Suboptimal T-cell Therapy Drives a Tumor Cell Mutator Phenotype That Promotes Escape from First-Line Treatment

Tumor cells escape from a variety of first-line therapies through multiple genetic and epigenetic mechanisms that are derived from the genomic instability inherent to cancer. This genomic instability confers phenotypic plasticity and allows for the selection and expansion of therapeutically resistant tumor cell subclones. Dysregulation of the APOBEC3 family of cytosine deaminase enzymes represents an endogenous source of DNA damage that mediates the evolution of cancer (1, 2). Originally identified as innate antiviral restriction factors (3), the APOBEC3 enzymes catalyze cytosine to uracil deamination of ssDNA to generate C to T transitions and, to a lesser frequency, C to G transversion mutations. APOBEC3A and B characteristic mutations exist in approximately half of all human cancers and correlate with APOBEC3A and B overexpression (1, 4, 5). The APOBEC3B mutation signature has been associated with poor prognosis, and may be associated with therapeutic resistance in multiple cancer types (6–8). For estrogen receptor–positive breast cancer, APOBEC3B expression inversely correlated with tamoxifen benefit both preclinically and clinically (7). APOBEC3B overexpression has been attributed to protein kinase C (PKC)-mediated activation of the NF-κB pathway (9, 10), however, it remains unclear which cell types or signals in the tumor microenvironment modulate APOBEC3B expression in tumor cells.

Although the human genome encodes seven distinct APOBEC3 enzymes (APOBEC3A through H), the mouse genome encodes a single gene, mAPOBEC3 (11, 12). It is unclear which of the human APOBEC3 activities in mAPOBEC3 mimics with respect to inducing cancer-promoting mutations. In this study, we have investigated whether mAPOBEC3 may have similar activities to those of human APOBEC3B (hAPOBEC3B) as an inducer of the tumor cell heterogeneity that contributes to cancer evolution.

Although radiation and chemotherapy lead to direct killing of tumor cells, their capacity to induce DNA damage can introduce mutations that may also promote resistance. Several evolutionary mechanisms to evade adoptive T-cell therapy, oncolytic viroimmunotherapy, and suicide gene therapy have been described in murine models (13–15).

Current views hold that generation of a tumor-specific adaptive immune response of any magnitude is beneficial to the patient, even if the tumor-reactive T cells are at low frequency or bear low affinity T-cell receptors for their cognate tumor antigens. Herein however, we show that the effector cytokine TNFα produced by tumor-reactive T cells acts on target tumor cells by upregulating the mAPOBEC3 enzyme, which can lead to an increased resistance to therapy. When tumor-reactive T cells were cocultured with tumor cells at low effector-to-target (E:T) ratios in the presence of additional selective pressures such as an oncolytic virus or suicide gene therapy, mAPOBEC3 upregulation increased the rate of treatment-resistant clonal outgrowth. In an immunocompetent mouse model of herpes simplex virus (HSV)-thymidine kinase (TK)–driven suicide gene therapy, hAPOBEC3B overexpression promoted accumulation of silencing mutations in the TK gene and reduced median survival. We conclude that suboptimal immunotherapies indirectly drive tumor cell mutations, in the same way as other mutagenic therapies such as radiation and chemotherapy, and should be optimized to prevent enhancement of tumor cell escape.

B16 murine melanoma cells were obtained from the ATCC prior to being modified with the relevant transgenes. Cell lines were authenticated by morphology, growth characteristics, PCR for melanoma-specific gene expression (gp100, TYRP-1, and TYRP-2), and biologic behavior, tested Mycoplasma-free, and frozen. Cells were cultured for less than 3 months after thawing. The B16OVA cell line was derived from a B16.F1 clone transfected with a pcDNA3.1ova plasmid obtained from Dr. Esteban Celis (Augusta University) in 2000 (16, 17). B16OVA cells were grown in DMEM (HyClone) + 10% FBS (Life Technologies) + 5 mg/mL G418 (Mediatech) until challenge. GL261 cells were obtained from Dr. Aaron Johnson (Mayo Clinic, Rochester, MN) in 2014. GL261OVA was obtained by transfection of parental GL261 cells with pcDNA3.1 OVA in 2015. LLC cells were obtained from Professor Ian Hart (ICRF, London, United Kingdom) in 1998. LLCOVA was obtained by transfection of parental LLC cells with pcDNA3.1 OVA in 2015. B16TK cells were derived from a B16.F1 clone transfected with a plasmid expressing the HSV-1 TK gene in 1997/1998 (18). Following stable selection in 1.25 μg/mL puromycin, these cells were shown to be sensitive to ganciclovir (Cymevene) at 5 μg/mL (19–21). Cells were tested for Mycoplasma using the MycoAlert Mycoplasma Detection Kit (Lonza Rockland, Inc.).

Six to 8-week-old female C57BL/6 mice were purchased from Jackson Laboratories. The OT-I mouse strain is on a C57Bl/6 background and expresses a transgenic T-cell receptor Vα2/Vβ5 specific for the SIINFEKL peptide of ovalbumin in the context of MHC class I, H-2K^b as described previously (22) and were bred at Mayo Clinic. The Pmel mouse strain is on a C57Bl/6 background and expresses a transgenic T-cell receptor Vα1/Vβ13 that recognizes amino acids 25–33 of gp100 presented by H2-D^b and were bred at Mayo Clinic.

