Detecting and targeting tumor relapse by its resistance to innate effectors at early recurrence

Abstract

Tumor recurrence represents a major clinical challenge. Our data show that emergent recurrent tumors acquire a phenotype radically different from that of their originating primary tumors. This phenotype allows them to evade a host-derived innate immune response elicited by the progression from minimal residual disease (MRD) to actively growing recurrence. Screening for this innate response predicted accurately in which mice recurrence would occur. Premature induction of recurrence resensitized MRD to the primary therapy, suggesting a possible paradigm shift for clinical treatment of dormant disease in which the current expectant approach is replaced with active attempts to uncover MRD before evolution of the escape phenotype is complete. By combining screening with second-line treatments targeting innate insensitivity, up to 100% of mice that would have otherwise relapsed were cured. These data may open new avenues for early detection and appropriately timed, highly targeted treatment of tumor recurrence irrespective of tumor type or frontline treatment.

Tumor dormancy is the time following frontline treatment in which a patient is apparently free of detectable tumor but after which local or metastatic recurrence becomes clinically apparent^1–11. It results from the net balance of tumor-cell proliferation and death through apoptosis, lack of vascularization and immune surveillance of persisting cells^{1–4,9,12–16}. It can also result from growth arrest, or dormancy, at a cellular level^1–3. In the clinic, many tumor types demonstrate periods of dormancy that can last for decades^1,4,17–19. This presents a major clinical challenge, as recurrence occurs at times that cannot be predicted. Recurrent tumors are often phenotypically very different from their primary counterparts, representing the end product of an in vivo selection during which the recurrence has gained or lost expression of genes that render it insensitive to the frontline treatment^20–25. Therefore, the ability to predict, at an early stage, when tumor cells will emerge from MRD and to understand the phenotype of these recurrent tumor cells would be clinically valuable, allowing for the early initiation and design of effective second-line therapies.

We have previously observed apparent cures of tumors following frontline therapy in preclinical models of adoptive T cell transfer^22,26–29, systemic virotherapy^30,31, systemic vesicular stomatitis virus (VSV) cDNA immunotherapy^32,33 and chemotherapy^34–37. However, upon prolonged follow-up, some of these apparently cured animals developed very late occurring local tumors. As shown here, both histology and PCR showed evidence of MRD over long periods of time, even in mice that were scored as cured. These findings have been mirrored in some of our clinical studies using systemic oncolytic virotherapy to treat large local tumors, in which impressive clinical responses have been followed by aggressive and fatal recurrences³⁸. Although transplantable models have several drawbacks as models for primary tumors³⁹, they may have considerable value for studying MRD, tumor dormancy and escape from dormancy.

We show here that across several models of tumor type and treatments, transition from MRD to proliferating recurrence stimulates an innate immune response similar to that seen upon natural infection^40–42. Newly emerging recurrent tumors acquired a substantial insensitivity to innate immune effectors, which allowed them to evade clearance during this transition. Notably, this host response could be screened for in mice undergoing the MRD-to-recurrence transition. This may lead to new screening strategies designed to better predict tumor recurrence in patients. We also show that carefully timed mechanism-based second-line therapies that target this innate immune insensitivity phenotype could effectively treat recurrences across tumor and treatment types.

RESULTS

Recurrence-associated inflammatory response

In about 20–50% of mice apparently cured by primary chemotherapy^34–37, immunotherapy^32,33, virotherapy^30,31 or T cell therapy^22,26–29 100 d previously, nests of tumor cells persisted at the site of injection (Fig. 1a). Following OT-I T cell (MHC class I–restricted ovalbumin-specific CD8⁺ T cells) therapy, which induced macroscopic cure of B16-Ova tumors, skin samples from the site of tumor injection free of palpable tumor 60 d after primary treatment were often positive (7 out of 15 over two independent experiments) for tumor-specific ovalbumin (Ova), consistent with MRD (Fig. 1b). Of those seven Ova⁺ skin samples, three showed evidence of an inflammatory response, including expression of tumor necrosis factor-α (Tnf-α), interleukin-1β (Il-1β) and Il-6.

Figure 1.

Transition from MRD to recurrence is characterized by a local inflammatory response. (a) Histological section at the site of tumor cell injection of the skin of a C57BL/6 mouse seeded with B16-Ova tumor cells 100 d previously and subsequently treated with OT-I adoptive T cell therapy that had no detectable palpable tumor, showing diffuse scattered melanin-containing tumor cells (arrows). Scale bar, 100 μm. (b) Expression (RT-PCR) of Ova antigen and inflammatory cytokines in the skin of one control C57BL/6 mouse that was never seeded with tumor and five mice macroscopically cleared of B16-Ova tumors at day 60 after being treated as in a. (c,d) RT-PCR shows expression of cytokines (c) and tumor antigens (d) at the site of tumor cell injection at day 60 for C57BL/6 mice (n = 6) developing large local recurrences with a diameter of >0.5 cm (+++, n = 2), no detectable recurrence (−, n = 2) or only palpable tumor nodules before progressive recurrence (+, <0.2 cm, n = 2) after complete macroscopic regression of their primary tumor. Asterisk denotes positive for Pmel with nested PCR. Representative of three separate experiments.

