Background
There has been rapid growth in development of vaccine approaches that direct the immune system to target tumor neoantigens for treatment [
1]. The most successful approaches have identified patient specific neoantigens to produce personal cancer vaccines (PCVs) that are in multiple clinical trials [
2‐
5]. Generation of PCVs relies on sequencing the DNA and RNA of the tumor to discover and validate mutations creating neoantigens. Algorithms or in vitro testing determine which mutations are likely to elicit an anti-tumor immune response and the vaccine is manufactured as peptides, DNA, RNA or viruses presenting the neoantigens [
6]. The patient is then administered the unique vaccine, often in combination with another immunotherapeutic. Early reports on clinical trials applying such a system are encouraging in that the vaccines appear safe and may have signs of efficacy [
2‐
5]. However, the PCV approach involves considerable time to produce the vaccine and may be costly. A pre-made vaccine would not suffer these limitations but may be less effective. Here we compare both types of vaccine in the mouse mammary tumor model.
Besides the time and cost issues, the biggest limitation of PCVs may be the dearth of DNA mutations in most tumors [
6‐
8]. While some tumors have 100–1000s of non-synonymous single nucleotide variants (SNVs), a large number have few. For example, in a recent report on PCVs to melanoma, two out of 10 enrolled patients did not have the high mutation rates normally observed in melanoma, disqualifying them for vaccination [
5]. The majority of other tumor types rarely have large numbers of mutations [
7]. Since less than 5% of the mutations are expected to be immunogenic, this is a severe limitation on the breadth of application of the PCV strategy [
8]. This has spurred exploration of other neoantigen sources focused on alternative splicing and from non-coding regions of the genome [
8,
9].
In contrast to SNV neoantigens, we have found frameshift neoantigens (FS) produced by mistakes in RNA production and processing are a rich neoantigen source [
10,
11]. Increasing evidence demonstrates that Fs neoantigens are efficacious, highly immunogenic antigens [
12]. While rare in DNA, they are frequently produced by transcription through microsatellites in coding regions or by mis-splicing of exons, processes that are highly error-prone in tumors relative to normal cells [
13‐
15]. We have shown that these RNA-based FS peptides (FSP) are protective in mouse tumor models and that these FSP elicit antibody and T-cell responses [
10,
11]. Based on these findings we designed a small array of 788 human FSPs that were predicted informatically from exon mis-splicing or indels from transcription through microsatellites and used these arrays to screen 2 different murine cancers for personalized and shared immunoreactive FSP neoantigens. We observed that many FSP antigens were recurrent at various frequencies across individuals within a tumor type. We then designed a shared antigen, FAST vaccine for both tumor types as well as a PCV for each mouse in order to test the relative efficacy of these two FSP neoantigen vaccines in the 4 T1 metastatic breast cancer mouse model. This murine model is widely used to assess novel therapies against stage IV human breast cancer due its poor immunogenicity and ability to metastasize, shared characteristics of human advanced cancers [
16,
17]. Metastatic breast cancer is the second most deadly cancer for US women, affecting approximately one in eight women in their lifetimes. With 41,000 deaths in 2018, current treatments and early detection have only decreased mortality rates by a modest 1.3% during 2011–2017 [
18,
19]. Despite many advances in cancer research and improvements in patient care, a more effective treatment is still desperately needed.
Discussion
In this study, we designed PCVs and an off-the-shelf, shared neoantigen vaccine (BC-FAST) composed of frameshift neoantigens against a triple-negative breast cancer cell line (4 T1) and compared their efficacy and immune responses. A peptide microarray composed of FS neoantigens predicted to be generated by errors in RNA processing, either from exon mis-splicing or transcription through INDELs in microsatellite regions, was used to identify neoantigens for both vaccine types. When used as monotherapies, both vaccines conferred comparable protection to primary tumor growth in a therapeutic model and both reduced lung metastases relative to a Mock vaccine control. Importantly, a shared antigen vaccine designed against a pancreatic tumor cell line (PC-FAST) and a vaccine composed of non-reactive (NR) peptides from the array did not reduce primary tumor growth. ICI co-treatment did not enhance protection to the primary tumor nor did it reduce metastasis for PCVs or BC-FAST vaccinated mice relative to vaccine alone. All vaccines elicited strong T-cell immune responses to their component antigens and to tumor cells, with variable antibody responses to the antigens.