Wild-type Reovirus type 3 (Dearing strain) was obtained from Oncolytics Biotech and stock titers were measured by plaque assay on L929 cells.

B16TK cells were seeded in 96-well plates in triplicate and treated with reovirus [multiplicity of infection (MOI) 0.1] or with ganciclovir (Cymevene) at 5μg/mL. CellTiter blue (Promega) was added to wells at 10% v/v and fluorescence was measured after approximately 4 hours incubation (560_Ex/590_EM). Relative viability of experimental conditions was normalized to untreated cells.

Spleens were immediately excised from euthanized C57Bl/6, OT-I mice and dissociated in vitro to achieve single-cell suspensions. Red blood cells were lysed with ACK lysis buffer. CD8⁺ T cells were prepared using the CD8α T Cell Isolation Kit (Miltenyi Biotech) and cocultured with target tumor cells at various E:T ratios as described in the text. Supernatants were assayed for TNFα and IFNγ by ELISA as directed in the manufacturer's instructions (Mouse TNFα or Mouse IFN-γ ELISA Kit, OptEIA, BD Biosciences).

OT-I or Pmel T cells were activated in IMDM (Gibco) + 5% FBS + 1% penicillin/streptomycin + 40 μmol/L 2-ME. Media was supplemented with the SIINFEKL or KVPRNQDWL peptides, respectively, at 1 μg/mL and human IL2 at 50 U/mL. Cells were used for in vitro assays following 4 days of activation.

CD8⁺ T cells were prepared as described above from C57BL/6 mice that had been cured of subcutaneous B16TK tumors following three weekly courses of ganciclovir (50 mg/kg on days 5–9, 12–16, and 19–23). Cells were harvested between 60 and 80 days post tumor implantation.

B16TK or B16OVA cells were plated in triplicate wells in the presence of ganciclovir (Cymevene) at 5μg/mL, reovirus (MOI 0.1) or 4-day in vitro–activated OT-I CD8⁺ T cells, or tumor-experienced (T.E.) CD8⁺ T cells (E:T ratio of 5:1) for 7 days in Iscove's Modified Dulbecco's Medium (IMDM; Gibco) + 5% FBS + 1% penicillin–streptomycin + 40 μmol/L β-mercaptoethanol. Wells were washed three times with PBS and cultured in normal medium for a further 7 days. Surviving cells were then cultured again in the presence of PBS, ganciclovir, reovirus (MOI 0.1) or 4-day in vitro–activated OT-I CD8⁺ T cells, or T.E. CD8⁺ T cells (various E:T ratios) for 7 days.

Tumor cells were treated with phorbol 12-myristate 13-acetate (PMA, 25 ng/mL). These coculture systems were also performed with anti-H-2K^b (AF6-88.5, 0.5 μg/mL; BioLegend), the inhibitor of PKC signaling (AEB071, 10 μmol/L; MedChemExpress) or anti-TNFα (AF-410-NA, 0.5 μg/mL; R&D Systems) or anti-IFNγ (MAB485, 0.5 μg/mL; R&D Systems).

RNA was prepared with the QIAGEN-RNeasy-MiniKit (Qiagen). One-microgram total RNA was reverse transcribed in a 20 μL volume using oligo-(dT) primers using the First Strand cDNA Synthesis Kit (Roche). A cDNA equivalent of 1 ng RNA was amplified by PCR with gene-specific primers using GAPDH as loading control (mgapdh sense: TCATGACCACAGTCCATGCC; mgapdh antisense: TCAGCTCTGGGATGACCTTG; and APOBEC3 sense: ATGGGACCATTCTGTCTGGGA; APOBEC3 antisense: TCAAGACACGGGGGTCCAAG). qRT-PCR was carried out using a LightCycler480 SYBRGreenI Master Kit and a LightCycler480 Instrument (Roche) according to the manufacturer's instructions. The ΔΔC_t method was used to calculate the fold change in expression level of APOBEC3 and GAPDH as an endogenous control for all treated samples relative to an untreated calibrator sample.

The OVA transgene was sequenced using the following primers:

Sense: ATGGGCTCCATCGGCGCAGC and antisense: CCGTCTACACAAAGGGGAATT and aligned to the reference sequence CAA23682.1. The HSV-TK transgene was sequenced using the following primers: CACGCAGATGCAGTCGGGGCGGCG (downstream of the EcoR1 site in the 5′UTR), CTGGTGGCCCTGGGTTCGCGCGA, GCGTTCGTGGCCCTCATCCC, GCCTGGGCCTTGGACGTCTTGG, and AGGGCGCAACGCCGTACGTCG and aligned to the reference sequence AB009254.2.

Murine APOBEC3 was measured by Western blotting with a rabbit polyclonal antibody (PA511430, Thermo Fisher Scientific) or rabbit monoclonal anti-human APOBEC3B (184990, Abcam) which react with both human APOBEC3B and murine APOBEC3 (Thermo Fisher Scientific) or by ELISA according to the manufacturer's instructions (Antibody Research Corporation). B16TK cells were treated with recombinant murine TNFα (R&D Systems). HSV-TK protein was detected by Western blotting tumor cell lysates with a goat polyclonal antibody (28038; Santa Cruz Biotechnology). β-Actin was detected using an horseradish peroxidase–conjugated mouse mAb (clone AC-15; Sigma).