To test the hypothesis that proinflammatory cytokines are associated with emergence of recurrence from dormancy, we analyzed cytokine production (RT-PCR) from skin samples harvested from mice in which recurrences were not evident, just developing or fully established. We detected Tnf-α and Il-1β only where no tumor was palpable (Fig. 1c). In contrast, we coordinately detected Il-6, interferon-γ (Ifn-γ) and serum amyloid P component (Apcs), a marker of an acute-phase protein response (APPR) to infection^40–42, in mice bearing emerging (<0.2 cm) recurrent tumors, but we could no longer detect these markers at the site of large (>0.5 cm) actively growing recurrences.

We could no longer detect Ova in large recurrences following OT-I treatment, a finding consistent with antigen loss for evading primary T cell therapy^20,22,26 (Fig. 1d). However, Ova was still expressed in small recurrences, suggesting that loss of the primary target of T cell therapy was not necessarily a prerequisite for the transition from MRD to actively growing recurrence.

Il-6 and vascular endothelial growth factor mark recurrence

As tumor recurrence resembles an infection-like event, which the host immune system recognizes with an APPR^40–42, it could potentially be detected by screening the innate host response initiated when tumors start to expand. Consistent with this hypothesis, we detected Il-6 transiently in the blood within a small window when palpable recurrence was at an early stage (<0.2 cm, Fig. 2a). In contrast, serum Il-6 did not exceed background levels in mice that had no tumor regrowth or in mice whose blood was sampled a few days before recurrence became macroscopically apparent. Notably, serum Il-6 was no longer detectable when we sampled blood at a later stage of recurrence (>0.5 cm). This was not a model-specific phenomenon, as we also detected Il-6 in the sera of three out of five Her-2–neu transgenic mice before the development of spontaneous breast tumors (Supplementary Fig. 1a) but not in five mice that did not develop any tumor (Supplementary Fig. 1b). Therefore, detection of serum Il-6 correlated with the onset and early progression of tumor growth in models of melanoma recurrence and spontaneous breast cancer.

Figure 2.

Il-6 and Vegf correlate with early tumor progression. (a) Correlation between B16-Ova tumor volumes and serum Il-6 concentration after OT-I therapy leading to macroscopic regressions. Serum sampling was performed at days marked with arrows. Each graph represents data from a single mouse. (b,c) Concentrations of Tnf-α and Il-6 (b) and Apcs (c) in lysates from the sites of tumor injection, 60 d after OT-I T cell therapy. MRD, no palpable tumor; <0.2 cm, small tumor nodules; >0.5 cm, actively growing local recurrence. (d) The same mice were assayed for serum Vegf. Mice that never developed recurrences were sampled 44 and 60 d after OT-I therapy; four other mice were assayed at different stages of recurrence progression. (e) Serum Vegf concentration at day 41 for 15 mice with actively growing B16tk recurrences and 15 mice free of palpable tumor after macroscopic regressions (ganciclovir treatment). The time taken to develop large tumors (~1.0 cm) is shown. DT, detectable tumor.

We also detected Il-6 locally at the site of tumor injection at the time of early recurrence (Fig. 2b). We sometimes (in two out of seven mice) detected Tnf-α locally before tumor regrowth but not consistently in the serum (data not shown). We detected Apcs locally in mice with small emerging recurrences (Fig. 2c) but not reproducibly in the serum (data not shown). In contrast, vascular endothelial growth factor (Vegf), a later marker of the APPR, was detectable in sera of mice bearing very small recurrences but not in the same mice before recurrence or in mice with large established recurrent tumors (Fig. 2d). We detected Vegf in three out of the four mice in which Il-6 was detected coincident with tumor regrowth (Fig. 2a) but not in any of the remaining animals, which never developed recurrences or which we failed to sample at the right time. We also observed serum spikes of Vegf coincident with the presence of Il-6 in two out of the three Her-2-neu transgenic mice that were positive for Il-6 (Supplementary Fig. 1c).

