Adaptive immunity does not strongly suppress spontaneous tumors in a Sleeping Beauty model of cancer

Abstract

The tumor immunosurveillance hypothesis describes a process by which the immune system recognizes and suppresses the growth of transformed cancer cells. A variety of epidemiological and experimental evidence supports this hypothesis. Nevertheless, there are a number of conflicting reports regarding the degree of immune protection conferred, the immune cell types responsible for protection, and the potential contributions of immunosuppressive therapies to tumor induction. The purpose of this study was to determine whether the adaptive immune system actively suppresses tumorigenesis in a Sleeping Beauty (SB) mouse model of cancer. SB transposon mutagenesis was performed in either a wild-type or immunocompromised (Rag2-null) background. Tumor latency and multiplicity were remarkably similar in both immune cohorts, suggesting that the adaptive immune system is not efficiently suppressing tumor formation in our model. Exceptions included skin tumors, which displayed increased multiplicity in wild-type animals, and leukemias, which developed with shorter latency in immune-deficient mice. Overall tumor distribution was also altered such that tumors affecting the gastrointestinal tract were more frequent and hemangiosarcomas were less frequent in immune-deficient mice compared to wild-type mice. Finally, genetic profiling of transposon-induced mutations identified significant differences in mutation prevalence for a number of genes, including Uba1. Taken together, these results indicate that B- and T-cells function to shape the genetic profile of tumors in various tumor types, despite being ineffective at clearing SB-induced tumors. This study represents the first forward genetic screen designed to examine tumor immunosurveillance mechanisms.

Introduction

Tumor immunosurveillance is an emerging hallmark of cancer supported by numerous human and mouse studies (1). For example, organ transplant patients undergoing immunosuppressive anti-rejection therapies, as well as patients with HIV or chronic lymphocytic leukemia, are at a greater risk of developing cutaneous squamous cell carcinomas and Kaposi’s sarcomas when compared to the general population (2–4). However, immunosuppressed patients are at a higher risk of contracting cancer-associated viral infections (e.g. HPV), and some studies indicate that drugs used to suppress the immune system are themselves carcinogenic (5). Due to conflicting reports, there is still ongoing debate regarding both the nature and strength of immunosurveillance in controlling tumor growth.

Many genetically immunocompromised mouse strains (e.g. Rag2^−/−) develop spontaneous tumors at higher rates than their immunocompetent counterparts (6). Immune control of tumor development is also supported by experiments using the chemical carcinogen, 3’-methylcholanthreme (MCA) (7). Koebel et al. recently demonstrated that antibody depletion of CD4/CD8 T-cells rendered mice more susceptible to MCA-induced sarcoma formation. Moreover, sarcomas produced in a Rag2^−/− background were also more immunogenic than those arising in the presence of a functional immune system, indicating that lymphocytes can shape tumor immunogenicity. Yet, these experiments lack genetic analysis of tumors arising in immunocompromised backgrounds. In fact, little is known about the mechanisms by which tumors escape immune suppression, and additional knowledge could be useful when designing cancer immunotherapies.

This study sought to independently determine if the adaptive immune system actively suppresses spontaneous tumorigenesis using a genetically engineered mouse model of cancer. Previous work has shown that Sleeping Beauty (SB) transposon mutagenesis produces a variety of spontaneous carcinomas (8). This approach facilitates high-throughput identification of candidate genes responsible for tumor induction (9). Because immunoediting would exert a selective force on developing tumors, we reasoned that tumors capable of immune escape would have an altered genetic profile as a result. By comparing mutation spectra of tumors produced in the presence or absence of an active adaptive immune response, we might identify key mutations involved in immune evasion.

Contrary to our expectation, we found that the adaptive immune system did not significantly impact tumor latency or multiplicity in our screen. However, tumor distribution and mutation spectra in solid tumors were slightly different, depending on immune status. Candidate genes with differential mutation between cohorts were observed in hepatocellular carcinomas and leukemias. Unfortunately, additional experiments were unable to confirm that one of these candidate mutations, Uba1 would result in increased immunogenicity. Our findings lead us to conclude that, in contrast with the MCA model, B- and T-cells do not play a major role in suppressing tumorigenesis in SB models of cancer. However, we do present evidence that the adaptive immune system is shaping the genetic spectra of tumors, despite a lack of widespread effects on kinetics of tumor formation.