Four separate mice with unique 29mer short hairpin RNA (shRNA) retroviral constructs (OriGene Technologies) were tested individually, or as a combination, for their ability to reduce expression of murine APOBEC3 in B16 cells compared with a single-scrambled shRNA encoding retroviral construct. Optimal knockdown for periods of more than 2 weeks in culture was achieved using all four constructs prepackaged as retroviral particles in the GP+E86 ecotropic packaging cell line and used to infect B16 cells at an MOI of approximately 10 per retroviral construct. In addition, a single-scrambled negative control noneffective shRNA cassette was similarly packaged and used to infect B16TK cells to generate B16TK (scrambled shRNA) cells.

B16TK cells were infected with a retroviral vector encoding either full length functional APOBEC3B or a mutated, nonfunctional form of APOBEC3B as a negative control obtained from Reuben Harris (University of Minnesota, Minneapolis, MN). Infected populations were selected for 7 days in Hygromycin to generate B16TK (APOBEC3B) or B16TK (APOBEC3B MUT) cell lines and used for experiments as described. In populations of B16TK (APOBEC3B) cells selected for more than 7–10 days in Hygromycin expression of APOBEC3B returned to basal levels associated with the toxicity of prolonged APOBEC3B expression. Murine APOBEC3 (Accession no. BC003314) was expressed from the pCMV-SPORT6 plasmid obtained from Dharmacon).

All in vivo studies were approved by the Institutional Animal Care and Use Committee at Mayo Clinic. Mice were challenged subcutaneously with 2 × 10⁵ B16TK melanoma cells, in 100 μL PBS (HyClone) or with 1 × 10⁴ cells in 2 μL intracranially into the frontal lobe as reported previously (23). Subcutaneous tumors were treated with a 2- or 3-week course of ganciclovir (50 mg/kg) administered intraperitoneally daily. Tumors were measured three times per week, and mice were euthanized when tumors reached 1.0 cm in diameter. Intracranial tumors were treated with a 3-week course of ganciclovir (50 mg/kg) administered intraperitoneally on days 6, 8, 10, 13, 15, 17, 20, 22, and 24. Mice were sacrificed upon emergence of neurological symptoms or weight loss.

Survival curves were analyzed by the log-rank test. Student t tests, one-way ANOVA, and two-way ANOVA were applied for in vitro assays as appropriate. Statistical significance was set at P < 0.05 for all experiments.

B16 cells expressing the HSV-1 TK (B16TK) were sensitive to treatment with ganciclovir at 5 μg/mL (Fig. 1A) or reovirus at an MOI of 0.1 (Fig. 1B), however, despite the induction of significant cell death, a small proportion of cells survived as escape variants following two consecutive weekly cycles of treatment with ganciclovir or reovirus (Fig. 1C and D). We have previously shown that clearance of B16TK tumors by ganciclovir in immunocompetent mice is dependent upon CD8⁺ T cells, and that tumor-cured mice have CD8⁺ T-cell responses against parental B16 cells (19–21). Purified CD8⁺ T cells from mice that had rejected B16TK tumors following ganciclovir therapy (T.E. CD8⁺ T cells) killed target B16TK cells and produced low concentrations of IFNγ in vitro (Fig. 1E). We therefore reasoned that the combination of ganciclovir, or reovirus, with T.E. CD8⁺ T cells would lead to enhanced cumulative cell killing. However, when purified T.E. CD8⁺ T cells were cocultured with B16TK cells at the time of treatment with ganciclovir or infection with reovirus, we observed a significant increase in the number of B16TK cells that survived compared with those treated in the absence of T cells. In contrast, we did not observe this enhanced outgrowth when B16TK cells were treated with ganciclovir or reovirus in the presence of CD8⁺ T cells purified from naïve C57/BL6 mice (Fig. 1F). We confirmed that the cytotoxicity and IFNγ section of the T.E. CD8⁺ T cells cocultured with parental B16 cells was not compromised by ganciclovir. (Supplementary Fig. S1A). In addition, the activation of OT-I CD8⁺ T cells was increased in the presence of reovirus (Supplementary Fig. S1B) showing that the virus does not diminish CD8⁺ T-cell function, eliminating one possible explanation for the increased survival of the target cells.

To investigate this phenomenon in a model with a defined antigenic target, we evaluated the potential of B16OVA cells to escape therapy when cocultured with in vitro–activated CD8⁺ OT-I T cells. At high E:T ratios (50:1 and 10:1), no surviving escape colonies of B16OVA were observed (Supplementary Table S1). At lower ratios (5:1 and 1:1) individual colonies of B16OVA cells could be isolated as escape variants; these colonies were resistant to further OT-I killing, even at high E:T ratios (Supplementary Fig. S2). We then introduced either naïve or T.E. CD8⁺ T cells into the OT-I CD8⁺ T cell B16OVA coculture system at an T.E. effector to OT-I E:T ratio of 10:10:1, and observed escape B16OVA clones emerge (Table 1, timeline outlined in Fig. 2A). In contrast, when the T.E. CD8⁺ T cells were replaced with naïve CD8⁺T cells, all target cells were killed. We isolated and expanded 15 clones from the T.E. CD8⁺ and OT-I T-cell coculture condition. Ten of these clones showed complete loss of the ova gene, consistent with our previous findings (14, 16); however, five escape B16OVA clones retained the ova gene. Sequencing revealed that four of five clones contained a TC to TT conversion in two locations (positions 406 and 457), both of which generated a premature stop codon upstream of the immunodominant MHC class I–binding SIINFEKL epitope (Fig. 2B). The B16OVA cells used in the experiment of Fig. 2 and Table 1 were originally derived from a single-cell clone of B16OVA with a sequenced ova gene and selected for strong recognition by OT-I T cells. Taken together, these data suggest that these OVA mutant–containing B16OVA cells were selected for by a gain of mutation induced in the OT-I/T.E. CD8⁺ T cell cocultures.