MRD followed by recurrence was also induced in mice bearing B16tk tumors by suboptimal ganciclovir (GCV) treatment^34–37. 15 out of 15 mice with recurrent tumors of >0.5 cm when blood was harvested (day 41, 24 d after last GCV injection) were negative for serum Vegf and died from their recurrent tumors within 6 d (Fig. 2e). In contrast, seven out of seven mice with detectable serum Vegf (day 41) had no palpable tumor at this time, but all seven subsequently developed tumors necessitating euthanasia within 9–14 d. Finally, eight mice that had neither Vegf in the serum nor detectable recurrence at the time of blood sampling (day 41) developed recurrent tumors more than 20 d later.

Screening for recurrence

We established a stable reporter assay for recurrence (Fig. 3a). Subcutaneous injection of a control cytomegalovirus (CMV)-luciferase plasmid⁴³ generated luciferase expression from 24 h after injection for up to 8–15 d (Fig. 3b). Luciferase under transcriptional control of 2.6 kb of the Vegf promoter⁴⁴ was expressed neither in mice that were not seeded with tumor (data not shown) nor in mice within ~15 d of cessation of GCV treatment of a primary tumor (Fig. 3c). Within 15–25 d, de novo luciferase expression became detectable only in mice that subsequently developed recurrences within 7–12 d of the first luciferase signal (Fig. 3d). Luciferase was expressed for 5–10 d and became undetectable just before or as recurrences became palpable (Fig. 3e). In mice with progressively growing recurrent tumors (luciferase-positive before palpable tumor), luciferase was often no longer detectable (Fig. 3f), but 2 out of 11 mice became newly positive, possibly owing to stimulation of the Vegf promoter by cytokines released by or released in response to the growing recurrences.

Figure 3.

Plasmid assay to detect tumor recurrence. (a–f) C57BL/6 mice (n = 30) with 5-d B16tk tumors were treated with GCV. Mice in which tumors regressed macroscopically were injected on the opposite flank with CMV-luciferase43 (b) or Vegf-luciferase44 (c–f) plasmids. Signal intensities (number of photons) in the regions of interest (ROI) are shown. Luciferase imaging 48 h after plasmid injection, representative of four (b) and eight (c) mice, respectively. Mice with no palpable tumors imaged on day 34 (d). Days between first detectable luciferase signal and detectable tumor (palpable) are indicated. Images are representative of 16 mice over three different experiments. Mouse 5 from d gave a positive signal on day 34 which had decayed by day 44; a palpable tumor was detected for the first time on day 45 (labeled 5′) (e). Mice with actively growing recurrent tumors, which had been positive for luciferase before recurrence, were either negative at day 52 (9 out of 11 mice) or occasionally newly positive (2 out of 11 mice) (f).

One-hundred percent of mice (n = 11 over three separate experiments) that developed a luciferase signal (Vegf-luciferase) developed recurrences (data not shown). Within 1 d of detection of this signal, all mice tested positive for serum Vegf. However, within 5 d of the first signal, serum Vegf was no longer detectable despite sustained luciferase signal, suggesting that additional cytokines to which the Vegf promoter is responsive may have been produced.

Premature induction of recurrence

We hypothesized that if the MRD-to-recurrence transition was induced prematurely, tumor cells may not have acquired the critical properties allowing escape from frontline therapy. Systemic administration of Vegf during MRD considerably decreased the time to recurrence in two out of three models (Fig. 4a–c). Notably, when we induced recurrences prematurely and administered a second round of initial treatment, survival was significantly (P = 0.01) enhanced relative to groups with no induction, irrespective of tumor type and initial therapy. Induction of recurrence using Vegf was associated with apparent angiogenesis and rapid tumor formation at the site of MRD (Fig. 4d) compared to control sites, where only scattered tumor cells remained 23 d after tumor inoculation (Fig. 4e).

Figure 4.

Premature induction of recurrence resensitizes tumors to frontline therapies. (a) Survival of C57BL/6 mice bearing B16-Ova tumors (n = 5 or 6 per group) treated sequentially with OT-I T cells, then Vegf or PBS, and finally OT-I or PBS. Time from the last injection of OT-I is shown. Representative of two separate experiments. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant (log-rank test). (b) Survival of mice bearing B16tk tumors (n = 8 per group) treated sequentially with GCV or PBS, then Vegf or PBS, and finally GCV or PBS. Time from the last GCV injection is shown. Representative of two separate experiments. (c) Survival of mice bearing B16 melanoma or TC2 prostate tumors (n = 6 per group) treated sequentially with an altered self epitope library (ASEL, a VSV cDNA library) or PBS, then Vegf or PBS, and finally ASEL or PBS. Time from the last day of treatment is shown. Representative of four separate experiments. (d,e) Histological sections of skins 23 d after B16-Ova challenge, treated as in a with OT-I + Vegf + OT-I (d) or OT-I + PBS + OT-I (e) Scale bars, 100 μm. (f–i) Histological sections (day 60) from the skin of one of six long-term survivor mice (n = 7 per group) treated as in a with OT-I + PBS + OT-I with four additional OT-I injections (d 24, 25, 31 and 32) (i). Scale bars, 250 μm (f,g and i) or 100 μm (h). (j) Expression (RT-PCR) of Pmel and Ova antigens (OT-I + Vegf + OT-I therapy) at the site of B16-Ova tumor injection. Asterisks indicate no palpable tumor.

