that the initial priming was due to cross-presentation by the few remaining TAP-competent APCs derived from the recipient BM.
Taken together, our results showed that peptide localization within the M45 protein, rather than targeted activity of immune evasion genes, is the limiting factor for CTL recognition of the HGIRNASFI peptide in MCMV-infected LSECs.

Role of peptide processing in the generation of inflationary responses
Since the MHC-I availability of the same peptide expressed by the same viral gene differed greatly based on its localization within the protein, we considered that this difference may be due to improved peptide processing prior to loading on MHC-I molecules. Antigenic peptide processing and subsequent surface presentation of peptide MHC-I complexes can be enhanced by altering amino acid residues flanking an epitope [29]. Thus, we generated a novel MCMV recombinant called MCMV M45ASL , carrying the SSIEFARL peptide at the C-terminus of the M45 protein, but preceded by 2 alanines (S4A Fig). Thus, these two alanines were the only difference between MCMV M45ASL and MCMV M45SL [8]. The in vitro and in vivo growth of MCMV M45ASL was comparable to MCMV WT (S4B Fig). To test the effect of these flanking residues on epitope recognition by CD8 T cells, we co-cultured CD8 T cells from transgenic gBT-I mice expressing a T-cell receptor specific to the K b -SSIEFARL complex [30], with IC-21 macrophages infected with MCMV M45SL or MCMV M45ASL , and assessed them for TNFα and IFNγ production. Infection with the recombinant MCMV containing the alanine spacer induced a stronger T-cell response than the one without it (Fig 6A). The same result was observed upon co-culture with in vitro infected LSECs (S4C Fig). This implied that peptide processing is a rate limiting step in activating CD8 T cells by our mutants. Therefore, we tested if the same would apply in vivo and affect the size of inflationary responses to SSIEFARL. We infected mice with MCMV M45SL , MCMV M45ASL or, as a non-inflationary control, with rVACV SL . At 7 dpi, the percentage of SSIEFARL-responding cells was identical in mice infected with either MCMV recombinant (Fig 6B). Likewise, both mutants induced SSIEFARL responses with inflationary phenotype (S4D Fig). On the other hand, the peptide-specific CD8 T cells showed an inflationary trend only in mice infected with the MCMV M45ASL recombinant (Fig 6B), and differences in peptide-specific responses were statistically significant at all times after 60 dpi. Hence, improved peptide processing resulted in stronger MI.
To test if C-terminal expression improved the processing of peptides by the proteasome, MCMV M45Cterm or MCMV WT -infected LSECs were co-cultured with CTLs in the presence of two proteasome inhibitors-MG-132 or Lactacystin. Both inhibitors impaired CTL activation upon co-culture with cells infected with the MCMV M45Cterm recombinant in a dose dependent manner (Fig 6C). On the other hand, CTL activation by cells that were loaded exogenously with the peptide remained unimpaired (S4E Fig) and MCMV WT infection did not activate CTLs in presence or absence of the inhibitors (Fig 6C). Finally, treatment with a protease inhibitor (leupeptin) did not influence CTL recognition of the MCMV M45Cterm -infected target cells (Fig 6C), confirming that the C-terminally expressed epitope was proteasomally processed for CTL recognition. We infected mice lacking the immunoproteasome component LMP7 with MCMV WT or MCMV M45Cterm , and compared the kinetics of their HGIRNASFI responses to the parental C57BL/6 strain. The response was absent in LMP7 -/mice upon MCMV WT infection (Fig 6D top panel), but fully maintained upon MCMV M45Cterm infection. (Fig 6D  bottom panel). Interestingly, the acute response at 7 dpi with MCMV WT was completely abrogated (p<0.001) in LMP7 -/mice (Fig 6D top panel), but only slightly impaired (p<0.05) upon MCMV M45Cterm infection (see Fig 6D bottom panel). This argued that even the acute CD8 Tcell response was largely immunoproteasome independent when the epitope was relocated to the C-terminal position. Taken together, our data strongly argued that the efficacy of peptide processing defines the rate of CD8 T-cell inflation for a given peptide in MCMV infection.