The first hotspot for C to T transition mutation in the ova gene was consistent with the previously reported murine APOBEC3 motif TXC. Both hotspots were consistent with that of the APOBEC3B cytosine deaminase with an A in the +1 position (TCA; refs. 4, 12, 24–26). We hypothesized that T-cell interaction may induce an equivalent murine APOBEC3B-like activity in tumor cells, which could promote the mutations that allow for escape from therapy. We therefore evaluated the expression of mAPOBEC3 by qRT-PCR in tumor cells following coculture with tumor-reactive T cells at E:T ratios at which escape variants were observed.

mAPOBEC3 mRNA expression rose sharply after 12 hours of coculture with OT-I or T.E. CD8⁺ T cells, as well as following treatment with the PKC activator PMA (ref. 9; Fig. 3A). Similarly, mAPOBEC3 protein was induced in B16OVA cells at suboptimal E:T ratios with OT-I CD8⁺ T cells, but not at a high E:T ratio (Fig. 3B), consistent with the outgrowth of escape variants (Table 1). This same effect was observed in B16TK cells cocultured with T.E. CD8⁺ T cells at low E:T ratios, with maximal upregulation of mAPOBEC3 at a ratio of 10:1 (Fig. 3C). The different T.E. CD8⁺ and OT-I E:T ratios required for maximal mAPOBEC3 induction likely reflects the lower frequency of antigen-specific T cells in the T.E. CD8⁺ cell population. We also confirmed the upregulation of mAPOBEC3 by Western blot analysis (Fig. 3D).

Upregulation of APOBEC3 in tumor cells at suboptimal E:T ratios inversely correlated with the secretion of the effector cytokine IFNγ (Fig. 3E). Coculture of CD8⁺ OT-I cells with B16OVA cells at an E:T ratio of 10:1 induced robust IFNγ production but little mAPOBEC3 expression. As the ratio was reduced, APOBEC3 expression rose. Naïve OT-I cells that had not previously been activated in vitro produced little IFNγ when cocultured with B16OVA cells, but stimulated mAPOBEC3 upregulation in target cells. Conversely, naïve OT-I cells cocultured with B16OVA cells in the presence of exogenous SIINFEKL peptide produced high IFNγ and did not induce mAPOBEC3 expression. High E:T ratios were also associated with high TNFα secretion from OT-I T cells (Fig. 3F), which decreased as the number of activated T cells was reduced.

To show the relationship between the E:T ratio of T-cell killing, TNFα levels, and mAPOBEC3 induction, we used a transwell coculture system, in which B16OVA target tumor cells were cocultured with effector OT-I T cells in the top chambers, and (bystander) B16OVA cells were plated in the bottom chambers (Fig. 3G). In the top chamber, at the E:T ratio of 10:1, most of the target B16OVA tumor cells were killed (Fig. 3H), TNFα was detected (Fig. 3I), and thus mAPOBEC3 could not be measured because insufficient cells survived (Fig. 3J). At the lower E:T ratios, direct T-cell–mediated tumor cell killing was diminished (Fig. 3H), and APOBEC3 expression in the surviving cells was elevated (Fig. 3J). Bystander B16OVA tumor cells not directly exposed to T cells in the bottom chambers were not killed (Fig. 3H). However, as a result of the T-cell activity in the top chambers, TNFα was nonetheless detected in the bottom chambers (Fig. 3I), and was associated with induction of mAPOBEC3 in the bystander B16OVA cells (Fig. 3J).

Those bystander B16OVA cells that survived in the bottom chambers following exposure to TNFα, and in which mAPOBEC3 had been induced (Fig. 3H–J), were significantly more resistant to killing by OT-I T cells when replated in fresh cocultures than were parental B16OVA cells (Fig. 3K). The bystander B16OVA cells recovered from the bottom chambers of the E:T 10:1 cocultures were the most resistant to subsequent OT-I cell killing (Fig. 3K). These bystander B16OVA cells had been exposed to the highest levels of TNFα as a result of T-cell activation and killing in the top chambers (Fig. 3I, bottom chambers) and had the greatest mAPOBEC3 overexpression (Fig. 3J, bottom chambers). Resistance to OT-I T-cell killing in these bystander B16OVA cells was equivalent to that induced by de novo overexpression of mAPOBEC3 in B16OVA cells (Fig. 3K and L), suggesting that mAPOBEC3 mediates T-cell–induced, bystander tumor cell escape from therapy. Thus, when ineffective T-cell killing clears only some tumor cells, the remaining bystander tumor cells can upregulate mAPOBEC3 expression and acquire mutations that confer a selective advantage.