In the OT-I model (Fig. 4a), two injections of OT-I were rarely curative. With no Vegf induction, three subsequent rounds of OT-I were completely ineffective, resulting in large recurrences (Fig. 4f) that had lost Ova expression in seven out of seven mice (data not shown). In contrast, induction with Vegf followed by six injections of OT-I facilitated recognition of MRD by OT-I T cells, as shown by the lymphoid aggregate in the subcutaneous tissue (Fig. 4g,h), and cleared residual tumor in six out of seven mice over 120 d macroscopically, histologically (Fig. 4i) and molecularly (Fig. 4j).

Recurrences evade the innate immune response

An early requirement for expansion of a recurrent tumor from MRD could be acquisition of resistance to an innate immune response against expansion of a few cells into an actively growing tumor mass. Early (<0.2 cm) primary or recurrent tumors from untreated mice or those treated with chemotherapy (B16tk, GCV treatment), T cell therapy (B16-Ova, OT-I T cells injection), immunotherapy (TC2, VSV cDNA library treatment) or virotherapy (B16tk, paclitaxel and reovirus treatments) to induce complete regression followed by recurrence were explanted. Whereas primary B16-Ova tumors were good targets for natural killer (NK) cells, early recurrences following OT-I therapy were almost completely insensitive as shown by their inability to activate Ifn-γ secretion from NK cells (Fig. 5a). Conversely, although cultured B16-Ova or early untreated tumors were poor targets for Pmel T cells⁴⁵, which are naturally able to target the melanoma-specific antigen gp100 on these cells, B16-Ova recurrences were reproducibly better targets for T cell recognition (Fig. 5b). We also found almost complete loss of NK cell recognition in three other models (Fig. 5c and data not shown). Major histocompatibility complex (MHC) class I expression was not significantly (P = 0.2) different between recurrent and primary tumors, except for a slight increase (P = 0.05) in three B16-Ova recurrences (OT-I therapy) compared to primary tumors (data not shown).

Figure 5.

Recurrent tumors are insensitive to innate immune effectors. (a,b) Ifn-γ secretion from NK cells (a) or Pmel T cells (b) cocultured with early (~0.2 cm) B16-Ova local recurrences (OT-I treatment) or primary tumors (PBS). Positive control: Pmel T cells + human gp100 peptide. Representative of two separate experiments. (c) Tnf-α secretion from NK cells cocultured with early B16 recurrences (paclitaxel + reovirus treatment) or primary tumors (PBS) previously infected in vitro with reovirus. (d) Reovirus titers 48 h after infection of early tumor explants in the presence or absence of Ifn-α. TCID₅₀, 50% tissue culture infectious dose. c and d are representative of two separate experiments. (e) Surviving TC2 cells (crystal violet staining) explanted from recurrent (ASEL treatment) or primary (PBS) tumors after in vitro infection by VSV in the presence or absence of Ifn-α. Representative of two separate experiments. (f) The experiment in b was repeated using OT-I T cells. Positive control: OT-I + SIINFEKL peptide from ovalbumin. (g) Thymidine-kinase expression (western blot) in B16tk recurrences (GCV treatment) and primary tumors (PBS).

Recurrences from these different models were also highly insensitive to Ifn-α, which protects against reovirus (Fig. 5d) and VSV (Fig. 5e) infections. When early recurrences were maintained for more than a month in culture, the innate insensitive phenotype was lost in about 50% of samples. Similarly, when late recurrences (>0.8 cm) were freshly explanted, they showed greater ability to activate NK cells (data not shown). Loss of innate immune sensitivity following virotherapy was observed in B16 melanoma, TC2 prostate cancer and orthotopically grown GL261 glioma (Supplementary Fig. 2) recurrences. Luciferase expression from an interferon-sensitive response element in tumors from PBS-treated mice was substantially higher than that in early recurrences of B16, TC2 or intracranial GL261 tumors (Supplementary Fig. 2), indicating a profound loss of type I interferon signaling across three different histological types of recurrence.

Recurrent tumors also acquired phenotypes specific to evasion of the frontline therapy. B16-Ova recurrences became insensitive to OT-I T cells due to loss of the Ova antigen^22,26 (Figs. 4j and 5f). Similarly, B16tk recurrences lost expression of the HSVtk protein (Fig. 5g) and became insensitive to GCV in vitro (data not shown). Neither Ova nor HSVtk were lost from B16-Ova or B16tk tumors treated with paclitaxel and reovirus (data not shown).

