Targeting public neoantigens for cancer immunotherapy

Abstract

Several current immunotherapy approaches target private neoantigens derived from mutations that are unique to individual patients’ tumors. However, immunotherapeutic agents can also be developed against public neoantigens derived from recurrent mutations in cancer driver genes. The latter approaches target proteins that are indispensable for tumor growth, and each therapeutic agent can be applied to numerous patients. Here we review the opportunities and challenges involved in the identification of suitable public neoantigen targets and the development of therapeutic agents targeting them.

The aim of cancer immunotherapy is to engage the immune system against targets that are present in cancer cells but not normal tissues. One such class of targets are mutation-associated neoantigens, which arise when somatic mutations generate altered peptides that are processed and presented by the major histocompatibility complex (MHC) on the cell surface¹. Such MHC-presented mutant peptides can be recognized by the immune system as distinct from their wild-type counterparts by several mechanisms (Fig. 1a–c). For instance, the altered peptide sequence can allow the mutant peptide to be processed differently for presentation, cause differential mutant versus wild-type peptide binding to the MHC, alter contacts with a T cell receptor (TCR) or change peptide conformation and the overall structure of the peptide–MHC–TCR binding interface. Because they are present exclusively in cancer cells, mutation-associated neoantigens are attractive targets for precision immunotherapies including vaccines, antibodies and cellular therapeutics.

Fig. 1 |. Generation and immune recognition of public neoantigens.

a, Amino acid alterations can enable a mutant peptide to be processed differently for presentation or to bind to the HLA, whereas the wild-type peptide does not bind. b, Differential contact with the TCR by a mutant amino acid residue allows discrimination between the mutant and wild-type peptides within the HLA. c, The overall structure of the peptide–HLA–TCR binding interface is altered through altered conformation of mutant peptide binding, distinguishing the mutant peptide from the wild-type peptide. d, Genetic mutations produce changes in the amino acid sequences of proteins. These proteins with altered sequences can be degraded by the proteasome and the resulting peptides can be processed and presented by HLA on the cell surface. The genes listed are those for which public neoantigens have been identified with defined HLA restriction and strong evidence for endogenous presentation (Supplementary Table 1).

Cancer driver mutations, in oncogenes or select tumor suppressor genes, tend to be localized in genome hotspots that change protein function and are recurrent among patients². Altered peptides produced from these mutations and presented by common human leukocyte antigen (HLA) alleles (the human MHC molecules) can therefore yield neoantigens that are shared among individuals whose tumors harbor the same genetic alterations and HLA, demarcating them as public (Fig. 1d). Conversely, private neoantigens are generated from passenger mutations or non-recurrent driver mutations that are observed in individual patients with cancer. The majority of driver gene mutations do not yield neoantigens that are presented by common HLAs, and the vast majority of presented neoantigens are private^3–5. Thus, identifying the subset of patients for whom public neoantigens are relevant targets is a major challenge. Nevertheless, the ability provided by the expansion of standard-of-care next-generation sequencing to screen patients affords the opportunity for off-the-shelf precision immunotherapies against public neoantigens that would be broadly applicable to many patients and would have tremendous advantages in scalability relative to targeting private neoantigens (Fig. 2). Here we discuss the rationale and recent progress in discovering and targeting public neoantigens, as well as future prospects for developing relevant immunotherapies.

Fig. 2 |. Strategies to target public neoantigens.

Sequencing of tumor specimens from patients enables mutation identification and HLA typing. Patients can then be matched to appropriate therapies for targeting their identified public neoantigens. These therapies include vaccines, TCR-T cells, CAR-T cells and bispecific antibodies.

The rationale for targeting public neoantigens

Current immunotherapeutic approaches against specific neoantigens, such as vaccines and autologous cell transfer, typically rely on a multistep development process that is personalized to each patient and may target non-recurrent private as well as putative public neoantigens¹. The logistical and financial realities of producing personalized therapies for each patient present a substantial obstacle to widespread accessibility⁶. The extended production timeline also increases the potential for disease progression before treatment can be initiated.

A common benefit of targeting public and private neoantigens is the cancer cell specificity conferred by the underlying genetic mutations, which would ensure minimal toxicity to normal tissue. However, targeting public neoantigens could circumvent many of the limitations, including the complex logistics, inherent in personalized immunotherapy approaches. In addition, because driver gene mutations drive tumorigenesis, their public neoantigens should be present in every cell within the cancer—something that is not necessarily true for private neoantigens⁷—with cancer cells unable to lose the antigen without a concomitant loss of fitness⁸.

One potential disadvantage of off-the-shelf public neoantigen-targeting therapies is that no patient is likely to have more than one or a few public neoantigens available to be targeted. In contrast, cancers generally have multiple private neoantigens, in principle facilitating combinatorial targeting and mitigating the risk of treatment resistance due to antigen loss during clonal evolution⁹.

