|Home | About | Journals | Submit | Contact Us | Français|
CTLs have the potential to attack tumors, and adoptive transfer of CTLs can lead to tumor regression in mouse models and human clinical settings. However, the dynamics of tumor cell elimination during efficient T cell therapy is unknown, and it is unclear whether CTLs act directly by destroying tumor cells or indirectly by initiating the recruitment of innate immune cells that mediate tumor damage. To address these questions, we report real-time imaging of tumor cell apoptosis in vivo using intravital 2-photon microscopy and a Förster resonance energy transfer–based (FRET-based) reporter of caspase 3 activity. In a mouse model of solid tumor, we found that tumor regression after transfer of in vitro–activated CTLs occurred primarily through the direct action of CTLs on each individual tumor cell, with a minimal bystander effect. Surprisingly, the killing of 1 target cell by an individual CTL took an extended period of time, 6 hours on average, which suggested that the slow rate of killing intrinsically limits the efficiency of antitumor T cell responses. The ability to visualize when, where, and how tumor cells are killed in vivo offers new perspectives for understanding how immune effectors survey cancer cells and how local tumor microenvironments may subvert immune responses.
Tumors rely on a wide array of mechanisms to escape destruction by infiltrating CTLs (1). Adoptive therapy relying on the transfer of a large number of activated antitumor CTLs offers a promising approach to circumvent this limitation (2) and has been shown to induce tumor regression in animal models and human clinical trials (3, 4). However, the mechanisms underlying tumor elimination after CTL transfer remain largely unknown. Studies using animal models with a deficiency in various T cell effector molecules have been instrumental in identifying the multiple roles of CD8+ T cells during tumor regression. In some models, the ability of CTLs to kill (at least some) tumor or stromal cells appeared essential for tumor elimination (5–9). Other studies have noted a critical role for CD8+ T cell–derived IFN-γ (10–15) and the recruitment of inflammatory cells. In addition, some CTLs act primarily by promoting the expansion of other anti-tumor T cell clones (16, 17). Finally, CTL effector functions can synergize as the production of IFN-γ increases MHC class I molecule expression on tumor cells, which thereby reduces the activation threshold required to trigger T cell cytotoxic activity (18). These studies, however, do not resolve the spatiotemporal orchestration of tumor elimination. Intravital 2-photon imaging offers the opportunity to dissect T cell behavior in lymphoid organs and was recently used to visualize intratumoral T cell motility during the course of tumor regression (19, 20). One of the critical unresolved issues is whether CTLs act primarily by killing all tumor cells or by initiating the recruitment of innate effectors that mediate tumor destruction. In addition, the rate at which tumor cells are killed by cytotoxic effectors in vivo and a possible role for this parameter in promoting tumor regression have yet to be determined. Addressing these questions has been limited in part by the technical difficulty in monitoring CTL killing activity in situ and in real time. Here, using static and intravital 2-photon imaging, we developed new approaches for assessing the cytotoxic activity of intratumoral effectors in a mouse model of solid tumor. We show that adoptive T cell therapy using in vitro primed CD8+ T cells induces tumor rejection. We demonstrated that the vast majority of tumor cells are killed by CTLs, not by innate effectors, and that killing of 1 target cell occupied an individual CTL for an average of 6 hours. Interestingly, the response mounted by in vivo primed CTLs remained ineffective despite evidence of intratumoral CD8+ T cell killing. Collectively, our results provide evidence that the antitumor T cell response is intrinsically limited by the extent of CTL infiltration and the slow rate of tumor cell killing by CTLs.
To visualize CTL responses against a solid tumor, we used the well-characterized EL4/EG7 subcutaneous tumor model (10, 21). C57BL/6 mice injected s.c. with EL4 tumor cells or with the OVA-expressing variant EG7 developed solid tumors that failed to be rejected in the absence of additional manipulation. Similar tumor growth patterns were observed with EL4 or EG7 tumor cells transfected with a membrane-targeted cyan fluorescent protein (mCFP) or membrane-targeted yellow fluorescent protein (mYFP), respectively. As shown in Figure Figure1,1, adoptive T cell therapy through the transfer of in vitro activated OT-I CTLs resulted in the complete regression of EG7-mYFP tumors, but had no detectable effect on the growth of EL4-mCFP tumors. Interestingly, transfer of equivalent numbers of naive OT-I T cells (even at early time points) failed to induce tumor regression with little to no tumor cell elimination being detected in tumor frozen sections (Figure (Figure1). 1).