The primary goal of this study was to test the relative efficacy of PCVs and shared frameshift antigen vaccines in a breast cancer model. This comparison was possible due to the predictable nature of FS antigens from RNA processing errors. This predictability enabled pre-synthesis of a small library of possible candidate antigens that could be screened against patient sera to discover immunogenic antigens that are either specific to the patient or shared across the group [
11]. This made it possible to quickly identify antigens for rapid PCV formulation for each mouse. This approach would not be feasible for neoantigens identified from DNA mutations as sequencing and vaccine synthesis could not be completed before mice succumbed to the tumors. Since there are only ~ 200 K possible FSP from exon mis-splicing, we were able to previously screened microarrays containing all possible FS against an number of cancer types which allowed us to choose 200 peptides to be synthesized for this effort [
11]. This system enabled PCV identification and vaccine production for each mouse in less than 5 days, as opposed to a month or more if from DNA mutations. Choosing commonly reactive peptides enabled BC-FAST vaccine production before challenge. For development of FAST vaccines for clinical application the full 200 K peptide arrays would be used to select the best peptides for production of a standard, off-the-shelf vaccine.
In humans the optimal application of the personal cancer vaccine approach would be that at diagnosis of cancer the process of production of the PCV would initiate. After sequencing of the tumor, assuming the patient had enough mutations to be eligible, the process would take at least 1–3 months before administration of the vaccine [
2‐
4]. During this time the tumor could evolve and advance. In contrast, with the FAST vaccine all patients could be administered the vaccine on diagnosis. The protocol we used to compare the FAST vaccine and PCV in mice was designed to reflect this difference in clinical protocols. The FAST vaccine, because it was pre-made, was administered on day 7. For the PCV the process of producing the vaccine was initiated on day 7 and the vaccine delivered on day 12, the fastest we could design and produce the vaccine.
These results demonstrate several possible advantages of FS neoantigens from RNA processing errors for cancer vaccines. While it is possible to discover PCV neoantigens with this approach, the more interesting aspect is the relative high occurrence of shared neoantigens across mice and in patient populations [
10,
11], in contrast to neoantigens from DNA mutations. This could enable development of off-the-shelf vaccines for specific cancers overcoming the logistical challenges of PCV discovery and manufacturing that cost patients precious time to treat. It is also noteworthy that these neoantigens were readily identifiable from a “cold tumor” with low mutational load [
7] suggesting it could be possible to develop FAST vaccines against other cold tumors.
Frameshift peptide microarrays as an antigen discovery technology, while an indirect measure of the presence of neoantigen in the tumor, are a direct measure of the immunogenicity of the neoantigens in the patient. This is in contrast to PCV approaches which identify the DNA-based neoantigen and then predict immunogenicity with mixed success at best [
2‐
5]. It should be noted that neoantigens can evolve over time as was recently noted in colorectal cancers [
26] and in pancreatic cancers which have different FSP neoantigen profiles in early stage and late stage [
10,
11]. The dynamic neoantigen profile could adversely affect PCVs while an off-the-shelf shared vaccine should not be as affected by this process.