The frequency of driver gene mutations (Fig. 3) indicates that targeting their public neoantigens could potentially benefit a large proportion of patients. Apart from oncogenes, some tumor suppressor genes also exhibit recurrent mutations, but tumor suppressor genes in general are unlikely sources for public neo-antigens as they are often inactivated by non-recurrent mutations or expressed at low levels because of nonsense-mediated decay of their messenger RNA transcripts. An exception is TP53. Among the several TP53 mutation hotspots, the most common is the one leading to the protein alteration p.Arg175His, which was recently shown to be targetable with a bispecific antibody in pre-clinical studies¹⁰.

Fig. 3 |. Frequency of recurrent mutations across cancer types.

For the top 12 solid tumor organ sites projected to be responsible for most new cancer cases in the United States in 2020 (National Cancer Institute Surveillance, Epidemiology, and End Results Cancer Statistics Factsheets; https://seer.cancer.gov/statfacts/html/common.html), the corresponding The Cancer Genome Atlas (TCGA) projects were selected¹⁵⁹. Shown are the frequencies of occurrence in each TCGA project of each substitution that accounts for at least 5% of cases in at least one TCGA project. Sum represents the summation of frequencies of all included mutations for a given TCGA project.

A broadly applicable public neoantigen-targeting therapy also requires the presence of common HLA alleles, many of which do occur at high frequencies (Table 1). The product of the frequency of a specific hotspot mutation and the frequency of a specific HLA allele yields the theoretical frequency of patients who could benefit from public neoantigen-based therapies. For example, when the most common HLA alleles (Table 1) and the most common driver gene mutations (Fig. 3) are considered, the fraction of patients with pancreatic cancer who could potentially benefit from targeting KRAS p.Gly12Asp public neoantigens is ~10%. Given that most driver gene mutations, such as those in PIK3CA, TP53 and KRAS, occur in a large number of tumor types, those shown to generate validated neoantigens with reasonably frequent HLA alleles would be common enough to enable the development of off-the-shelf immunotherapies. For the 12 most common solid tumor organ sites in the United States, ~1% of patients could in theory benefit from an immunotherapeutic approach targeting a KRAS p.Gly12Asp peptide in the context of an HLA-A*03 allele (Table 2). This number of patients is comparable to or higher than those who might benefit from other targeted therapies used in clinical practice or currently in development. Examples include larotrectinib for solid tumors with NTRK fusions (which may be present in up to 1% of all solid tumors) and avapritinib for PDGFRA exon 18 mutant gastrointestinal stromal tumors (which occur in a few thousand patients per year)^11,12. If immunotherapeutic agents could be developed just to the top ten public neoantigens, at least one of these drugs could in theory be used to treat more than one million patients with cancer per year at the international level.

Table 1.

Frequencies of common HLA alleles

HLA	Phenotype frequency in population (%)
	Asian	Black	Hispanic	White	Weighted average
A*02:01	18	23	35	50	42
C*07:01	8	23	20	31	26
C*07:02	27	13	21	28	25
A*01:01	10	9	13	31	24
C*04:01	15	34	30	20	23
A*03:01	5	16	15	27	22
B*07:02	5	14	11	26	20
B*08:01	3	8	9	23	17
A*24:02	33	4	23	17	17
C*06:02	13	17	11	18	16

Phenotype frequencies of the ten most common HLA alleles in the United States. The alleles are ranked by average phenotype frequency^153,154, weighted according to ethnicity representation in the United States (United States Census Bureau QuickFacts; https://www.census.gov/quickfacts/fact/table/US/PST045219).

Table 2.

Top ten public neoantigens

Gene	Protein alteration	HLA	Top cancer	HLA frequencya (%)	Mutation frequencyb (%)		Neoantigen frequencyc (%)		Incidence of cancers carrying neoantigend		Reference for neoantigen validation
					All cancerse	Top cancerf	All cancerse	Top cancerf	All cancerse	Top cancerf
BRAF	p.Val600Glu	A*02	Melanoma	41.9	7.1	43.9	3.0	18.4	40,000	18,000	155
KRAS	p.Gly12Asp	A*03	Pancreas	21.7	3.8	32.4	0.8	7.0	11,000	4,000	24
KRAS	p.Gly12Val	A*03:01	Lung	21.7	3.3	6.7	0.7	1.5	10,000	3,000	40
TP53	p.Arg175His	A*02:01	Colorectum	41.9	1.6	6.5	0.7	2.7	9,000	4,000	10,42,46,156
KRAS	p.Gly12Asp	A*11:01	Pancreas	10.8	3.8	32.4	0.4	3.5	6,000	2,000	45
KRAS	p.Gly12Val	B*35	Lung	11.4	3.3	6.7	0.4	0.8	5,000	1,700	41
HRAS/KRAS/NRAS g	p.Gln61Arg	A*01:01	Melanoma	23.6	1.6	12.4	0.4	2.9	5,000	2,900	24,157
KRAS	p.Gly12Val	A*11:01	Lung	10.8	3.3	6.7	0.4	0.7	5,000	1,700	45
BRAF	p.Val600Glu	B*27:05	Melanoma	4.8	7.1	43.9	0.3	2.1	5,000	2,100	158
KRAS	p.Gly12Asp	C*08:02	Pancreas	7.5	3.8	32.4	0.3	2.4	4,000	1,400	79