Materials and Methods

Generation and Aging of Tumor-Prone Mice

Rag2^−/− mice (RAGN12) were obtained from Taconic (10). T2/Onc3 and RosaSB mice were previously described (8). Animals were aged in a specific pathogen-free facility until 60 weeks of age, unless moribund or with skin lesion > 2 cm in diameter. Full necropsy was performed upon euthanization. All animal experiments followed guidelines approved by the University of Iowa Institutional Animal Care and Use Committee.

Flow Cytometry

Peripheral blood was collected from adult mice. After red blood cell lysis, cells were labeled with antibodies: anti-CD45R/B220-PE-Cy™7 (BD Bioscienes, cat #561881), anti-CD8a-eFluor® 450 (eBioscience, cat #48-0081-82) and anti-NK1.1-PE-Cy™7 (BD Biosciences, cat #552878). Flow cytometry was performed with the BD™ LSR II, and data were analyzed using FlowJo software (Tree Star).

Illumina Sequencing of Tumors

Transposon insertion sites were identified using a modified protocol previously described (9, 11). Briefly, tumor DNA was isolated using the GenElute™ Mammalian Genomic DNA Miniprep Kit (Sigma), sheared to 300 bp using a Covaris sonicator, then end-repaired using T4 DNA polymerase and PNK (New England Biolabs). Adapters (5’-GTAATACGACTCACTATAGGGCTCCGCTTAAGGGAC-3’ and (5’-Phos-GTCCCTTA AGCGGAG-C3spacer-3’) were annealed and ligated to repaired DNAs and digested with BamHI to eliminate junctions from unmobilized transposons. Primary PCRs were performed using Platinum Taq and IRR (5’-GGATTAAATGTCAGGAATTGTGAAAA-3’), IRL (5’-AAATTTGTGGAGTAGTTGAAAAACGA-3’), and linker (5’-GTAATACGACTCA CTATAGGGC-3’) primers. Nested secondary PCRs using barcoded primers were performed in preparation for sequencing by the University of Iowa DNA Facility with the HiS eq 2000 platform (Illumina). Statistical analysis was performed as previously described (9).

Uba1 Insertion PCR

Mouse tumor DNA was amplified using IRL/IRR.F (5’-TGTATGTAAACTTCCGA CTTCAACTG-3’), Uba1.0256.R (5’-CCGCTTGCCTAGTGACTCTT-3’), Uba1.1804.R (5’-GGGGGAATGTAGACCCTGAA-3’), OncAL (5’-TCACAATTCCAGTGGGTCAG-3’) and OncAR (5’-TTTCATCATCGGCTGAAGTG-3’) primers.

B16-F0 Transduction and Growth Curve

The Uba1 ORF was PCR-amplified from normal mouse liver cDNA using primers Uba1cloneF (5’-GTTAACCTTCGGCTTGTCTCCAGAAG-3’) and Uba1cloneR (5’-GTTAACTCAGCGAATGGTATATCG-3’). The 3.2kb product was cloned into piggyBac vector PB-EF1alpha-IRES-Neo. B16-F0 cells were obtained from ATCC and stably transfected with the Uba1 or empty vector using Effectene Reagent (Qiagen). Selection with 1 mg/ml G418 began two days post transfection and resistant cells were pooled and passaged in the presence of G418. Growth rates were measured by plating 1×10⁴ cells per well on a 6-well plate with media changes every 2–3 days. Cells were trypsinized and counted at indicated time points.

Western Analysis

Total protein from homogenized tissues was subjected to SDS-PAGE. Protein was transferred to nitrocellulose membranes for western blotting with anti-Uba1 (1:1000, Aviva Systems Biology Corp., cat #ARP58687), anti-β-tubulin (1:1000, Sigma, cat #T8328), or anti-Ub (1:1000, Santa Cruz, cat #47721). Proteasomal degradation was inhibited by culturing cells in 20 µM MG132 (Sigma) for 24 hours before lysis.