Nevertheless, both in vitro and in vivo primed CD8+ T cells were found to infiltrate the tumor and displayed enhanced effector functions at the tumor site (Supplemental Figure 1; supplemental material available online with this article; doi:10.1172/JCI34388DS1). These hallmarks of effector activity were highly reduced or absent in OT-I T cells found in the draining lymph node or within intratumoral OT-I CTL infiltrating the nonantigen bearing EL4-mCFP tumors (Supplemental Figure 1). These results strongly suggest that transferred CTLs enhanced their effector functions at the tumor site as a result of antigen recognition. Intratumoral CTL density, however, was different. After transfer of activated OT-I T cells, the CTL density was 1.3 ± 0.2 × 104 T cells/mm3 on day 2 and reached 2.5 ± 0.3 × 104 T cells/mm3 on day 3. CTL infiltration was significantly lower and less homogenous at all time points after transfer of naive OT-I T cells, with only 0.2 ± 0.03 × 104 OT-I T cells/mm3 on day 7 (Figure (Figure2,2, A and B).
We next focused on the efficient antitumor response mounted by in vitro activated CTLs by visualizing the distribution of intratumoral OT-I CTLs during the course of tumor destruction. Two days after being transferred, CTLs accumulated in multiple discrete areas of EG7 tumors found in the vicinity of tumor microvessels (Figure (Figure2C),2C), their likely port of entry. On day 3, CTLs were found more evenly distributed, and most of the tumor cells had been eliminated. Tumor areas containing low numbers of CTLs usually tended to have a higher density of tumor cells (Figure (Figure2C).2C). On day 5, the residual mass harvested at the tumor injection site did not show any viable EG7-mYFP. These observations suggest that intratumoral CTL infiltration is initiated in multiple regions of the tumor that progressively extend and ultimately merge. Intravital 2-photon imaging performed in tumor-bearing mice 2–3 days after CTL adoptive transfer revealed that CTL located inside regions where tumor cells had been eliminated were usually motile (Supplemental Movie 1). In contrast, many CTLs were sequestered at the border of these regions, being engaged in long-lasting interactions with tumor cells (Supplemental Movie 1). Thus, intratumoral CTL dissemination occurred concomitantly with tumor cell elimination. Noteworthy, many CTLs in the vicinity of tumor cells had a high granzyme B content (Supplemental Figure 2).
The close relationship between the topography of tumor elimination and that of CTL infiltration suggested that CTLs were playing a continuous role during tumor elimination. However, it was still unclear whether CTLs were acting by eliminating tumor cells directly or by recruiting innate effector cells that were responsible for most of the tumor destruction. To discriminate between these possibilities, we took advantage of the observation that mice injected with a mixture of EL4-mCFP and EG7-mYFP cells develop chimeric tumors (Figure (Figure3),3), which are composed of small individual patches of EL4-mCFP and EG7-mYFP cells. We reasoned that if the primary effectors of tumor cell killing were the CTLs themselves, then EG7-mYFP cells should be solely targeted. In contrast, if innate effectors were largely involved in tumor destruction, both EG7-mYFP and EL4-mCFP should be destroyed. Strikingly, 3 days after injection of OT-I CTLs, virtually all EG7-mYFP patches were cleared, whereas EL4-mCFP patches appeared minimally, if at all, affected (Figure (Figure3A).3A). This observation was further confirmed by analyzing the tumor cell composition as a percentage (Figure (Figure3,3, B and C) in absolute numbers (Supplemental Figure 3). Tumor cell mixtures containing as little as 10% EL4 tumor cells were not rejected following adoptive transfer of OT-I CTLs, which confirmed that tumor destruction proceeded with minimal nonspecific lysis. Thus, elimination of each individual tumor cell required the local action of antigen-specific CTLs, and tumor elimination occurred with little to no bystander effect.
To characterize the dynamics of tumor cell lysis by intratumoral CTLs, we transfected EG7 tumor cells with a fluorescent probe containing the CFP and YFP molecules linked by a peptide containing the caspase 3 cleavage motif DEVD (referred to as EG7-DEVD) (22). Apoptosis-induced caspase 3 activation resulted in substrate cleavage and subsequent Förster resonance energy transfer (FRET) disruption. As shown in Figure Figure4,4, FRET loss, which translated into a higher apoptosis index (see Methods), was readily detected in cultured EG7-DEVD tumor cells subjected to UVB irradiation or that cocultured with activated OT-I CTLs (Figure (Figure4).4). Importantly, FRET loss correlated with Annexin V staining on tumor cells cocultured with OT-I CTLs (Supplemental Figure 4). No FRET changes were detected when EG7 tumor cells were transfected with a control substrate containing a non-cleavable peptide DEVG (Figure (Figure4). 4).