This table lists public neoantigens identified with defined HLA restriction and with evidence for endogenous processing and presentation, ranked by the number of new patients annually in the United States for whom the public neoantigen applies. This analysis includes public neoantigens generated by substitution mutations recurrent in the top 12 most common solid tumor organ sites and identified with HLA class I restriction.

^aCalculated according to phenotype frequencies in the United States, weighted for proportional ethnicity representation^153,154 (United States Census Bureau QuickFacts; https://www.census.gov/quickfacts/fact/table/US/PST045219).

^bCalculated according to The Cancer Genome Atlas mutation frequencies, weighted according to the 2020 projected incidence of the top 12 most common solid tumor organ sites in the United States¹⁵⁹ (National Cancer Institute Surveillance, Epidemiology, and End Results Cancer Statistics Factsheets; https://seer.cancer.gov/statfacts/html/common.html).

^cCalculated by multiplying the HLA frequency by the mutation frequency.

^dNumber of new patients annually whose tumors carry the neoantigen, calculated by multiplying the cancer incidence by the neoantigen frequency.

^eAll of the top 12 most common solid tumor organ sites in the United States.

^fThe cancer type with the greatest number of applicable new patients annually.

^gHRAS, KRAS and NRAS p.Gln61Arg mutations are considered together because the protein sequences flanking the mutation site are identical.

The actual number of patients who could benefit from public neoantigen-targeted therapies could be lower than calculated maximum estimates because the existence of recurrent mutations and common HLA alleles does not guarantee the existence of public neoantigens. For a mutated gene to produce a neoantigen, the resulting altered peptide must be expressed, processed by cellular machinery, presented stably by an HLA on the cell surface and show a distinct feature recognizable to the immune system. Thus, the paramount question is whether specific driver gene mutations generate neoantigens that are presented by common HLAs. The high frequency of hotspot mutations has been suggested to be partly due to the inefficient neoantigen presentation of their product peptides, thereby permitting cells harboring the mutation to evade elimination by the immune system^13,14. Others have reported the lack of evidence supporting the depletion of mutations that generate neoantigens under negative selective pressure¹⁵. Evidence for a tumor’s ability to induce T cell tolerance to its own mutations further suggests that neither failure to present nor negative selection of peptides derived from driver mutations is universal¹⁶. Computational means have been employed to predict whether peptides containing hotspot mutations identified from sequencing data are likely to be presented. These efforts reveal that neoantigens generated from recurrent driver mutations are indeed predicted to exist (for example, for driver mutations in EGFR, KRAS, PIK3CA, TP53 and others)^17–19. However, computational methods to assess HLA binding of specific epitopes can generate both false negative and false positive predictions^1,20. Experimental validation is therefore critical for a more accurate estimate of the potential applicability of public neoantigen targeting to patients.

Experimental validation of public neoantigens

The most direct way to demonstrate the existence of public neoantigens is through mass spectrometry to detect HLA-complexed mutant peptides (pHLA), although the low sensitivity of detection may lead to false negatives²¹. Mass spectrometry experiments have been performed in primary tumor samples and in cell lines endogenously expressing the neoantigen or engineered to overexpress it^22–24, providing unequivocal evidence that neoantigens can be processed and presented by specific HLAs. However, such experiments do not provide evidence that these antigens are presented by tumors that harbor the mutation in every patient with that particular HLA allele. To demonstrate such presentation in an individual patient, mass spectrometry would have to be performed on pHLA complexes purified from the neoplastic cells of the patient, which is technically challenging but possible^21,25.

Another approach to identify or validate public neoantigens is to do so indirectly, through the existence of reactive T cells, as this suggests presentation of the putative neoantigen on the tumor cells^20,26,27. Neoantigens validated by this approach are immunogenic by definition. Although the majority of characterized public neoantigens recognized by T cells are HLA I restricted, HLA II-restricted T cell responses have also been identified, such as against the common BRAF mutation that leads to a p.Val600Glu alteration²⁸. Reactive T cells do not necessarily show that the neoantigen is actually processed and presented by tumor cells²⁹. Specifically, if rather than expressing the relevant gene the target cells are experimentally loaded with exogenous peptide, endogenous processing is not taken into account³⁰. More definitive evidence is therefore provided by detecting T cell reactivity against cells that express the driver gene and comparing that with isogenic cells engineered in a way, such as with knockout, that they do not express the gene. In vivo T cell clonal expansion also indicates presentation of a neoantigen that is sufficiently immunogenic to drive this expansion^31,32.