Vegf induction of premature recurrence generated tumors that were both intact in type I interferon signaling and sensitive to NK cells (Supplementary Fig. 3). In contrast, spontaneous B16tk recurrences were completely insensitive to NK recognition and had almost completely lost the integrity of their interferon signaling. This demonstrates that Vegf-induced recurrences retained a phenotype much more similar to the parental primary tumor than to naturally occurring recurrences, at least with respect to sensitivity to innate effectors.

Therapeutic targeting of the recurrence-specific phenotype

We sought to treat recurrent tumors with second-line treatments targeting the recurrence-specific phenotype. Intratumor reovirus had moderate therapeutic effects against primary B16tk. (Fig. 6a). Therapy largely depended on NK cells, as reovirus treatment was no longer more efficient than PBS injection when NK cells were depleted before therapy. Consistent with insensitivity of recurrent tumors to type I interferons, conferring enhanced sensitivity to viral oncolysis^46,47; although emerging B16tk recurrences had effectively failed initial reovirus therapy, reovirus was highly efficacious against recurrences (Fig. 6b,c). Therapy was no longer dependent upon NK cells and was instead associated with higher levels of virus replication within the tumors (Fig. 6d).

Figure 6.

Targeting the recurrence-specific phenotype of innate immune insensitivity. (a) Mice with B16 tumors (n = 7 per group), depleted or not of NK cells, were treated with reovirus or PBS. Survival with time is shown (log-rank test), representative of two separate experiments. (b) Mice with B16 tumors (n = 5 or 6 per group) were treated with paclitaxel plus reovirus. As tumors recurred, mice were depleted or not of NK cells and injected with reovirus. Long-term survival (log-rank test) is shown with time from the last injection of reovirus, representative of two separate experiments. (c) Histological section of the skin of a long-term (d 120) survivor from b (reovirus + NK). Scale bar, 500 μm. (d) Primary (PBS) or recurrent (PAC + reo) B16 tumors were injected at ~0.2- to 0.3-cm diameter with reovirus in mice ± NK cells (n = 3 per group) or PBS (n = 2 per group). Reovirus titers from explanted tumors are shown. Data are expressed as mean ± s.d. (e) Mice with B16-Ova tumors (n = 5–7 per group) were treated with PBS or OT-I. Mice in which the primary tumor regressed were injected (opposite flank) with Vegf-luciferase (Luc) plasmid 15 d after the last injection of OT-I. During MRD, second-line therapy was initiated either early (luc⁻, day 20) with Ifn-α, Vegf, Pmel T cells or Vegf followed by Pmel or late (luc⁺, ~d 30) with Ifn-α, Pmel T cells or PBS. Survival with time after the last injection of OT-I (day 0) is shown (log-rank test), representative of two separate experiments. *P < 0.05, **P < 0.01, ***P < 0.001. NS, not significant.

We used our ability to detect triggering of recurrence from MRD to treat relapse prospectively. By screening B16-Ova tumors treated with OT-I to establish MRD with the Vegf-luciferase reporter assay, we were able to initiate treatment with second-line therapy at early (luciferase-negative) or late (luciferase-positive) time points at which no recurrences were palpable in the mice. Consistent with the decreased sensitivity to interferon characteristic of recurrences, systemic Ifn-α given early was curative and significantly (P = 0.01) more effective than treatment started late (Fig. 6e). Conversely, late treatment with Pmel T cells was more effective than early treatment, thus confirming an increased susceptibility of recurrences to T cell recognition. The therapeutic impact of early Pmel treatment was considerably enhanced by inducing premature release from MRD with systemic Vegf.

DISCUSSION

Our data are compatible with a model in which the transition from MRD into an actively growing recurrent tumor stimulates a host-mediated APPR similar to that seen in a natural infection^40–42. It seems probable that, to expand from MRD, recurrent tumor cells must evade the potent innate immune response triggered by their proliferation and the release of inflammatory signals.

The ability to predict and detect when patients will relapse would be of great clinical value, allowing rapid intervention with second-line therapies that specifically target the phenotype of recurrent tumors^2–4. We show that it may be possible to detect the critical stage at which a dormant tumor has successfully acquired the portfolio of phenotypic changes required to expand in vivo as a viable recurrence. Spikes in serum Il-6 and Vegf, possibly as a result of the host response to the MRD-to-recurrence transition, served as predictors of the onset of recurrent tumor growth in mice. Vegf promoter induction may also be caused by the secretion of additional factors related to the innate immune response over several days, although endogenous levels of Vegf are rapidly controlled in vivo. Notably, 11 out of 11 mice (from three independent experiments) that converted from luciferase negative to luciferase positive before clinical signs of recurrence developed recurrent tumors.