Uba1 qRT-PCR

Total RNA was extracted from the stably-transduced cell lines or human tissues using the PerfectPure RNA Tissue Kit (5–Prime). Total cDNA was generated with the SuperScript® III First-Strand Synthesis System (Invitrogen) using the oligo dT primer. qPCR was performed using the Platinum® SYBR® Green qPCR SuperMix (Invitrogen) with the following primers: Uba1F (5’-TGACCAGGAGTTGCAGAGTG-3’), Uba1R (5’-GGTCTGCAGGGGAAATATCA-3’), HprtF (5’-TGATCAGTCAACGGGGGACA-3’), and HprtR (5’-AGAGGTCCTTTTCACCAGCA-3’). Uba1 C_t values were first normalized to Hprt C_t values, and then to B16-F0 using the ΔΔC_t computational method (12).

Syngeneic Transplant Model

B16-F0 cells were diluted to 1×10⁶ cells/ml in PBS and 200ul were injected into the tail vein of adult C57BL/6 females. Mice were euthanized 3 weeks after injection and lungs were examined for melanocyte colony formation (13).

Results

Generation and Characterization of Mouse Cohorts

It was previously reported that tumor immunoediting is a lymphocyte-mediated process (7). Based on this report, we sought to define a genetic signature specific to edited tumors, thereby identifying a potential mechanism by which edited tumors may escape immune suppression. We previously described an SB model of cancer in which mice developed a wide variety of tumor types (8). By generating a tumor cohort that developed in the absence of an adaptive immune response, we hypothesized that genetic analysis of transposon-induced mutations originating in these tumors might reveal differences in mutation spectra that could be attributed to selective pressure of the adaptive immune system.

Two large cohorts of SB-mutagenized immunocompetent (Rag2^+/−) and immunocompromised (Rag2^−/−) mice (n=96 and n=91, respectively) were generated (Supplemental Fig. S1). A control cohort of immunocompromised mice lacking transposon mutagenesis was also generated (n=30). Peripheral blood from adult mice in all three cohorts was analyzed by flow cytometry to verify that transposon mutagenesis did not rescue the Rag2^−/− phenotype in developing lymphocytes. As expected, mature B- and T- cells were lacking in both the Rag2^−/− tumor-prone mice and the Rag2^−/− controls (Fig. 1A and B). It is known that Rag2^−/− mice are still able to produce functional NK cells, which have also reported to play a role in immunosurveillance (14). To verify NK cell production in our mice, peripheral blood was again analyzed by flow cytometry and NK1.1 positive cells were normalized to the non-lymphocyte peripheral blood cell populations. All cohorts were found to maintain NK cell production, though the Rag2^−/− cohorts displayed a trend toward fewer NK cells produced (Fig. 1C).

FIGURE 1. Rag2-null mice produce NK, but not mature B- or T-cells.

Flow cytometry was performed on peripheral blood collected from adult mice. A, Representative histograms depict the mean fluorescence intensity (MFI) of CD8 or B220. Rag2^−/− (+SB) (blue line) and Rag2^−/− (−SB) (black line) MFIs were lower than Rag2^+/− (+SB) (red line) MFIs, indicating the absence of mature T- and B-cells in the immunocompromised cohorts. The gray curve represents unlabeled control cells. Gates on histograms were used to generate panel B. B, CD8- or B220-positive cells are given as a percentage of total cells counted. Rag2^+/− (+SB) mice (n=7, red circles) show a significant increase in circulating T-cell and B-cell populations compared to the Rag2^−/− (+SB) (n=9, blue circles) and Rag2^−/− (-SB) (n=8, black circles) cohorts (P<0.05, 1-Way ANOVA with Bonferroni posttest). C, NK1.1-positive cells were counted and normalized to non-lymphocytic cells. Rag2^+/− (+SB) mice (n=5, red circles) did not show a significantly different number of circulating NK cells compared to the Rag2^−/− (+SB) (n=4, blue circles) and Rag2^−/− (−SB) (n=4, black circles) cohorts (P=0.0624, unpaired t).