Discussion
We reported here several findings that were counterintuitive in light of the existing literature. Previous publications showed that immune sensing of an antigenic epitope may impair the transcription of viral genes expressed later in the process of viral reactivation [31], and that antigenspecific inflationary responses compete with each other [8,32]. Taken together, the evidence was unified in the Immune Sensing Hypothesis [20], where intermittent MCMV transcription during viral latency results in antigen expression, T-cell sensing and suppression of genes expressed later during reactivation, thus defining the immunodominance hierarchy. On the other hand, the Immune Sensing Hypothesis did not explain the inflationary responses against epitopes expressed by the early genes M38 or m139 [18]. We showed here that the conventional responses against the natural M45 epitope are not due to its silenced transcription in latency. In fact, we showed that the very same viral gene can simultaneously induce inflationary and conventional responses to two epitopes expressed within its protein product, thus demonstrating that promoter activity cannot alone predict which epitopes are inflationary. We observed that sequences flanking an antigenic peptide and its position within a protein may critically define the type of responses and that the efficacy of peptide processing by the constitutive proteasome is a rate-limiting factor for MI. Our results are not necessarily in conflict with the immune sensing hypothesis, but rather complement it with a secondary mechanism defining epitope dominance. Namely, while we show that immunodominance and MI depend on processing efficacy, the epitopes cannot be antigenic unless expressed, and hence the pattern of gene expression during viral latency is necessarily an important contributing mechanism [8,20].
Previous publications showed that CTL priming upon HSV-1 infection requires cross-presentation [33], and that the same might be true for the MCMV responses [34,35], but we showed that this holds true only for conventional CTL responses against a non-inflationary epitope. Transferring the conventional HGIRNASFI epitope to the C-terminus of the M45 protein allowed its direct presentation on virus-infected LSECs, and induced inflationary responses. Impaired cross-presentation (Fig 5D) abrogated the conventional HGIRNASFI response, but not the inflationary one. Previous data showed that non-hematopoietic cells, such as LSECs, are sites of MCMV latency [36,37] and that peptide presentation on nonhematopoietic cells is required for inflationary, but not for conventional immune responses [38,39]. However, our work differs in experimental design from published evidence in two important aspects, and thus allows us to posit novel conclusions. Firstly, we generated recombinant viruses that allowed us to compare the response to the same epitope expressed in two different locations of the same gene, while published evidence was based on cross-comparison of two distinct MCMV epitopes, expressed by different viral genes. Therefore, previous evidence could have been explained by peptide-intrinsic properties, or by cell-type specific differences in promoter activity, whereas our data excluded these scenarios and allowed us to focus on processing as the determinant of memory inflation. Secondly, we performed bone-marrow reconstitution of C57BL/6 mice with TAP deficient bone-marrow, while previous work was based on MHC-deficient recipient mice of wild-type bone marrow. Therefore, prior evidence showed that direct presentation is required for memory inflation, whereas our work might indicate that it is sufficient for this phenomenon. Namely, while the response to the HGIR-NASFI epitope in wild-type MCMV infection required cross-presentation, because it was completely abrogated in mice transferred with TAP -/bone-marrow, responses were undiminished when the epitope was expressed from the C-terminus, arguing that cross-presentation on APC might be dispensable for the induction of inflationary responses. If that was true, our result would point to direct priming by antigen presented on non-hematopoietic cells, which would defy a key dogma in immunology. It is important to note, however, that the irradiation and bone-marrow transfer procedure resulted in chimerism, where approximately 10% of PBMCs remained TAP-competent (S3E Fig). While this was not enough for the priming of conventional CD8 T cell responses in the setup of infection with MCMV WT , our data cannot exclude the possibility that the few remaining TAP-competent cells were sufficient to provide cross-priming and jump-start the inflationary immune response. In conclusion, while our data are intriguing, more detailed analyses are necessary to ascertain if inflationary responses may occur in the absence of an initial bout of cross-priming.