However, how best to implement such tests in patients whose risk of recurrence may span several years is yet to be determined^1,4,17–19. Cytokine spikes were unpredictable and resolved quickly before clinical evidence of relapse in our experiments here. We are developing protocols to monitor patients at high risk of relapse, such as patients with multiple-node–positive melanoma, by taking monthly blood samples with prompt repeat samples at the first suspicion of relapse to seek initial indications of the existence and the nature of human cytokine spikes. This approach is consistent with frequent monitoring protocols that are being developed to detect circulating tumor cells.

If a tumor transitioning from MRD to recurrence is recognized by a host innate response but not cleared by it, a prerequisite of escape is acquisition of a phenotype that is insensitive to innate immune recognition and clearance. We observed a recurrent tumor–specific treatment-nonspecific phenotype associated with a dramatic loss of sensitivity to innate immune effectors (type I interferons and NK cells) common to recurrent tumors from different tumor types. Interestingly, the innate immune insensitivity was not completely stable with time. We hypothesize that the phenotype was associated with a subset of cells that predominate in early recurrence but become less predominant in the overall tumor population with time, as we showed recently⁴⁸.

Recurrent tumors also acquired a treatment-specific recurrence-nonspecific phenotype as part of the escape process, such as loss of HSVtk or Ova expression^22,26. It was possible to cure mice of secondary tumor by inducing premature recurrence of MRD with systemic Vegf before full evolution of this treatment-specific recurrent phenotype was complete. We are currently investigating the precise role of Vegf in promoting accelerated appearance and therapy of recurrent tumors, which may include direct effects on tumor growth, enhanced angiogenesis of small foci of MRD or enhanced access of therapeutic agents or tumor-specific T cells.

These data raise the possibility that promoting recurrence would allow for more effective treatment of patients at high risk of relapse. Persisting vulnerabilities of preemergent early-stage recurrences could drive a paradigm shift away from current clinical practice of a passive vigilance for disease recurrence. Our data suggest that it may be more clinically effective to actively seek to uncover dormant MRD by administering transient tumor-stimulating signals that force it to reveal itself when it is not yet optimally equipped to evade therapies. Although controversial, a similar approach of administering thyroid-stimulating hormone to patients with high-risk thyroid cancer who are apparently in complete remission is currently recommended as a means of revealing residual subclinical tumors by driving them to secrete the marker thyroglobulin^49–51. Thyroglobulin-positive patients can then be further treated when disease burden is still very low, illustrating the potential value of a more generalizable clinical approach of flushing out occult MRD and treating it for effective cure.

Our characterization of a requirement for tumors to acquire an innate-insensitive recurrent-specific phenotype provided a potentially translatable opportunity to design rational second-line therapies to target early recurrences of different etiologies. Recurrences were treated very effectively with oncolytic virotherapy by exploiting their increased insensitivity to type I interferon^46,47. Antitumor activity has been observed in several clinical trials of oncolytic virotherapy, usually in patients who have been heavily pretreated with other therapies. It will be interesting to see whether these pretreated tumors, which may resemble recurrences, are more sensitive to virotherapy than the primary untreated tumors. Tumor cells still in the state of MRD were also very sensitive to the antitumor effects of Ifn-α^52,53, provided that it was administered before the MRD-to-recurrence transition. These results may help to explain some of the variability of patients' responses seen clinically in trials of systemic Ifn-α^52,53 by emphasizing how timing of follow-up treatments during the transition from MRD to recurrence may be critical.

In summary, our data suggest that it may be possible to develop new screening methods for subclinical MRD, instigate new clinical approaches that flush out occult MRD before treatment and design a few common therapies to treat a variety of recurrences of different etiologies.