Latency, Multiplicity, and Tissue Distribution of SB-induced Tumors

Mice in both SB-mutagenized Rag2^−/− and Rag2^+/− cohorts developed a variety of tumor types with average latencies of 47 and 51 weeks, respectively (Fig. 2ASupplemental Fig. S2). Previous reports indicate that Rag2^−/− mice are more susceptible to development of spontaneous epithelial carcinomas (15). Interestingly, solid tumors in our cohorts developed with equivalent latencies regardless of immune status (Fig. 3A and B), and only leukemias developed with a shorter latency in the immunocompromised group (Fig. 3C). Leukemias also demonstrated a difference in disease severity, such that Immunocompromised mice had greater tumor burden as measured by the number of different tissues affected (Fig. 3D).

FIGURE 2. Sleeping Beauty mutagenesis produces subtle phenotypic differences.

Aging of SB-mutagenized animals was carried out in a specific pathogen-free facility. A, Tumor-free survival plots (all tumor types) show a significant difference in latency (P=0.044, Log-Rank Mantel-Cox test) between SB-mutagenized Rag2^+/− (solid grey line) and Rag2^−/− (solid black line) populations. Difference in latencies was largely due to leukemias (see Fig. 3). Rag2^−/− controls without ongoing SB mutagenesis were also aged (broken black line). B, Tumor multiplicity (all tumor types) is not significantly different between immune cohorts (P=0.3535, unpaired t-test). Error bars represent one standard deviation away from the mean. C, Tumor distribution of major tumor groups was similar, with the exception of gastrointestinal (GI) tumors and hemangiosarcomas (H.s.). Immunodeficient mice were more prone to developing GI tumors (P=0.0210, Fisher’s exact test) and less prone to developing hemangiosarcomas (P=0.0381, Fisher’s exact test) than immunocompetent mice (Table I).

FIGURE 3. Tumor-free survival and disease severity differ by tumor type.

A and B, Tumor latencies were not different between Rag2^+/− (solid grey line) and Rag2^−/− (solid black line) in hepatocellular carcinomas (A) or keratoacanthomas (B). C, A significant difference in latency was observed specifically in mice with leukemia (P=0.0106, Log-Rank Mantel-Cox test). D, Leukemias were more widely disseminated in Rag2^−/− animals than Rag2^+/− animals (P=0.0034, unpaired t-test).

Similar to latency, tumor multiplicity was also nearly equivalent in both cohorts, with Rag2^+/− and Rag2^−/− mice developing an average of 2.19 and 1.97 tumors per mouse, respectively (Fig. 2B). Only one significant difference in tumor multiplicity was observed when examined by individual tumor type. Specifically, keratoacanthoma (KA) multiplicity was increased in immunocompetent mice, where Rag2^+/− mice developed 1.31 KAs per mouse versus 1.05 in the Rag2^−/− cohort (P=0.03). Similarly, tumor incidence was equivalent in most tumor histologies, with the exceptions of hemangiosarcomas and tumors of the gastrointestinal (GI) tract. Rag2^−/− mice showed significantly decreased susceptibility to hemangiosarcomas and an increased susceptibility to GI tumors (Fig. 2CTable I).

Table I.

Distribution and frequency of SB-induced tumors in aged cohorts.