Intravital 2-photon imaging of mice bearing an EG7-DEVD tumor and adoptively transferred with OT-I CTL revealed a close juxtaposition of CTLs and apoptotic tumor cells (Figure (Figure55 and Supplemental Movie 2). As shown in Figure Figure5,5, tumor cells found in contact with CTLs were much more likely to have undergone apoptosis than were tumor cells with no CTL in their vicinity. By tracking the apoptosis index of individual tumor cells over time, we found that, of the tumor cells that remained alive during the imaging period (n = 869), only 13% were associated with CTLs (Figure (Figure55 and Supplemental Movie 3). In contrast, 92% (n = 13) of the tumor cells that underwent apoptosis during the imaging period (as detected by FRET loss) were stably engaged by at least 1 CTL (Figure (Figure55 and Supplemental Movie 4). This observation provides additional evidence that CD8+ T cell–mediated cytotoxicity accounted for the bulk of tumor cell destruction during T cell adoptive therapy. Finally, we tracked 129 stable interactions between CTLs and live tumor cells and enumerated the number of killing events (as detected by FRET loss in individual tumor cells) as a function of the cumulated time of imaging. Using this approach, we estimated the rate of cell killing to be 1 tumor cell every 6 hours per CTL (Figure (Figure55 and Movie 4). In summary, these results demonstrate that CTLs directly mediate tumor elimination, but require, on average, several hours to kill 1 target cell.
Finally, we asked whether an impaired T cell cytotoxic activity accounted for the inefficient antitumor response mounted by naive antigen-specific CD8+ T cells. To this end, we visualized the response of adoptively transferred naive OT-I T cells in EG7/EL4 chimeric tumors. Although CTL infiltration was quite variable in the different regions of the tumor (Figure (Figure6A),6A), EG7 patches were eliminated in CTL-rich areas, which was evidence that in vivo primed CTLs were not grossly impaired in their ability to kill target cells (Figure (Figure6A,6A, right). A similar conclusion was reached when we visualized the apoptosis of EG7-DEVD tumors induced by in vivo primed OT-I T cells, because we detected a close association between OT-I CTLs and apoptotic tumor cells (Figure (Figure6B).6B). Thus, the low level of CD8+ T cell infiltration, rather than a defect in the cytotoxic activity, appeared to be responsible for the inefficient response mounted by in vivo primed OT-I T cells.
Whether CTLs or other host effector cells are responsible for most of the killing during adoptive T cell therapy is controversial. Here, we used 2 novel and independent strategies to demonstrate that CTLs, but not innate effectors, were responsible for the bulk of tumor cell destruction in a model of T cell therapy (Figure (Figure44 and Figure Figure6).6). Tumor killing occurred with minimal bystander activity (Figure (Figure4)4) — a feature that may play a role in the emergence of antigen-loss variants seen after T cell therapy in some clinical trials (23). Thus, in our system, tumor regression occurred with each individual tumor cell being eliminated by CTLs. The primary role of CTLs as cytotoxic effectors does not preclude an indirect contribution of other immune cells locally. For example, evidence for intratumoral T cell–macrophage interactions have been reported (19) and could possibly enhance CTL effector functions. In addition, other cells, such as stromal cells, could be targeted as well by CTLs during this process, as reported previously (9).
In some cases, isolated CTLs were found in contact with 2 apoptotic tumor cells, which suggests that serial (24) or simultaneous (25) killing previously described in vitro may also occur in vivo. The strong enrichment for granzyme-positive CTLs at the tumor site (Figure (Figure22 and Supplemental Figure 1) suggests that the perforin/granzyme pathway may contribute to tumor cell killing. The important role of CTL cytotoxicity revealed here by imaging approaches was surprising given that the transfer of perforin–/– FasL–/– OT-I CTLs was shown previously to efficiently induce regression of EG7 tumors (11). This observation might be explained by the redundancy of CD8+ T cell cytotoxic pathways. Alternatively, a distinct mechanism for tumor regression may be involved when CD8+ T cells are experimentally prevented to kill target cells. Thus, although our approach does not discriminate between the various CTL killing pathways, it provides, to our knowledge, the first direct assessment of intratumoral CD8+ T cell cytotoxic activity.