Previous evidence showed that the immunoproteasome is not strictly required for processing of inflationary epitopes [40]. Similarly, a recent publication showed that the non-independent experiments (in total 8 mice per each group) Significance on indicated time points was assessed by a Mann-Whitney U test. *p < 0.05. ***p < 0.0001, ns-not significant. inflationary HGIRNASFI epitope may induce inflationary responses when expressed as a minigene within an adenovirus-based vaccine vector [41]. However, these experiments could still be explained within the Immune Sensing Hypothesis, where memory inflation is predicated by the context of gene-expression of the antigenic epitope. By testing the same epitope in the context of the same protein, we excluded confounding factors and showed directly that peptide processing by the constitutive proteasome and direct antigen presentation on nonprofessional APC are the rate-limiting events for MI.
Predictive algorithms to identify epitope insertion sites with optimized antigenic processing [42], have been recently applied to score peptide processing of natural and modified epitopes expressed by MCMV [43]. However, published data argue that the proteasome needs to cleave peptides very precisely behind the anchoring amino acid at the C-terminus of an epitope (reviewed in [44]), whereas additional amino acids present on the N-terminal end of the epitope precursor peptide may be trimmed by aminopeptidases upon TAP-mediated transport into the ER [45]. Therefore, designing vectors that already carry the antigenic peptide at the Cterminus of a protein abrogates the need for proteasomal cleavage on the sensitive C-terminus of the peptide and thereby substantially improves the inflationary CD8 T-cell response. This insight likely explains why the surface presentation of the HGIRNASFI peptide, that T cells do not recognized in vitro (Fig 5B) or in vivo [24] at its native site, could be substantially optimized by this tweak in vector design. Accordingly, the E7 antigen, just like the prostate specific antigen [7], induced weak CD8 T-cell responses and poor protection when expressed within full length proteins in recombinant MCMVs, but a single E7 peptide fused to the C-terminus of the ie2 MCMV protein induced not only robust CD8 T-cell responses, but also absolute anti-tumor immunity (Fig 1). Similarly, SSIEFARL epitopes expressed on the C-terminus of two MCMV proteins provided protection against viral challenge (Fig 2A). In conclusion, our data showed that simple shifting of an epitope sequence to the C-terminus of a protein optimizes peptide processing and inflationary CD8 T-cell responses and circumvents the need for predictive algorithm scoring of the epitope insertion site.
Although our results unequivocally demonstrate that response magnitude and MI depend on the peptide context within a viral protein, one should take into account that additional parameters are likely to contribute to memory inflation and epitope immunodominance. These may include peptide intrinsic properties, such as its avidity of binding to MHC-I molecules and the avidity of TCR binding to pMHC complexes [46], but also extrinsic properties, such as the promoter strength and epitope competition [8,31,32], or the expression of additional viral genes, as has been recently described in the rhesus CMV model of infection [47]. Therefore, the mechanisms driving the exceedingly strong CMV responses remain a field of active research. Considering the huge potential that CMV-based vectors may have for the control of numerous lethal pathogens [9,12,13], understanding the mechanisms that optimize antigen presentation and the induction of T-cell responses remains of paramount scientific and clinical relevance. By providing the first direct evidence for a causal link between antigen processing efficacy, epitope presentation, memory inflation and immune protection, our study makes a fundamental contribution towards this goal.