	Rag2−/− +SB	Rag2+/− +SB	Rag2−/− -SB	Fisher's Exact
	(n=91)	(n=96)	(n=30)
Skin
keratoacanthoma	22	32	-	ns
squamous papilloma	2	3	-	ns
dermal sarcoma	-	1	-	ns
squamous cell carcinoma	2	-	-	ns
trichoepithelioma	-	1	-	ns
melanocytoma	5	5	-	ns
malignant melanoma	2	2	-	ns
Blood
leukemia	27	20	1	ns
marginal zone lymphoma	-	1	-	ns
myeloid leukemia	-	1	-	ns
Liver
hepatocellular carcinoma	26	34	-	ns
hepatocellular adenoma	22	29	-	ns
cholangiocellular adenoma	1	-	-	ns
foci of cellular alteration	2	4	1	ns
Gastrointestinal Tract				P=0.021
stomach neuroendocrine tumor	2	2	-	ns
small intestinal polyp	4	-	-	ns
intestinal adenoma	3	1	-	ns
intestinal adenocarcinoma	2	1	-	ns
colon polyp	1	-	-	ns
intraepithelial neoplasia	1	-	-	ns
Brain
astrocytoma	4	1	-	ns
suprasellar germ cell tumor	-	1	-	ns
spinal neural tumor	-	1	-	ns
Airway
pulmonary adenoma	3	-	-	ns
pulmonary adenocarcinoma	-	3	-	ns
nasal gland adenoma	1	2	-	ns
lung papillary adenoma	4	1	-	ns
lung papillary adenocarcinoma	-	2	-	ns
nasal papillary adenoma	-	1	-	ns
nasal cavity adenoma	1	3	-	ns
nasal cavity adenocarcinoma	2	1	-	ns
Miscellaneous
hemangiosarcoma	8	19	-	P=0.0381
hemangioma	1	-	-	ns
mammary gland adenoma	1	-	-	ns
mammary gland adenocarcinoma	2	2	-	ns
mammary gland carcinoma	-	2	-	ns
prostate PIN	3	3	-	ns
luteoma (ovary)	-	1	-	ns
lipoma	-	2	-	ns
pancreas focal nodular hyperplasia	1	-	-	ns
salivary gland adenoma	-	1	-	ns
salivary gland carcinoma	-	1	-	ns
adrenal gland pheochromocytoma	-	1	-	ns
Harderian gland adenoma	2	3	-	ns

All tumor types produced by our screen and the number of animals with each tumor type are listed. Immunocompromised animals were more susceptible to developing tumors in the gastrointestinal tract (P=0.0210, Fisher’s exact test) and less prone to developing hemangiosarcomas (P=0.381, Fisher’s exact test) than immunocompetent mice.

Genetic Profiling of SB-induced Tumors

One advantage of using SB mutagenesis instead of chemical carcinogenesis is the ability to easily identify insertional mutations. Thus, we sought to determine if host immune status impacted genetic selection in tumors. Analysis of transposon integration sites was performed on the most abundant tumor types observed in both cohorts: leukemias, KAs, and hepatocellular carcinomas (HCCs).

Despite having similar tumor latencies (KAs and HCCs) and multiplicities (HCCs), genetic profiling of the solid tumors revealed some significant differences in transposon-induced mutations (Table IISupplemental Fig. S3, Supplemental Table SI). For example, Rtl1 was mutated with higher frequency in HCCs from immunocompetent mice, with transposons scattered throughout the locus in both transcriptional orientations (Supplemental Fig. S3A). Higher mutation frequency of Rtl1 in the immunocompetent cohort suggests that these mutations might be instrumental in enabling tumor escape from adaptive immunity. Interestingly, KAs did not display any significant difference in mutation spectra. Despite the fact that KA multiplicity was higher in immunocompetent mice, mutation of Zmiz1 (encoding an E3 SUMO-ligase) was the predominate insertion event in both immune cohorts (Table IISupplemental Fig. S3B).

Table II.

Common insertion site genes

Tumor type	Gene	Mutation freq		p-value
		Immuno- compromised	Immuno- competent
Hepatocellular carcinoma	Rtl1	58% (n=33)	82% (n=39)	0.036
Keratoacanthoma	Zmiz1	85% (n=26)	88% (n=41)	0.727
Leukemia †	Pten	84%	6%	----
	Uba1	65%	0%	----
	Nr3c1	19%	0.9%	----
	Rfx7	15% (n=26)	0% (n=232)	----

Fisher’s exact test was used to determine p-values in table.

^†Immunocompetent tumor frequencies combine data from previous leukemia screens (11, 33, 34).