Recent reports have used 2-photon microscopy to depict intratumoral T cell migration, and one study reported an example of tumor cell disintegration following interaction with a T cell. To extend these observations and characterize the dynamics of CTL killing in tumors, we combined the use of intravital 2-photon imaging and that of a fluorescent reporter of caspase 3 activity. Interestingly, we estimated the killing rate (per CTL) to be 1 tumor cell every 6 hours. Some heterogeneity in the killing ability of CTLs may exist and could not be fully appreciated here because imaging periods were limited to 1–2 hours. However, CTL-target cells almost always remained in interaction during the entire imaging period, confirming that during this process, CTLs are occupied for hours, not minutes. In addition, the period during which CTLs are sequestered by tumor cells often extended beyond the killing, because CTLs remained in contact with apoptotic tumor cells for up to several hours (Supplemental Movies 2 and 5).
The estimated length required for a CTL to kill a tumor cell was much longer than that described for CTLs to kill other target cells in vivo. Two-photon imaging in lymph nodes showed that CTL killing of peptide-pulsed B cells and subsequent CTL detachment from the target cell requires less than 25 minutes (26). Consistent with a rapid killing kinetic in vivo, the half-life of adoptively transferred peptide-pulsed splenocytes was found to be approximately 1 hour in lymphocytic choriomeningitis virus–immune animals (27). These differences suggest that factors such as the tumor microenvironment and/or the cell type of the target (tumor cell versus splenocytes) may influence the rate of killing. Defects in TCR proximal signaling among intratumoral CD8+ T cells have been documented and could also possibly account for these differences (28). In the case of adoptive therapy with activated CD8+ T cells, the relatively slow rate of killing was compensated by an elevated density of infiltrating CTLs. In other contexts, however, the kinetics of CTL-tumor cell lysis could represent a limiting factor for tumor rejection and could account for the dose- and time-dependent efficiency of adoptive T cell therapy (29, 30). This view is further supported by a pattern of response mounted by naive CD8+ T cells. We found that naive antitumor CD8+ T cells were primed in vivo, infiltrated the tumor, and exerted cytotoxicity in situ. Yet, they failed to eliminate the tumor, most likely because their intratumoral density was too low, which left many areas of the tumor free of CTLs. Thus, whereas tumors use many mechanisms to prevent CTL-mediated cytotoxicity, our results provide evidence that, in some cases, functional CTLs may be present at the tumor site but in insufficient numbers to induce tumor regression. The presence of a low-to-moderate number of functional CTLs could possibly contribute to a temporary state of equilibrium between the tumor and the immune response (31).
In summary, we provided direct evidence that CTLs can mediate most of the tumor cell killing during adoptive T cell therapy in situ, but at a relatively slow rate. Achieving a high and relatively uniform density of functional CTLs in the tumor could therefore be critical to counterbalance this limitation during T cell–based cancer immunotherapeutic strategies. As illustrated in the present study, real-time imaging tumor cell apoptosis in live animals should provide new opportunities for monitoring antitumor immune responses and for dissecting their modulation by the tumor microenvironment.
C57BL/6 (B6) mice were purchased from Charles River Laboratories. Transgenic mice expressing the enhanced GFP under the human ubiquitin C promoter were purchased from the Jackson Laboratory and crossed to OT-I TCR transgenic mice. Ubiquitin C–GFP × OT-I TCR transgenic mice were bred in our animal facility. All animal experiments were performed according to institutional guidelines for animal care and use.
EL4 and EG7 tumor cell lines were purchased from the American Type Culture Collection. Constructs encoding mYFP or mCFP molecules (Clontech) or the CFP-DEVD-YFP apoptosis sensor (kindly provided by J.M Tavare, University of Bristol, Bristol, United Kingdom) were cloned in hygro pcDNA3.1 (Invitrogen). EL4 or EG7 tumor cells were transfected by electroporation, grown in complete RPMI in the presence of the appropriate selection agent, sorted using a FACS Aria (Becton Dickinson), and cloned. One clone was then selected for each transfected cell line. Tumor cells were harvested at exponential phase, and 2 × 106 tumor cells were resuspended in 50 μl of PBS and injected s.c. in the leg. Tumor volumes were recorded every 2–3 days.
CD8+ T cells were purified from the lymph nodes and spleens of OT-I TCR Tg or Ubiquitin C–GFP × OT-I TCR Tg mice using the depleting CD8 isolation kit (Miltenyi). For in vitro activation, OT-I T cells were stimulated using anti-CD3/anti-CD28 coated beads (Dynal) in the presence of 25 U/ml of recombinant human IL-2. Recipient mice were adoptively transferred with 5 × 106 purified naive or in vitro activated OT-I CD8+ T cells.