In this study, PCVs and FAST vaccines were discovered from a small library of 200 FSPs. We have shown with peptide microarrays of all 220,000 possible FSP, that ~ 20% of all peptides show reactivity for any particular cancer with many public antigens that are shared across 10% to up to 70% of patients [
10,
11]. The shared FSP are not strictly tumor type specific and the amount of overlap between tumor types varies considerably, similar to the results of this study. The peptides chosen for the 4 T1 BC-FAST vaccine were shared between 29 and 83% of the 24 mice used. This was also true for the PC-FAST vaccine for the pancreatic cell line. As we had seen for the human samples, some FS peptides were reactive in both the 4 T1 (breast) and KPC (pancreatic) cell lines, which could explain the partial protection of the PC-FAST vaccine in the 4 T1 mouse model. In contrast to the BC-FAST vaccine which was potent as a monotherapy, the PC-FAST vaccine required combination with ICIs to protect against the primary tumor. This could be due to the PC-FAST vaccine containing only five 4 T1 reactive antigens versus the 10 antigens contained in the BC-FAST vaccine. This implies that the small number of 4 T1 antigens in the PC-FAST are not enough to limit the primary tumor but can reduce metastases and are more effective in combination with ICIs.
Immune analysis of FSPs selected for vaccination revealed that all groups had broad antibody responses to FS neoantigens after tumor challenge, consistent with the antigen discovery by FSP microarrays. It is interesting that the vaccines did not boost the antibody response but this is consistent with reports that the poly (I:C) adjuvant plus peptides preferentially enhances cellular immune responses [
27,
28]. Importantly, antibody levels did not correlate with tumor size, as demonstrate by groups with larger tumors did not have higher antibody responses to all peptides tested. In contrast to the antibody response, T-cell analysis revealed a weak pre-existing T-cell reactivity to the FSPs in the PCVs, BC-FAST, or PC-FAST vaccines. However, vaccination induced a strong T-cell response whether T-cells were stimulated by peptides or the 4 T1 cells themselves. Others have also observed that most neoantigens chosen for vaccines have low or no pre-existing T-cell responses [
2‐
5,
29]. While more mice remained tumor free in the BC-FAST and PCV vaccinated groups than in the controls, these tumor free mice grew tumors when re-challenged with twice the 4 T1 dose of cells. These secondary tumors did grow significantly slower in the BC-FAST group compared to the other re-challenged mice. One interpretation of these results is the BC-FAST induced enough T-cell response to kill or contain the initial dose of 4 T1 cells in most mice. However, that level of immune response could only slow tumor growth in the re-challenge with more tumor cells. For both FAST and PCVs it will remain a challenge to create enough effective T-cells to kill all the tumor cells.
Additionally, the shared frameshift peptides selected for the BC-FAST induced polyfunctional antigen-specific T cells, both CD4+ and CD8+, in naïve mice, indicating the capability of this peptide vaccine to elicited a protective and functional immune responses. Also, the pre-existing anti-FSP T cell immunity in tumor bearing mice (mock) presented a dysfunctional phenotype. On the other hand, the vaccination boosted FS antigen-specific effector T CD4+ and CD8+ cells, important for an effective antitumor immune response [
30]. A sustainable and optimal effector antitumor response has been associated to a polyfunctional phenotype cells [
31]. Most of our vaccine-induced T cells were monofunctional, suggesting repeated vaccination may be required to maintain a protective immune response.
We conducted these studies to determine if there may be value in pursing the FAST vaccine concept. The model we used has limitations relative to the clinical situation. We used the immune reaction to cell lines to define the FAST vaccines. In patients the variation in tumors will be much more than in the cell lines. A FAST vaccine for breast cancer would have to be defined across widely variable tumor types. It will be of interest to test different FAST vaccines against different mouse mammary tumor lines. Recently, we showed that for 5 different human cancers, it may be possible to define a FAST vaccine for each tumor type [
11]. We also used inbred mice, where in humans the immune response to the vaccine will be much more variable, as has already been evident in personal cancer vaccine trials [
2‐
5]. The personal vaccines were developed from RNA sourced FSPs, not DNA sourced neoantigens. Considering the potential advantages of FAST vaccines, the pre-clinical data presented here justify their further investigation.