By contrast, Uba1 was mutated only in leukemias from immunocompromised mice. Transposon insertions were clustered in intron 1 in the same transcriptional orientation as Uba1 likely resulting in overexpression of the E1 ubiquitin-conjugating enzyme (Supplemental Fig. S3D). Upon closer examination, subclonal Uba1 insertions in intron 1 were also detectable in a subset of KAs and HCCs from Rag2^−/− mice, indicating that Uba1 overexpression could be important for development of multiple tumor types in a Rag2^−/−but not Rag2^+/−background (Fig. 4A).

FIGURE 4. Syngeneic transplant of Uba1-expressing cells does not result in immune clearance.

A, Two independent Uba1 insertions (pictured top) were detected in keratoacanthomas (KA) (n=3), leukemias (Ly) (n=3), and hepatocellular carcinomas (HCC) (n=4) by PCR amplification of the transposon/Uba1 junction. Solid tumors were taken from animals that were leukemia-free at time of necropsy. B, Western analysis of B16-F0 cells stably transfected with a Uba1-overexpressing vector revealed these cells had 1.8-fold higher expression of Uba1 compared to cells expressing the empty vector control (Uba1 normalized to β-tubulin, HepG2 cells served as positive control). C, Western analysis of total protein from B16-F0 cells (+/− MG132 proteasomal inhibition) shows similar levels of ubiquitinated proteins in Uba1-overexpressing and empty vector control cells. D, Quantitative RT-PCR shows that Uba1 mRNA levels are comparable in normal mouse tissues and B16-F0 melanoma cells. E, Uba1-overexpressing B16-F0 cells proliferate at a similar rate to empty vector control cells (error bars represent standard deviation). F, Lung colony counts from B16-F0-injected immunocompetent recipients. Mice that received the empty vector had an average of 20.2 lung colonies per animal, while lungs from animals that received Uba1-overexpressing cells had an average of 8.1 colonies per animal. This difference was not statistically significant (P=0.1281, unpaired t-test).

Syngeneic Transplantation of Uba1-overexpressing Cells

Because Uba1 insertions were detected in multiple tumor types from the immunocompromised background (Fig. 4A), we hypothesized that Uba1 overexpression in a wild-type background might trigger immunological rejection. To test this hypothesis, we chose to utilize a well-characterized syngeneic transplant model of melanoma using B16-F0 cells (13). B16-F0 cells were found to express levels of Uba1 comparable to normal mouse tissues, including liver, skin and thymus (Fig. 4D). Cells were stably transfected with a Uba1-overexpressing construct or empty vector. Protein expression was nearly 2-fold higher in cells transfected with Uba1 than in empty vector control cells (Fig. 4B). Overexpression of Uba1 did not impact global ubiquitination as measured by western blot analysis of total protein, despite the fact that Uba1 acts as an E1 ubiquitin-activating enzyme (Fig. 4C). Importantly, Uba1 overexpression also had no effect on proliferation rate of B16-F0 cells in culture (Fig. 4E).

Stably transfected cells were injected into the tail vein of wild-type C57BL/6 mice. Eighteen mice received Uba1-overexpressing B16-F0 cells, and 15 received empty vector control cells. Recipients were aged for three weeks, at which point animals were euthanized, and the number of melanoma colonies in the lungs were counted. If Uba1 overexpression triggers an immune response, we would expect fewer colonies in the lungs of animals that received Uba1-expressing cells compared to control cells. We found that lung colony numbers, while trending toward a reduction, were not significantly different (Fig. 4F), leading us to conclude that Uba1 overexpression does not produce a strong antigenic response in this context.

Discussion

A recent report using 3’-methylcholanthrene (MCA) chemical carcinogenesis found that the adaptive immune system is responsible for a process called tumor immunoediting, whereby tumors that are initially recognized by lymphocytes acquire mutations that eventually allow escape from immune suppression (7). Our current study sought not only to independently verify these findings, but to also identify key mutations involved in immune escape using a forward genetic screen. To this end, we generated tumor-prone mice undergoing ubiquitous Sleeping Beauty (SB) transposon mutagenesis on either a wild-type or Rag2^−/− background. The identification of driver mutations that are selected specifically in the presence of a functional adaptive immune system (i.e. wild-type vs. Rag2^−/−) may provide a better understanding of the cellular mechanisms used by tumor cells to escape immune detection and could greatly improve the efficacy of current immunotherapies.