ONLINE METHODS

Mice, cell lines and viruses

We purchased 6- to 8-week old female C57BL/6 mice from The Jackson Laboratory (Bar Harbor, Maine, USA). The OT-I mouse strain is on a C57BL/6 background (H2-K^b) and expresses a transgenic T cell receptor Vα14/Vβ12-1 (IMGT nomenclature) specific for the SIINFEKL peptide of ovalbumin in the context of MHC class I H-2K^b as previously described⁵⁴. The mice were bred at the Mayo Clinic. Pmel-1 transgenic mice (C57BL/6 background) express the Vα7-D5/Vβ14 T cell receptor, which recognizes amino acids 25–33 of gp100 of Pmel-17 presented by H2-D^b MHC class I molecules⁴⁵. We purchased Pmel-1 breeding colonies from The Jackson Laboratory at 6 to 8 weeks of age. The B16-Ova cell line was derived from a B16.F1 clone transfected with a pcDNA3.1ova plasmid^22,26. B16-Ova cells were grown in DMEM (HyClone, Logan, UT, USA) + 10% FBS (Life Technologies Grand Island, NY, USA) + 5 mg ml⁻¹ G418 (Mediatech, Manassas, VA, USA) until challenge. B16tk cells were derived from a B16.F1 clone transfected with a plasmid expressing the herpes simplex virus thymidine kinase (HSVtk) gene. Following stable selection, these cells were shown to be sensitive to GCV (Cymevene, Roche) at 5 μg ml⁻¹ (refs. 35–37). Following in vitro cultures or harvest from mice, tumor lines were grown in DMEM + 10% FBS + 1% penicillin-streptomycin (Mediatech). We determined wild-type reovirus type 3 (Dearing strain) titers by plaque assays on L929 cells. For in vivo studies, we administered reovirus i.v. at 2 × 10⁷ TCID₅₀ per injection. For virus titration, tumors were harvested from mice, weighed and lysed (three freeze-thaw cycles within 2 h of removal). Virus titers are expressed as TCID₅₀ per mg of tissue.

In vivo experiments

All in vivo studies were approved by the Mayo Institutional Animal Care and Use Committee. Mice were challenged subcutaneously with 5 × 10⁵ B16-Ova, B16tk or B16 melanoma cells, or with 2 × 10⁵ TC2 prostate tumor cells, in 100 μL PBS (HyClone). We measured tumors three times per week, and mice were euthanized when tumors reached one centimeter in diameter.

For luciferase imaging, we injected mice intraperitoneally (i.p.) with 200 μl of d-luciferin substrate. 10 min later, mice were anesthetized with 2% isoflurane (Novaplus) and imaged using a Xenogen IVIS200 bioluminescent imager (PerkinElmer, Waltham, MA, USA).

Serum harvested from mice was assayed for Il-6, Vegf, Tnf-α, Il-1β and Apcs using ELISA detection kits (BD Biosciences, San Diego, CA, USA, or Kamiya Biomedical Company, Seattle, WA, USA) according to the manufacturer's instructions.

For apparently curative adoptive therapy experiments, we injected i.v. 1 × 10⁷ in vitro–activated OT-I T cells in 100 μL PBS 5 and 7 d after tumor seeding. This regimen typically led to tumor regression and tumor-free `cure' of ~50% of mice for over 60 d (refs. 22,26–29). For suboptimal adoptive T cell therapy, in which more than 50% of treated mice would undergo complete macroscopic regression followed by local recurrence, mice seeded subcutaneously with B16-Ova tumors 5 d previously were treated i.v. with PBS or with 1 × 10⁶ 4-d activated OT-I T cells on days 6 and 7.

For GCV chemotherapy experiments, we treated C57BL/6 mice seeded with B16tk tumors 5 d previously with GCV i.p. at 50 mg ml⁻¹ on days 6–10 and then on days 13–17. With this regimen, 100% of tumors regressed macroscopically, followed by ~50–80% of the mice undergoing later local recurrence.

For VSV cDNA library–based immunotherapy experiments, C57BL/6 mice seeded with either B16 melanoma or TC2 prostate tumors 5 d previously were treated with three i.v. injections on days 6, 8 and 10 with the ASEL, a VSV cDNA library previously shown to cure TC2 tumors using nine i.v. injections³².

For suboptimal systemic oncolytic reovirus virotherapy experiments, C57BL/6 mice with 5-d established B16 tumors were treated i.p. with PBS or paclitaxel (Mayo Clinic Pharmacy, Rochester, MN, USA) at 10 mg per kg body weight per injection for 3 d followed by i.v. reovirus (2 × 10⁷ TCID₅₀) or PBS for 2 d. We repeated this cycle (paclitaxel-reovirus, five injections, 2-d rest), modified from a more effective therapy described³⁰, once.

In vivo NK depletion was performed by two i.v. injections (0.6 mg per injection) of asialo-GM1–specific antibody (CL8955, Cedar Lane, Burlington, NC, USA) or control IgG on days 5 and 6 after tumor challenge. We subsequently injected reovirus (1 × 10⁸ TCID₅₀) intratumorally on days 8 and 9.

For prospective therapy, we treated 5-d primary B16-Ova tumors with three daily injections of 1 × 10⁶ OT-I T cells or PBS. Mice showing tumor regression were injected on the opposite flank with Vegf-luciferase plasmid (10 μg) 15 d after the last injection of OT-I. Second-line therapy was administered 5 (early treatment) or 10 (late treatment) d later. Treatment included Ifn-α (10,000 U per injection, 3 consecutive d), Vegf (10 μg per injection, 3 consecutive d), Pmel T cells (1 × 10⁶ cells per injection, 2 consecutive d) or Vegf followed by Pmel.