We also confirmed that a low E:T ratio of tumor antigen–specific T cells to tumor cells induced mAPOBEC3 not only in the B16 cell line but also in both the GL261 glioma and the lung carcinoma LLC cell lines (Supplementary Fig. S3A and S3B). Taken together, these data show that there is a threshold where suboptimal T-cell activation and limited effector function induces APOBEC3 upregulation in tumor cells.

mAPOBEC3 induction and tumor cell outgrowth from ganciclovir and T-cell therapy was dependent on MHC class I recognition of tumor cells, TNFα secretion, and activation of PKC signaling (9), as antibody blockade of H-2K^b, TNFα, or pharmacologic inhibition of PKC ablated the effect (Fig. 4A and B). In contrast, IFNγ blockade had no significant effect on tumor cell mAPOBEC3 expression following CD8⁺ T-cell coculture (Fig. 4C). mAPOBEC3 induction in B16TK cells was not affected by the DMSO solvent or by the IgG control antibody (Supplementary Fig. S4A). None of the treatments in Fig. 4A significantly inhibited the growth of B16TK cells alone (Supplementary Fig. S4B). Consistent with a role for TNFα in T-cell–mediated mAPOBEC3 induction in tumor cells, exogenous TNFα was sufficient to induce mAPOBEC3 (Fig. 4D and E), an effect which was almost completely inhibited by AEB071 action upon the tumor cells themselves (Fig. 4E). Although the cytotoxicity of the T.E. CD8⁺ T cells was not significantly altered in the presence of AEB071 compared with PBS, the levels of IFNγ were significantly decreased in the presence of AEB071 (Supplementary Fig. S4C). Therefore, the attenuated induction of mAPOBEC3 seen in the presence of T.E. CD8⁺ T cells and AEB071 in Fig. 4A may indeed be the result of reduced induction of APOBEC3 in the tumor cells through inhibition of PKC signaling (Fig. 4E), and/or the result of partial inhibition of CD8⁺ T-cell function by AEB071.

To confirm that mAPOBEC3 was required for the outgrowth of escape variants in this model, we generated a stable B16TK cell line expressing four unique 29mer shRNA constructs targeting mAPOBEC3, as well as a stable B16TK cell line with a single scrambled shRNA construct. mAPOBEC3 expression was significantly reduced in B16TK shRNA mAPOBEC3 cells, both at basal levels, and upon induction with PMA (Fig. 5A). In cultures not treated with CD8⁺ T cells or cultures that received naïve CD8⁺ T cells, fewer B16TK sh mAPOBEC3 cells survived either with ganciclovir or reovirus treatments compared with parental B16TK, or B16TK (scrambled shRNA) cells (Fig. 5B and C). These data are consistent with a role for mAPOBEC3 expression in mediating mutagenesis that facilitates escape from either of these therapies. We did not observe a statistically significant increase in the number of surviving B16TK sh mAPOBEC3 cells treated with ganciclovir or reovirus when cells were cocultured with T.E. CD8⁺ T cells (Fig. 5B and C). B16TK parental or B16TK-scrambled shRNA cells treated with ganciclovir and cocultured with T.E. CD8⁺ cells exhibited the pattern of enhanced escape, in contrast to coculture with naïve CD8⁺ T cells. These results were recapitulated with reovirus infection, where more treatment-resistant clones arose from parental B16TK and B16TK cells with scrambled shRNA than from B16TK sh mAPOBEC3 cells cocultured with T.E. CD8⁺ T cells (Fig. 5C). This same pattern of escape that was prevented with shRNA to mAPOBEC3 was also seen when B16TK cells were cocultured with activated Pmel T cells cultured in the presence of ganciclovir (Fig. 5D). These results were not attributable to different rates of cell growth in vitro (Supplementary Fig. S5A). Similarly, more treatment-resistant clones could be isolated from parental B16OVA and from B16OVA cells with scrambled shRNA than from B16OVA sh mAPOBEC3 cells cocultured with OT-I CD8⁺ T cells (Supplementary Fig. S5B).

These results were validated in vivo where B16TK cells transduced with the scrambled shRNA or the mAPOBEC3-targeting shRNA were implanted subcutaneously and treated with a suboptimal course of ganciclovir. Although B16TK (scrambled shRNA) tumors all eventually escaped therapy (0/7 long-term survivors; Fig. 5E), recurrence was delayed or did not occur in B16TK tumors transduced with mAPOBEC3 shRNA (4/7 long-term survivors; Fig. 5F). These results could not be attributed to different growth rates in vivo, as untreated tumors that expressed either scrambled shRNA or sh mAPOBEC3 grew at equivalent rates (Supplementary Fig. S5C). This confirms that mAPOBEC3 expression after suboptimal therapies can contribute to the generation of treatment- resistant clones in vivo.