Tumors were excised, fixed in periodate-lysine-paraformaldehyde, progressively dehydrated in sucrose gradients ranging from 10% to 30%, and then frozen in OCT compound (Tissu-Tek Sakura Finetek, Europe). Tissue sections (8-μm thick) were rehydrated and blocked with normal mouse serum and anti-CD16 mAb (eBioscience) in the presence of 1% Triton X100 (Sigma-Aldrich). Tissue sections were stained with biotin-conjugated anti-PECAM mAb (RMA) or with APC-conjugated anti-granzyme B mAb (Caltag). Sections were mounted using the Vectashield medium (Vector Laboratories) and then imaged using a confocal microscope (Zeiss). Images were processed using ImageJ software version 1.38.
Tumors and lymph nodes were digested at 37°C in RPMI 1640 containing 1 mg/ml collagenase and 0.05 mg/ml DNAse. Single-cell suspensions were subjected to intracellular staining using the Cytofix/cytoperm kit (BD Pharmingen) and either APC-conjugated anti-granzyme B mAb (Caltag) or Alexa fluor 647-conjugated anti-mouse IFN-γ mAb (clone XMG1.2; Becton Dickinson Biosciences). Cells were then analyzed with a Cyan ADP (DAKO). Data were analyzed using FlowJo software version 8.6.
Mice were anesthetized and prepared for intravital imaging. Briefly, the tumor was surgically exposed and the mouse was placed on a custom-designed heated stage. To immobilize the region of interest, plaster bandages were placed on each side of the posterior leg. A coverslip was placed on top of the tumor and glued onto the plaster cast. The temperature was maintained at 37°C by a heated metal ring placed onto the coverslip and filled with water to immerge a 20X/0.95 NA dipping objective (Olympus). Two-photon imaging was performed using an upright microscope (DM 6000B) with an SP5 confocal head (Leica Microsystems). Excitation was provided by Chameleon Ultra Ti:Sapphire laser (Coherent) tuned at 880 or 940 nm. Emitted fluorescence was split using a 495-, 510-, or 560-nm dichroic mirror and passed through 465/30, 525/50, or LP560 filters (Chroma Technology) to nondescanned detectors (Leica Microsystems). Typically, 7 z-planes spaced 5 μm apart and located at least 100 μm below the tumor surface were imaged every 60 seconds. Movies were further processed using Imaris (version 5.7; Bitplane) and ImageJ software.
The apoptosis index represents the ratio of signal detected in the CFP channel to that detected in the YFP channel after normalization. In each experiment, the average value for 5 live tumor cells was used to normalize the apoptosis index to 1.
We compiled all interactions (n = 179) between CTLs and live tumor cells observed using 2-photon imaging of mice bearing EG7-DEVD tumors and adoptively transferred with activated OT-I T cells. Overall, these individual interactions represented a cumulative time of imaging of 74 hours and 41 minutes, with an average duration of imaging of 35 minutes. A total of 12 EG7 tumor cells were seen undergoing apoptosis during the course of the imaging period (as detected by FRET disruption), which indicated that, on average, 1 tumor cell killing was detected per 6 hours of CTL-tumor cell interaction (killing rate = cumulated time of imaging/number of killing events). This calculation assumes that the kinetics of killing remain relatively constant during tumor regression.
Data are presented as mean ± SEM. Statistical analysis was performed using the Mann-Whitney U test. A P value less than 0.01 was considered significant.
We thank J.M. Tavare for kindly providing the CFP-DEVG-YFP and CFP-DEVG-YFP constructs; M. Albert, E. Robey, and J. Di Santo for helpful comments on the manuscript; and the Plate-forme de Cytométrie and the Plate-forme d’imagerie Dynamique, Institut Pasteur. This work was supported by Institut Pasteur, INSERM, Mairie de Paris, Fondation de France, and Association pour la Recherche sur le Cancer and by a Marie Curie Excellence grant.
Nonstandard abbreviations used: FRET, Förster resonance energy transfer; mCFP, membrane-targeted cyan fluorescent protein; mYFP, membrane-targeted yellow fluorescent protein.
Conflict of interest: The authors have declared that no conflict of interest exists.
Citation for this article: J. Clin. Invest. 118:1390–1397 (2008). doi:10.1172/JCI34388