Ethic statement
Mice were housed and handled in accordance with good animal practice. All animal experiments involving HPV immunization and challenge were performed at OHSU according to federal (U.S. Animal Welfare Act) and institutional guidelines, following the Institutional animal care and usage committee (IACUC) requirements, under the protocol (IACUC Study #IS00003413). Experiments involving LMP7 -/mice were performed according to U.K. Animal Welfare Act of 2006 and Home Office regulations (project license no. PPL 30/3293) after review and approval by the local Ethical Review Board at the University of Oxford. All other animal experiments were performed at HZI in compliance with the German animal protection law (TierSchG BGBI S. 1105; 25.05.1998) and were approved by the responsible state office (Lower Saxony State Office of Consumer Protection and Food Safety) under permit number 33.9-42502-04-11/0426. The mice were housed and handled in accordance with good animal practice as defined by FELASA and the national animal welfare body GV-SOLAS.

Peptides, primers
The peptides M45 (H-2D b -restricted, 985 HGIRNASFI 993 ) and the HSV-1 glycoprotein Bderived epitope (H-2K b -restricted, 498 SSIEFARL 505 ) [50] were synthesized and HPLC purified (65-95% purity) at the HZI peptide-synthesis platform. Primers used in the study are listed in the S1 Table. Viruses and viral mutagenesis MCMV M45ASL , MCMV M45Cterm , MCMV M45I->A and MCMV ie2E7 recombinants were generated by En passant mutagenesis as described by Tischer and colleagues [51], with modifications described by us [52]. In brief, linear PCR products were generated using the plasmid pGP704 I-SceIKan [52] as template and primers containing MCMV-homologue sequences and constructs to be introduced in the MCMV genes on their 3' ends. Each linear construct was generated by 2 rounds of PCR amplification [52], and contained sequences of the antigenic peptides inserted in the MCMV genome and sites of homologies allowing targeting to specific sites in the virus genome. All primers used in the study are listed in S1 Table. Upon cloning on chloramphenicol LB-agar plates insertion was confirmed by colony PCR and sequencing of the insertion site.
To generate MCMV E6+E7 , the fragments containing the entire E6 and E7 genes of HPV16 were cloned into pOriR6K-ie-zeo [53]. The resulting plasmid pO6-ie-E6+E7 was inserted into pSM3fr-Δ1-16-FRT [21] via Flp-mediated recombination [53] resulting in pSM3fr-Δ1-16-FRT-ie-E6+E7. The same E6 and E7 gene sequence was fused on the 3'-end of the ie2 gene in a pSM3fr MCMV BAC [54] by homologous recombination to generate pSM3fr-ie2E6+7. All viruses used in this study are listed in S2 Table. Reconstitution of MCMV from recombinant BACs was done by transfection of MEFs as described previously [8]. BAC-derived wild-type [55] and all recombinant MCMVs used in the study were propagated on M2-10B4 cells and virus stocks were purified on sucrose gradient as described previously [37]. Virus stocks were titrated on MEFs as described before [8]. MCMV from organ homogenate or tissue culture supernatants were titrated on MEFs as described previously [56].

In vitro infections
In vitro growth fitness of the recombinant viruses was determined on NIH 3T3 cells as described before [8]. For infection with centrifugal enhancement, plates were centrifuged at 2000rpm for 30 min, incubated at 37˚C, 5% CO 2 for another 30 min, upon which the supernatants were replaced with fresh medium. This procedure increases virus infection by a factor of 20 and thus a nominal MOI of 0.2 with centrifugal enhancement equals an MOI of 4 without it.

In vivo infection
Mice (6-12 weeks old) were intraperitoneally infected and housed in SPF conditions throughout the experiment. Infected mice showing very weak immune priming (2 standard deviations below average) at 7 dpi were regarded as outliers due to suboptimal infection and were excluded from the study.
Immune protection assays 129Sv mice were i.p. infected with 2x10 5 PFU of MCMV WT or MCMV M45SL . Eight months later, mice were i.p. challenged with 10 6 PFU of recombinant VACV SL . Ovaries were harvested at 7 days post challenge and infectious VACV titers established by plaque assay on Vero cells. For tumor challenge experiments animals were i.p. primed and boosted at 4 weeks interval with recombinant MCMV as indicated. Prior to challenge with 2.5x10 4 exponentially growing TC-1 tumour cells [22] mice were shaved on the hind left flank and s.c. injected. Mice were followed for tumor growth and tumor volume calculated by length x height x 0.5.

M45 (D b ) CTL generation
Splenocytes from C57BL/6 mice (latently (>3 months) infected with MCMV WT ), were isolated for CTL generation as described before [49] with minor modifications. IL-2 was added at a concentration 200U/ml, the peptide at a concentration 10 -10 M. CTL were used after the 2 nd round of re-stimulation.

Co-culture experiments
Target cells were seeded in 96-well plates. Infection was performed at an MOI 0.2 with centrifugal enhancement (~MOI 4 in standard conditions) as described above. After 1h infection, infectious supernatants were removed and effector cells (either in vitro generated M45(D b ) or ex vivo harvested gBT-I.1 CD8 T-cells) were added at an E:T ratio of 3:1. The cells were incubated at 37˚C for 1h, upon which brefeldin A (Cell Signaling Technology) was added at a concentration 10μg/ml and cells were incubated for 14h. For proteasomal inhibition experiments, target cells were pretreated for 5h with indicated inhibitors.