In vivo data were analyzed using GraphPad Prism 4 software (GraphPad Software, La Jolla, CA, USA).

Premature induction of recurrence

For all these experiments, we set day 0 as the last day of treatment. For the adoptive T cell therapy model, we treated C57BL/6 mice bearing 5-d subcutaneous B16-Ova tumors i.v. with PBS or 1 × 10⁶ activated OT-I T cells (days −12 and −11). Mice were subsequently treated with PBS or recombinant mouse Vegf₁₆₅, encoded by the most abundant splice variant of Vegfa, (ProSpec, East Brunswick, NJ; 10 mg per injection) on days −6, −5 and −4, then PBS or two additional injections of 1 × 10⁶ OT-I T cells on days −1 and 0.

For the chemotherapy model, we treated mice bearing 5-d B16tk tumors i.p. with GCV (days −15 to −11), i.v. with PBS or Vegf on days −7, −6 and −5 and then with PBS or GCV for 5 additional d (days −4, −3, −2, −1 and 0).

For our viroimmunotherapy model, mice bearing 5-d B16 melanoma or TC2 prostate tumors were treated with three i.v. injections (days −15, −13 and −11) of the ASEL. We subsequently treated mice with PBS or Vegf on days −7, −6 and −5 and then with PBS or three additional ASEL injections on days −2, −1 and 0.

RT-PCR

Skin at the site of tumor cell injection or tumors were immediately excised from euthanized mice and dissociated in vitro to achieve single-cell suspensions. We extracted RNA from cells using the Qiagen RNeasy kit (Qiagen, Valencia, CA, USA). cDNA was made from 1 μg total cellular RNA using the First Strand cDNA Synthesis Kit (Roche, Indianapolis, IN, USA). We amplified a cDNA equivalent of 1 ng RNA by PCR with gene-specific primers. Gapdh was used as a positive control for expression. The following primers were used: Ova sense (CACAAGCAATGCCTTTCAGA), Ova antisense (TACCACCTCTCTGCCTGCTT), Pmel sense (CCAGCCCATTGCTGCCCACA), Pmel antisense (CCCGCCTTGGCAGGACACAG), Gapdh sense (TCATGACCACAGTCCATGCC) and Gapdh antisense (TCAGCTCTGGGATGACCTTG). Other primers available on request.

Immune cell activation

Spleens were immediately excised from euthanized OT-I or Pmel mice and dissociated in vitro to achieve single-cell suspensions. We lysed red blood cells with ACK lysis buffer for 2 min and resuspended cells at 1 × 10⁶ cells ml⁻¹ in Iscove's Modified Dulbecco's Medium (Gibco, Grand Island, NY, USA) + 5% FBS + 1% penicillin-streptomycin + 40 μM 2-mercaptoethanol. Cells were pulsed with 2.5 μg ml⁻¹ of the synthetic H-2K^b-restricted peptides Pmel_25–33 (KVPRNQDWL) or Ova_257–264 (SIINFEKL) synthesized at the Mayo Foundation Core Facility, Rochester, MN, USA, or medium for 48 h. We then cultured tumor cells with activated T cells at different effector/target ratios of 100:1, 50:1 and 10:1. We collected cell-free supernatants 48 h later and tested them for Ifn-γ production by ELISA as directed in the manufacturer's instructions (BD Biosciences, San Diego, CA, USA).

We prepared NK cells from spleens of naive C57BL/6 mice using the NK Cell Isolation Kit II (Miltenyi, Auburn, CA, USA) and cultured them with target tumor cells 48 h after obtaining explants at different effector/target ratios of 100:1, 50:1 and 10:1. 72 h later, we assayed supernatants for Ifn-γ or Tnf-α by ELISA. For the reovirus-NK assay, explanted tumors were cultured for 48 h, infected with reovirus (multiplicity of infection (MOI) 1.0) for 24 h and then cultured with NK cells at indicated ratios.

Virus sensitivity assays

For reovirus, explanted tumors were cultured for 72 h before being infected (MOI 1.0) with or without Ifn-α (100 U). Reovirus titers were determined after 48 h.

For VSV (Indiana strain), tumors were cultured for 48 h and infected with VSV (MOI 0.01) with or without Ifn-α (100 U). 15 h later, we stained surviving cells with crystal violet.

Histopathology

Skin at the site of initial tumor-cell injection or tumors were harvested, fixed in 10% formalin, paraffin-embedded and sectioned. Two independent pathologists blinded to the experimental design examined H&E sections. Images were acquired using an Olympus microscope and AnalySIS 5.0 software.

Statistical analyses

Survival curves were analyzed by the log-rank test. Statistical significance was set at P < 0.05 for all experiments. *P < 0.05, **P < 0.01, ***P < 0.001.