As knockdown of mAPOBEC3 significantly reduced the ability of B16TK cells to escape ganciclovir treatment (Fig. 5B–F), we evaluated how overexpression of mAPOBEC3 could promote therapeutic escape in the 21-day treatment cycle (Fig. 5G). The T-cell–mediated induction of a TCA-TTA mutation in the OVA gene, which allowed escape from OT-I T-cell therapy (Fig. 2), suggested that mAPOBEC3 functions may overlap those of its human APOBEC3B counterpart. Therefore we overexpressed hAPOBEC3B in B16TK cells. We observed that overexpression of either mAPOBEC3 or hAPOBEC3B promoted the outgrowth of treatment-resistant clones compared with unmodified B16TK cells. Moreover, overexpressed mAPOBEC3 or hAPOBEC3B rescued the phenotype induced by shRNA knockdown of APOBEC3 in B16TK cells treated with ganciclovir. In contrast, overexpression of a catalytically inactive form of the protein, hAPOBEC3B MUT, did not promote resistance. Although anti-mAPOBEC3 shRNA expressed in the B16TK (shRNA mAPOBEC3) cells was able to target the pCMV-APOBEC3 plasmid used for overexpression of mAPOBEC3 in these experiments, it was unable to prevent mAPOBEC3 induction, especially at early time points (Supplementary Fig. S6A). Taken together, these data confirm that mouse APOBEC3 activity, induced through T-cell–derived TNFα (Fig. 3), induces resistance to ganciclovir killing.

We further investigated the mechanism of ganciclovir escape by sequencing the bulk populations from each condition (Fig. 5D and G). B16TK parental cells that escaped ganciclovir treatment in vitro retained the wild-type HSV-TK sequence, suggesting that failure to eradicate these cells was not due to mutation of the therapeutic gene (Fig. 5H). In contrast, B16TK cells engineered to overexpress either hAPOBEC3B, or mAPOBEC3, both selected populations of cells in which the HSV-TK gene contained an ATCA-ATTA mutation at position 22 (Fig. 5H). This mutation introduced a stop codon in the first eight amino acids of the protein, thereby preventing expression of functional HSV-TK protein to maintain susceptibility to ganciclovir therapy. This mutation was not observed in the B16TK sh mAPOBEC3 cells, but the shRNA knockdown could be overwhelmed by the plasmid overexpression of mAPOBEC3 (Supplementary Fig. S6B). We also investigated whether the CD8⁺ T-cell–mediated mutator phenotype, which we have shown to be at least partly dependent upon APOBEC3 activity, was also associated with this mAPOBEC3 mutation signature by sequencing surviving cells from Fig. 5D. B16TK cells that survived the ganciclovir selection regimen during coculture with either no CD8⁺ T cells, or with naïve CD8⁺ T cells, showed no evidence of the APOBEC3-driven ATCA-ATTA mutation in the HSV-TK gene at either 7 or 21 days in the selection process (Fig. 5H), indicating that resistance to ganciclovir in these cells was not associated with this particular mechanism of escape. Ganciclovir -resistant B16TK cell populations selected in the presence of tumor-specific Pmel CD8⁺ T cells consisted of a mixed population of HSV-TK wild-type (ATCA) and HSV-TK-mutated (ATTA) cells 7 days after coapplication of both T cell and ganciclovir selective pressure and 2 further weeks of ganciclovir therapy forced the evolution of a clonal population of B16TK ESC cells with this mutation. In contrast, knockdown of mAPOBEC3 prevented induction and further selection/fixation of the mutation. These data were confirmed in two replicate experiments (Supplementary Fig. S6C). Taken together, these results show that suboptimal T-cell therapy can promote the emergence of resistance to a coapplied first-line therapy through an APOBEC3-driven mutational mechanism. These data (i) show that mAPOBEC3 has overlapping mutational specificity with hAPOBEC3B and (ii) provide a mechanism by which the acquired resistance to ganciclovir therapy that follows suboptimal (incompletely cytotoxic) T-cell activity can be attributed to mAPOBEC3 activity.

We evaluated the role of APOBEC3B in the acquisition of therapeutic resistance using the retroviral overexpression system of human APOBEC3B used in Fig. 5G. B16 tumor cell lines were engineered to overexpress hAPOBEC3B or the catalytically inactive hAPOBEC3B MUT. Forty-eight hours post transduction, bulk populations of cells were selected in Hygromycin for 2 weeks and used for experiments. Elevated levels of hAPOBEC3B are seen in cultures within 72 hours post transfection/infection; expression then returns to levels seen in parental unmodified cells. We believe that this is because mutagenesis by hAPOBEC3B is tolerable only within limits. Only cells that do not excessively overexpress hAPOBEC3B or generate mutations in critical genes will survive and carry hAPOBEC3B-induced mutations. The increased frequency of ganciclovir-resistant tumor cell outgrowth as seen when B16TK cells were cocultured with T.E. CD8⁺ T cells was recapitulated by the overexpression of hAPOBEC3B but not hAPOBEC3B MUT in B16TK cells (Fig. 6A). In parallel, hAPOBEC3B overexpression in B16TK cells increased the rate of outgrowth following reovirus infection (Fig. 6B). Thus, hAPOBEC3B function is sufficient to support therapeutic escape.