Methods
Mice and tumor cell line
All mouse procedures and protocols were approved by the Arizona State University Institutional Animal Care and Use Committee (protocol #1568R). Animals were purchased from The Jackson Laboratory and housed at ASU specific pathogen free (SPF) at the Interdisciplinary Science and Technology Building 1 (ISTB1) administrated by the Department of Animal Care and Technologies (DACT). Mice were caged in a ventilated Thoren with 250 cages or less, room temperature of 74 ± 2 °F (23 ± 1 °C), light cycle of 12 h of light /12 h of dark and 10–15 air changes per hour. The 4 T1 cells were purchased from ATCC (ATCC® CRL-2539™) in 2006, cultured as specified and aliquots stored at − 180 °C until use. For the experiments, cells were used with no more than 5 passages in vitro and authentication was only made by the supplier (ATCC). Serum samples from the KPC pancreatic tumor mouse model were obtained from C57/BL6 mice challenged by Dr. Haiyong Han’s team, at TGen.
4 T1 BALB/c breast cancer mouse model and vaccination regimen
4 T1 tumor cells were grown in RPMI-1640 culture medium supplemented with 10 units/ml penicillin-streptomycin (Sigma Aldrich) and 10% fetal bovine serum (FBS) at 37 °C in 5% CO
2 until 80% confluence. Cells were detached with trypsin (TrypLE, Thermo fisher Scientific), washed 3 times and suspended in sterile PBS 1X at 5 × 10
3 cells/ml. 4 T1 cell suspension (500 cells in 100 μl PBS 1X) was inoculated subcutaneously (s.c.) into the right flank, near the mammary glands, of each female BALB/c mice (6–8 weeks-old) [
19,
32]. Mice were randomized into the treatment groups (
n = 10 mice/group) (5 mice/cage). Tumors were measured twice per week. For the FAST vaccine experiments, 4 T1-tumor bearing mice were immunized at day 7 and boosted at day 14. For the PCV experiments, 4 T1-tumor bearing mice were immunized at day 12 and boosted at day 19. Both vaccines were administrated subcutaneously in the left flank (opposite to the tumor injection). Each immunization was composed of 50 μg vaccine peptide pool (5 μg/FS peptide) and 50 μg Poly (I:C) (Polyinosinic–polycytidylic acid potassium salt, Sigma Aldrich) in 100 μl sterile PBS 1X. PCV peptides were conjugated to KLH according to the manufacturer’s protocol (Imject Maleimide-Activated mcKLH™, Thermo Scientific) and purified by dialysis. For the immunotherapeutic treatment, mice were treated with anti-mouse PD-L1 (200 μg/mouse, clone: 10F.9G2, BioXCell, West Lebanon, NH) and anti-mouse CTLA-4 (100 μg per mouse, clone: UC10-4F10–11, BioXCell, West Lebanon, NH) on days: 8, 15 and 22 (for the FAST vaccines) and days: 13, 15, 20 and 22 (for the PCV vaccines). The control group was immunized with PBS 1X at the same schedule and injection volumes as the vaccine groups. All tumor challenges, immunizations and tumor measurements were performed in the morning. Tumor growth was monitored and the volume was calculated as: tumor volume = (length × width
2)/2. Animals were euthanized with carbon dioxide overdose (CO
2) followed by confirmation with cervical dislocation when tumor volume reached 2000 mm
3 or when animals became moribund with severe weight loss or ulceration. Values were reported as means ± S.E., with statistical analysis performed using Prism 8.0 (GraphPad Software). All statistical analyses were performed in: 8/10 mice in mock group, since 2 animals did not develop tumor as described before by [
19]; 10/10 mice in all other treated groups.