Liquid chromatography-mass spectrometry analysis (LC-MS)
LSECs were infected with indicated viruses at an MOI 2 with centrifugal enhancement. Cells were harvested 24 hpi and cell lysates were prepared with complete RIPA buffer with 2x protease inhibitor cocktail mix (Roche). The lysates were sonicated with Bioruptor Plus at low intensity and MHC-I molecules were immunoprecipitated with GammaBind Plus Sepharose Beads (GE Healthcare) coupled with anti-mouse H-2D(b) (Clone 28-14-8, BD Pharmingen). Immunoprecipitates were stored at -70˚C until MS analysis. Peptides were eluted from the immunoprecipitated pMHC complexes with 0.2% trifluoroacetic acid (TFA) in water, subjected to ultrafiltration with a cut-off of 10 kDa (Vivacon 500 filters, Sartorius Stedim Biotech), desalted with OMIX C18 10-100 μL pipette tips (Agilent Technologies) and vacuum dried. Samples were re-suspended in 3% acetonitrile (ACN), 0.1% formic acid (FA) and 0.01% TFA in water prior to LC-MS analysis.
Nano ultra-performance liquid chromatography mass spectrometry (nano-UPLC-MS) analysis was performed using a NanoAcquity UPLC system (Waters Corp.) coupled to a QTRAP6500 (AB SCIEX) mass spectrometer equipped with a nano-ESI (electron spray ionization) source. Samples were separated on a nanoAcquity UPLC BEH C18 analytical column (0.075 x 250mm) (Waters). The LC separation started with 97% eluent A (0,1% FA and 0,01% TFA in water) to 10% eluent B (0,1% FA and 0,01% TFA in ACN) by 1 min with a linear gradient and then to 40% eluent B by 50 min with a linear gradient. The flow rate was set to 300nL/ min. The mass spectrometer was operated in a low mass hardware profile operating in positive mode. The nano-ESI voltage was set at 2700 V, curtain gas at 30 L/min, ion source gas at 15 L/min, collision gas (CAD) high and interface heater temperature at 150˚C. The resolution of the first (Q1) and third quadrupole (Q3) was set at unit resolution.
Synthetic reference peptides for target and control epitopes were provided by the in-house DKFZ core facility with an analytical purity of >95%. Known cytoskeletal and housekeepingprotein derived H2-D b -restricted epitopes were used as positive controls [59,60].
A minimum of four fragments with the best signal-to-noise ratio were assigned per peptide during direct injection of synthetic peptides. Critical MS parameters (e.g., declustering potential, collision energy) were manually optimized to achieve the best sensitivity for the following peptides (precursor ion m/z: fragment ion m/z): HGIRNASFI (507.78: y 8 877. 49 The MS results of synthetic peptides were manually compared to the MS results acquired in the IP sample using the Analyst 1.6.2 (AB SCIEX) software. Identity of the targeted peptides was confirmed by their retention times, chromatographic profiles and MS 3 spectra of each fragment. Targeted nanoLC-MS 3 was performed to ensure sufficient sensitivity and specificity of MS results. Retention times (Fig 5C) and MS 3 spectra (S3C Fig) of the IP samples were compared to the results of the synthetic HGIRNASFI reference peptide.

Cell surface, intracellular staining and flow cytometry
Blood sample collection, processing and subsequent stimulation of blood lymphocytes with indicated peptides and intracellular cytokine staining was performed as described before [8]. The antibody panels used for the samples stimulated with the SSIEFARL peptide or stained with SSIEFARL-Kb tetramers shown previously [8] were expanded with anti-CD62L-eFluor605NC (MEL-14, eBioscience) for tetramer staining or with anti-KLRG1-Biotin (Clone 2F1; Biolegend) and anti-CD127-PE (Clone A7R34; Biolegend) for peptide re-stimulation.

Statistics
Statistical analysis was performed using GraphPad Prism program (version 5.04). Kruskal-Wallis analysis followed by Dunns post-analysis was used to compare multiple samples at single time points. Comparisons between two groups were performed using the Mann-Whitney U test (two-tailed).