When implanted intracranially in a model of metastatic melanoma, B16TK-hAPOBEC3B tumors grew significantly more quickly following ganciclovir treatment compared with B16TK-hAPOBEC3B MUT tumors (Fig. 6C). These results could not be attributed to different growth rates of the tumors, as untreated tumors that overexpressed either hAPOBEC3B or mutated APOBEC3B grew at equivalent rates (Supplementary Fig. S6D). Western blot analysis of intracranial tumors recovered from mice showed that four of four B16TK-hAPOBEC3B MUT tumors continued to express the HSV-TK protein, despite eventually failing therapy, whereas zero of four B16TK hAPOBEC3B tumors continued to express HSV-TK protein (Fig. 6D). Sequencing of the HSV-TK gene identified a C to T mutation at position 22 that generated a premature stop codon as seen in Fig 5H in all four B16TK hAPOBEC3B tumors that was not observed in the parental B16TK cell line (Fig. 6E). None of the B16TK-APOBEC3B MUT tumors contained this mutation or any detectable characteristic APOBEC3 C to T mutation within the TCW motif in the HSV-TK gene. B16TK-hAPOBEC3B subcutaneous tumors also grew more quickly following ganciclovir treatment compared with parental B16TK or B16TK-hAPOBEC3B MUT tumors, even though untreated tumors grew at equivalent rates (Fig. 6F). These data support the hypothesis that hAPOBEC3B expression drives a mutator phenotype in tumor cells which allows for selection of a tumor cell clone that can resist treatment.

We began our study with the hypothesis that suboptimal treatments would be improved by combination with T-cell therapies in which tumor-reactive T cells were used as an adjunct to the first-line treatment. However, we report here a mechanism by which CD8⁺ T cells undergoing low E:T ratio interactions with tumor cells can promote an increase in the mutational burden of the tumor cells and drive escape from a coapplied therapy via induction of APOBEC3. Blocking MHC class I, TNFα secretion, or PKC signaling, or knocking down endogenous mAPOBEC3 expression negated the T-cell–induced mutator phenotype in vitro and in vivo. Consistent with the hypothesis that the single murine APOBEC3 protein may, among other properties, mimic the cancer-driving activity of the human APOBEC3B protein, overexpressing hAPOBEC3B in murine tumor cells enhanced escape from either suicide gene therapy or oncolytic virotherapy in vitro and in vivo.

Threshold levels of T-cell interaction with target cells are required for cell killing to occur (27). Thus, successful T-cell killing of target cells results from a combination of affinity of the TCR for the antigen–MHC complex, as well as the number of engagements. Because of the nature of self, or near-self, tumor-associated antigens, most therapies that induce antitumor T-cell responses either directly (vaccines and oncolytic viro-immunotherapy) or indirectly (chemotherapy and radiation) generate T cells with a low affinity TCR and/or low frequency T-cell responses. Despite the fact that the T.E. CD8⁺ T cells had a relatively low recognition of target B16TK cells, we still predicted that therapy with ganciclovir or reovirus virotherapy would enhance cumulative cell killing in combination with these tumor-reactive CD8⁺ T cells. However, we observed the opposite effect, in which weakly reactive CD8⁺ T cells promoted tumor cell acquisition of a mutator phenotype that enhanced their ability to acquire resistance to a first-line treatment of chemotherapy or virotherapy. The same effect could be mimicked using B16OVA cells and low E:T ratios of OT-I cells. Two recurrent mutations that were observed in the OVA coding sequence (C to T transitions) were consistent with the enzymatic signature of the APOBEC3 family of DNA deaminase enzymes and suggested a candidate effector mechanism that may play a role in tumor immunoediting. We hypothesize that evidence of APOBEC3 enzymatic activity in tumor cells cocultured with the polyclonal T.E. CD8⁺ T cells may be observed more broadly in the tumor genome, including genes involved in antigen processing and presentation. Indeed, the presence of APOBEC3 signature mutations has been associated with the loss of heterozygosity of human leukocyte antigen and thus has been proposed as a mechanism of immune-evasion (28). The expression of a variety of APOBEC3s across multiple cancer types has been associated with lymphocyte infiltration, mutational burden, and PDL-1 and PDL-2 expression, in The Cancer Genome Atlas (TCGA) datasets (18, 29, 30). This type of post hoc TCGA analysis cannot distinguish between T cells which were present and responding to neoantigens generated by prior APOBEC3 activity, or the presence of tumor-reactive T cells which produce TNFα and in turn promote APOBEC3B activity. Nonetheless, the clinical picture is consistent with a correlation between the presence of T-cell activity and APOBEC3 family members playing a role in increasing the mutagenic fuel that may promote immunoediting.

B16 escape from a low MOI reovirus infection is almost certainly mediated by an antiviral response, which itself may involve APOBEC3 activity against the virus in the B16 tumor cells. The presence of tumor-specific T cells, but not naïve T cells, mediated both increased mAPOBEC3 expression in the tumor cells, and increased resistance to reovirus oncolysis, over and above that induced within the B16 cells themselves. In the case of the OVA or HSV-TK models, a defined selective pressure enabled us to identify a mechanism driven by APOBEC3 whereby the integrity of the transgene is compromised. However, the mechanism by which APOBEC3 impedes oncolytic viral replication, spread, or oncolysis, may include the mutation of any number of cellular innate antiviral genes, or indeed, the mutation of the virus genome. To distinguish between these two nonmutually exclusive possibilities, we will sequence the genomes of the cells and viruses that have escaped treatment as a result of APOBEC3 induction.