FS peptide array and vaccine peptide selection
To identify FSP for both vaccine approaches, we collected pre-immune sera and sera from 7-days post 4 T1 tumor challenge (
n = 24 mice) or KPC tumor challenge (
n = 18). A FS peptide microarray similar to our previous FS peptide microarray [
10] with 788 peptides representing 200 Fs antigens as 20mer peptides was used for these studies. The FS antigens were from FS transcripts that could be generated by the errors of the RNA processing: insertion and deletion (INDEL) of the microsatellite (MS) region during the transcription and mis-splicing of the exons. The majority of these FS antigens were predicted MS INDEL FS antigens from mono-repeat MS regions (minimum repeat length was 7 nt). MS FS antigens were selected based on the length of the homopolymer and the length of the predicted FSP length. The FS antigens with longest microsatellite repeats were selected, since they have higher INDEL rates during the transcription [
14] and FSPs longer than 17 amino acids were chosen. The remaining FS antigens were selected from human EST library analysis where FS with high frequencies in tumor EST libraries and low frequencies in normal EST libraries were selected [
33]. Peptides were synthesized (Sigma-Aldrich, St. Louis, MO and WatsonBio, Houston, TX) without purification and spotted (Applied Microarrays, Tempe, USA) on NSB-9 amine slides (NSB Postech, Seoul, South Korea) using our previously developed methods [
11]. Dilute sera was spotted on filter paper (903, Whatman) and dried overnight at room temperature. A 6 mm spot was punched and the filter paper was added to 150 μl blocking buffer (PBS 1X, 0.05% Tween-20, 3% BSA (bovine serum albumin)) with
E. coli extract (1 mg/ml) (dilution 1:200) and probed on peptide array for 1.5 h at room temperature (R.T). Arrays were washed 3 times with PBST (PBS 1X, 0.05% Tween-20), incubated with 200 μl of 3–5 nM goat anti-mouse IgG-AlexaFluor 647 (Thermo Scientific, Waltham, MA), washed and scanned on an Innoscan 910 (Innopsys, Carbonne, France) at 80% gain at 647 nm excitation and 20% gain at 532 nm excitation. Microarray data was image-processed with GenePix Pro-6.0 (Molecular Devices, Sunnyvale, CA) and exported to Excel prior to analysis with JMP 12 (Statistical Discovery Software). Raw intensities were median normalized by slide and reactive peptides were defined as post-challenge signal two times higher than the standard deviation of the naïve mice (average). FSP positive rates were calculated for each positive peptide and the top 10 peptides with highest incidence chosen to compose the FAST vaccine. For PCVs, from the positives peptides for each mouse we selected 10 peptides with highest fluorescence intensities and that represented different frameshifts antigens. For the non-reactive FS vaccines, we selected 10 non-reactive peptides for each mouse.
To evaluate the 4 T1 clonogenic spontaneous metastasis, lungs were aseptically removed at the time of sacrifice as determined by the endpoints criteria (tumor size (larger than 2000 mm
3) and/or clinical signs of illness (lethargy, hunched posture, ulceration and etc.), ranging from 40 to 78 days post-challenge. Lungs were minced and digested with a solution of 10 mg/ml collagenase type I and 10 mg/ml hyaluronidase for 20–30 min at 37 °C under slow rotation. The suspension was filtered through 70 μm cell strainers and washed two times with complete RPMI-160 culture medium. Cells were suspended in the same medium supplemented with 60 μM 6-thioguanine (Sigma Aldrich) (10 ml/plate) and cultured in petri dishes for 14 days at 37 °C and 5% CO
2. Plates were fixed with methanol for 5 min, carefully washed with water and stained with 0.03% methylene blue and counted. Data are expressed as total number of metastatic colonies per mouse [
32].