Based on the results of Koebel et al. we expected to observe decreases in tumor latency and tumor multiplicity in our immunocompromised mice compared to their immunocompetent counterparts. However, we observed only minor differences in latency and multiplicity that were each isolated to a single tumor type, suggesting that lymphocytes do not strongly suppress tumor development in our model. Data from two recent papers may help to explain this discrepancy. First, exome analysis of MCA-induced sarcomas revealed that the mutation rate in these tumors is substantially higher than even human tumors that exhibit a hypermutator phenotype (16). Another recent publication showed that only strong (e.g. engineered) tumor-specific antigens were sufficient to trigger lymphocyte-mediated immunosurveillance (17).

Thus, it appears that strongly antigenic mutations may not occur frequently in tumors with lower mutation rates. In fact, the majority of identified endogenous tumor-specific antigens in human cancers are not strongly immunogenic, and current cancer vaccination approaches also require adjuvants to encourage an adaptive immune response (18, 19). Consequently, immunoediting mechanisms in human tumors may not rely solely upon adaptive immunity. Furthermore, if the development of antigenic signals is more likely in tumors with high mutation rates, then the low-copy transposon mouse strains used in our study may not generate a large enough mutation burden to produce strong antigens. This could explain the lack of widespread phenotypic differences in tumor development between immune cohorts.

The MCA and SB models differ in the nature of mutation produced in addition to mutation rates. Immunogenic endogenous tumor antigens in humans can be split into two classes: unmutated proteins with dysregulated expression, and somatically mutated neoantigens (20–22). SB-induced insertional mutations function to alter transcription of a target gene, and would be expected to produce tumor antigens akin to those with dysregulated expression. In contrast, MCA mutagenesis primarily causes point mutations, likely resulting in production of neoantigens (23, 24). It is important to note that immune recognition of either unmutated or mutated antigens may differ mechanistically. Nonetheless, altered expression levels of unmutated proteins can be immunogenic, and our SB model should be able to identify these antigens. In fact, transposon insertion analysis revealed that the mutational status of Rtl1 in hepatocellular carcinomas (HCCs) from Rag2-null and wild-type cohorts was disproportionate, hinting that the adaptive immune system does exert some selective pressure on tumor cells, but that selection is not strong enough to cause widespread differences in tumor latency and multiplicity.

We also observed that Uba1 was frequently mutated in Rag2^−/− leukemias. Unfortunately, interpretation of this mutation signature is clouded by the fact that lymphocytes are directly affected by the Rag2 mutation (10). While leukemia incidence was equivalent in both cohorts, leukemias were more widely disseminated and often the only tumor present in immunocompromised animals, such that leukemia was the primary cause of morbidity (Fig. 3D). This is in contrast with the majority of Rag2^+/− animals, which were generally older and often had multiple solid tumors in addition to leukemia. Therefore, leukemias from the Rag2^−/− cohort are not necessarily equivalent to those from the Rag2^+/− group based on tumor distribution and disease severity.

However, subclonal Uba1 insertions were also observed in HCCs and keratoacanthomas (KAs) in immunocompromised mice, leading us to hypothesize that Uba1 overexpression is immunogenic. It is known that loss of UBA1 leads to defects in DNA double strand break repair, and that pharmacologic inhibition delays tumor growth in a mouse model of leukemia (25, 26). One can imagine that global dysregulation of ubiquitination pathways or deficiencies in DNA repair might contribute to the production of tumor antigens. We chose to test Uba1 immunogenicity with the B16 melanoma model, since there are not many other C57BL6 syngeneic models available. Unfortunately, syngeneic transplant of Uba1-overexpressing B16 melanoma cells failed to support our hypothesis, indicating Uba1 overexpression alone is not enough to increase B16-F0 immunogenicity. One caveat of this experiment is our choice of transplant model. Specifically, Uba1 expression might lead to immune clearance of HCCs and KAs, but not melanomas. B16-F0 cells express endogenous Uba1 transcript levels consistent with normal mouse tissues. However, a two-fold Uba1 increase may not be sufficient to elicit a strong immunological response. Alternatively, B16-F0 cells may have already acquired mechanisms to evade an immune response such that introduction of Uba1 expression was not enough to overcome these other evasion strategies. Yet, our data trends toward significance even without additional genetic modulations or adjuvant stimulation. This might indicate that Uba1 overexpression is very weakly immunogenic, and that immunogenicity could be increased by layering on additional mutations.