T-cell interaction induced APOBEC3 expression through a mechanism dependent upon MHC class I, TNFα, and PKC, consistent with previous reports of APOBEC3 induction through the PKC pathway (9, 31). Although in our system APOBEC3 expression in tumor cells was not dependent upon IFNγ, other studies have shown that IFNγ produced by tumor-reactive T cells can drive tumor heterogeneity through genome editing mechanisms associated with DNA damage response pathways (32). Therefore, the tumor-reactive T cells can shape the tumor response to therapy in multiple ways. We hypothesize that APOBEC3 inhibition may in fact be an adjuvant to all direct and indirect immunotherapeutic platforms to reduce the risk of treatment failure and prevent recurrence. This may be particularly relevant to oncolytic viral therapies, as well the treatment of virally induced cancers associated with human papillomavirus (33) or EBV (34) infection, to which APOBEC3 induction is a natural response to restrict viral infection. Although no direct APOBEC3B inhibitors have been identified, indirect inhibition of PKC with AEB071 (9), or B-Myb with EGFR inhibitors (35) represent feasible clinical approaches.

Our sequencing studies confirmed that the parental B16TK and B16OVA cell lines do not contain mutations in either HSV-TK or OVA, despite the fact that they both have detectable basal APOBEC3 expression in vitro. In contrast, upon induction of de novo mAPOBEC3 expression (through an external stimulus such as viral infection, suboptimal T-cell activity, or forced overexpression), the induced mAPOBEC3 becomes mutagenically active leading to genomic mutation, subsequent selection, and clonal outgrowth of cells best suited to survive the applied treatment pressure. We observed fewer surviving B16TK cells in which mAPOBEC3 was knocked down by shRNA than in parental B16 cells, even with no CD8⁺ T cells or with naïve T cells present. These data suggest that reducing steady state endogenous mAPOBEC3 expression still prevents acquisition of resistance to selection by ganciclovir, although not through mutation of the HSV-TK transgene, as parental B16TK cells, expressing steady state mAPOBEC3, maintained a wild-type functional HSV-TK sequence through the ganciclovir selection process both in vitro and in vivo. It may be that endogenous, steady state expression of mAPOBEC3, in addition to induced mAPOBEC3, still contributes to the emergence of resistance, perhaps through the mutation of genes associated with other aspects of ganciclovir transport and/or metabolism by the cell.

The therapies studied here were susceptible to resistance, mediated by a single-point mutation that generated a premature stop codon within an APOBEC3 motif in OVA and TK. As this mechanism of acquired resistance requires the presence of the APOBEC3 motif, our observations are only generalizable to coapplied first-line therapies to the extent that the motif is present in genes of interest.

Additional cell types in the tumor microenvironment that produce TNFα, such as natural killer cells or macrophages, could also upregulate APOBEC3. Indeed, the central role of TNFα in APOBEC3 regulation is consistent with our finding that TNFα derived from CD11b cells promoted tumor recurrence (36).

In summary, our data here suggest that the generation of weak affinity and/or low frequency, suboptimal T-cell responses against tumor-associated antigens may actively drive a mutator phenotype in tumors through APOBEC3 activity and promote the emergence of treatment-resistant tumor populations. Hence, potentially immunogenic first-line therapies administered to patients should be optimized to generate potent CD8⁺ T-cell responses with high frequencies of high affinity circulating antitumor T cells to reduce T-cell–mediated induction of tumor escape and recurrence.

K.K. Smith is an employee at ArticulateScience LLC. R.S. Harris has ownership interest in ApoGen Biotechnologies and is a consultant/advisory board member for ApoGen Biotechnologies. M. Coffey is President and CEO at and has ownership interest in Oncolytics Biotech Inc. K.J. Harrington reports receiving commercial research funding from AstraZeneca, Bristol-Myers Squibb, MerckSerono, and MSD; has received speakers bureau honoraria from AstraZeneca, Bristol-Myers Squibb, MerckSerono, and MSD; and is a consultant/advisory board member for AstraZeneca, Boehringer-Ingelheim, Bristol-Myers Squibb, MerckSerono, MSD, and Pfizer. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Evgin, A.L. Huff, M. Coffey, P.J. Selby, K.J. Harrington, R.G. Vile

Development of methodology: A.M. Molan, R.S. Harris, P.J. Selby, R.G. Vile

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Kottke, K.G. Shim, K.K. Smith, P.J. Selby, R.G. Vile

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Evgin, A.L. Huff, T. Kottke, C.B. Driscoll, M. Schuelke, E.J. Ilett, H. Pandha, K.J. Harrington, R.G. Vile

Writing, review, and/or revision of the manuscript: L. Evgin, A.L. Huff, M. Schuelke, K.G. Shim, P. Wongthida, R.S. Harris, M. Coffey, H. Pandha, P.J. Selby, K.J. Harrington, A. Melcher, R.G. Vile

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Thompson, A.M. Molan

Study supervision: J.S. Pulido, P.J. Selby, R.G. Vile

The authors thank Toni L. Woltman for expert secretarial assistance. This work was funded in part by The European Research Council (to P.J. Selby), The Richard M. Schulze Family Foundation (to R.G. Vile), the Mayo Foundation (to R.G. Vile), Cancer Research UK (to K.J. Harrington and A. Melcher), the NIH (R01CA175386, R01CA108961 to R.G. Vile), The University of Minnesota and Mayo Clinic Partnership (to R.G. Vile) and a grant from Terry and Judith Paul (to J.S. Pulido), and a research grant from Oncolytics Biotech Inc (to R.G. Vile).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.