Peptide ELISAs
Specific IgG antibody responses in the sera of immunized mice were determined by ELISA in 96-well MaxiSorp plates (Nunc) coated with each vaccine peptide. Peptides were coated (10 μg/ml peptide/well) in carbonate-bicarbonate buffer (pH 9.6), overnight at 4 °C. Peptide-coated plates were blocked with blocking solution (PBS 1X, 0.05% Tween-20, 3% BSA) for 1.5 h a 37 °C, and washed thrice with PBS-T 1X (PBS 1X, 0.05% Tween-20). Mouse sera at the endpoint, either pooled sera by group or from individual mice, was diluted 1:200 in blocking solution, added to the plates, and incubated for 1.5 h at room temperature. Plates were washed and bound IgG was detected with horseradish peroxidase-conjugated anti-mouse IgG (Bethyl Laboratories Inc.) followed by TMB Substrate Solution (Thermo Fisher Scientific). The reaction was stopped with 0.5 M HCl and the final absorbance at 450 nm was measured in a plate reader (SpectraMax 190, Molecular devices). For the final absorbance, the pre-immune response was subtracted.
IFN-γ ELISPOT
At endpoints, the vaccinated mice were euthanized and the spleens were aseptically removed, minced and filtered with 100 μm screens in complete RPMI culture medium (10% FBS, HEPES, L-glutamine, ß-mercatoethanol, sodium pyruvate and penicillin/streptomycin). Red blood cells were removed by lysis with BD Pharm Lyse™ (BD biosciences), splenocytes were suspended and then counted. Cells were diluted in freeze medium (complete RPMI with 10% DMSO) and stored in liquid nitrogen until use. ELISPOT plates (BD biosciences) were coated with anti-mouse IFNγ (10 μg/ml) as described by manufacturer and incubated overnight at 4 °C. Plates were washed and blocked with complete RPMI for 2 h at 37 °C with 5% CO2. Splenocytes were thawed and diluted to 5 × 106 cells/ml in complete RPMI medium, and 100 μl cells added to each well and stimulated with FS peptide pools (3–4 FS antigens/well) (1 μg/well each FS peptide) or 4 T1 tumor cells (1 × 105 cells/well). As a negative control, splenocytes were stimulated with medium only. Concanavalin A (Sigma Aldrich) was used as positive control. Plates were incubated 20–24 h for the peptide stimulation and 72–96 h for tumor cell stimulation, at 37 °C with 5% CO2. Plates were washed, incubated with biotinylated anti-IFN-γ according to manufacturer’s protocol, developed with AEC substrate set (BD biosciences) and spots counted by the AID EliSpot Reader System (Autoimmun Diagnostika GmbH, Germany). Final numbers are represented as sum of the spots for the all vaccine peptides.
Flow cytometry
Frozen splenocytes prepare as described before were thawed and washed twice with complete RPMI culture medium. Cells were counted and prepared to 1 × 10
7 cells/ml, and one million cells were re-stimulated in vitro as follows. Lymphocytes were identified by forward and side scatter, and dead cells and doublets were excluded. Cells were stained extracellularly with amine reactive viability dye (Ghost Dye™ Red 780) (TonBio Biosciences) and fluorochrome-conjugated Abs specific for: CD8a (PerCP-Cy5.5, clone 53–6.7), CD4 (PE, clone GK1.5), IFN-γ (APC, XMG1.2), Granzyme B (FITC, clone NGZB), IL-2 (PE-CF594, clone JES6-5H4), TNF-α (BV421, Clone MP6-XT22), and PD-1 (PE-Cyanine7, clone J43). Cells were surface stained ex vivo, then fixed and permeabilized for intracellular staining (Fixation/Permeabilization Solution Kit with BD GolgiStop; BD Bioscience). For the intracellular cytokine analysis, cells were stimulated for 3–4 h with: FS peptide pools (3–4 FS antigens per pool), or PMA (50 ng/ml) and ionomycin (500 ng/ml) or medium only in the presence of GolgiStop (BD Bioscience), as recommended by the manufacturer. Data were acquired on an Attune NxT Flow cytometer (Thermo Fischer Scientific) and analyzed with FlowJo™ v10.6.1 software. Gating strategy were confirmed on unstimulated control samples or fluorescence minus one controls, as appropriate. Presentation of cytokine production was performed using SPICE version 6.0 software (National Institute of Allergy and Infectious Diseases, National Institutes of Health) [
34].
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.