While our data show that the adaptive immune system can shape the genetic profile of spontaneous tumors, it does not seem to heavily impact the transformation rate. It is important to remember that Rag2-null mice retain functional natural killer (NK) cells, albeit trending toward slightly lower numbers (Fig. 1C). It is unclear whether this subtle difference is biologically significant. NK cells have been shown to be instrumental in tumor immunosurveillance in some contexts (14). Perhaps NK cells in our immunocompromised mice are able to compensate for the lack of mature B- and T-cells, such that tumors were cleared by NK cells instead. Unidentified lymphocytes were observed in sections of solid tumors from Rag2^−/− mice, despite a confirmed lack of mature B- or T-cells (Supplemental Fig. S2Fig. 1B). It would be interesting to see whether similar results would be obtained if this screen were repeated using a mouse strain that also lacks NK cells.

Our data also indicate that immune recognition mechanisms vary with tumor type. For example, we observed that immunocompromised mice were more susceptible to developing tumors in the GI tract, and less susceptible to hemangiosarcoma development (Fig. 2CTable I). It is possible B- and T-cells are key players in immunoregulation of sarcomas and GI malignancies, but play less important roles in suppressing growth of other cancer types.

Interestingly, only one of our control Rag2^−/− mice lacking SB components (n=30) developed cancer (leukemia). This is also in disagreement with previous reports that Rag2^−/− mice are inherently more tumor-prone than wild-type counterparts (6), an inconsistency that may partially be explained by the fact that our animals were housed in a specific pathogen-free facility. It is known that increased innate inflammation, perhaps in response to infection, promotes tumorigenicity in humans (27). It has also been suggested that one way in which tumors escape immune destruction is by modifying cytokine secretion to create a pro-tumor environment (28). Thus, it is important to understand the role inflammation may be playing in tumor immunoediting. Unfortunately, separating inflammation- from lymphocyte-driven immunoediting mechanisms will be impossible using the MCA model of sarcomagenesis, as MCA administration introduces local inflammation (29), whereas SB-induced tumors are spontaneously produced in the absence of local inflammation. We propose that our model could be used in the future to test the relative contribution of chronic inflammation to the tumor immunoediting processes.

Skin tumors (i.e. KAs) were one of the most commonly produced tumor type in our SB model. Immunosuppressed humans have a nearly 65-fold increased risk for developing nonmelanoma skin cancer (2, 3). Based on this, we expected that mice in the Rag2-null group would produce more KAs, but tumor multiplicity was about half that of immunocompetent mice instead. In humans, pathogens like human papillomavirus and UV exposure increase risk for nonmelanoma skin cancer (30). It may also be important that both infection and UV damage, components that are lacking in our model, can result in tumor-promoting inflammation (31, 32).

This study is the first to investigate the role of the adaptive immune system in tumor immunosurveillance using a large-scale forward genetic screen in mice. Moreover, our experiment allowed the study of immunosurveillance in a variety of spontaneous tumor types, including carcinomas. This differs from the majority of immunosurveillance studies, which have relied heavily upon inflammation-inducing sarcoma and transplant models. In our model, we report that the adaptive immune system does not substantially modulate tumor phenotypes, but does influence genetic selection of driver mutations. Our findings underscore the complexity of the immune response to cancer, and indicate that immune recognition mechanisms are not necessarily conserved across tumor models.