Overall, each experimental approach provides a different level of substantiation that a putative public neoantigen is presented by cancer cells and is immunogenic. We propose an eight-point hierarchy for levels of evidence, ordered from the strongest to the weakest:

1. Generation of a clinical response in a randomized, controlled trial of a public neoantigen-targeted therapy. No public neoantigen has yet achieved this level of validation.

2. Documentation of a clinical response to a public neoantigen-targeted therapy in an uncontrolled trial^33–35.

3. Experimental confirmation of neoantigen presentation by mass spectrometry of primary cancer cells from patients with the expected neoantigen^23,36,37.

4. Identification of T cell reactivity against primary cancer cells in patients with the expected neoantigen^28,38,39. This is less rigorous than mass spectrometry, because the absence of false positives due to cross-reactivity cannot be guaranteed, but it does provide evidence of immunogenicity rather than presentation alone.

5. Experimental confirmation of presentation in model cell lines endogenously expressing the public neoantigen of interest^{10,22–24,40–42}.

6. Experimental confirmation of presentation in model cell lines engineered to overexpress the public neoantigen^{4,5,24,31,42–46}.

7. Confirmation of presentation in peptide-loaded cell lines.

8. Computational prediction.

As our understanding of the roles of cross-presented public neoantigens and ways to exploit them increases, it will also be useful to integrate evidence for cross-presentation into the validation framework. Supplementary Table 1 lists the public neoantigens identified to date that meet at least level 6 evidence for validation and for which the HLA restriction has been defined. The top ten of these neoantigens have also been ranked according to the number of new patients with cancer per year in the United States for whom a targeted therapy would be applicable (Table 2). The ranking is based on the overall incidence, including early-stage cancers. Although immunotherapies targeting public neoantigens are currently most relevant for advanced cancers, the prospective safety profile of these precision therapies, owing to their specificity, may enable them to eventually be applicable even in early disease settings.

Future identification of public neoantigens

As noted above, validating a candidate neoantigen as a bona fide one is still challenging, and showing proper processing and presentation is a major bottleneck. For immunotherapeutic strategies that make use of T cell responses, such as vaccination or adoptive cell transfer, a second bottleneck lies in determining whether presented antigens are immunogenic. Indeed, the majority of neoantigens confirmed by mass spectrometry as presented by HLAs do not induce T cell responses^21,25,47. The failure of T cells to recognize many neoantigens may be due to the similarity of the latter to their wild-type counterparts and the corresponding gaps in the functional TCR repertoire induced by central or peripheral tolerance mechanisms⁴⁸. For immunotherapies such as engineered antibody-based constructs, immunogenicity is not a limiting factor, providing a major advantage to such modalities compared with those that are dependent on a natural immune response, including immune checkpoint inhibitors.

Several strategies could help to overcome these bottlenecks. First, more accurate prediction of the binding of candidate peptides to specific HLA alleles is essential. Traditional computational methods for predicting the presentation of neoantigens were developed using affinity data of peptide binding to HLA. However, more recent prediction tools using machine learning methods trained on expanded mass spectrometry data of endogenously presented peptides account for additional factors in the presentation process beyond affinity and have achieved increased accuracy in predicting neoantigen presentation for both HLA class I and class II^49–52. The growing volume of publicly available mass spectrometry data should facilitate continued improvements in neoantigen prediction methods^53–55. Similarly, using binding affinity alone to forecast immunogenicity may lead to poor predictions. Indeed, peptides predicted to have low affinity for HLA can still be immunogenic, as is the case for the TP53 p.Arg175His alteration⁴⁶. In contrast, many neoantigens predicted to have reasonable binding affinity have not been identified in empirical searches of presentation or the ability to induce T cell responses^3,5,25. Evidence suggests that features other than affinity, such as pHLA stability, greater difference between altered versus wild-type peptide binding to HLA, or peptide foreignness, could better predict immunogenicity and prioritize targets^56–58. An advantage of some of the public neoantigen-targeted approaches described here, such as those using antibodies or cells directed to specific pHLA complexes, is that they do not require immunogenicity of the pHLA complex, but rather only its presence on the surface of the cancer cell.

Second, neoantigen discovery could be improved by screening with new innovations in recombinant HLA reagents. One traditional strategy is to measure T cell binding to soluble recombinant HLA molecules complexed with neoantigen peptides, as the existence of reactive T cells implies neoantigen presentation. Reagents with combinatorial fluorescent, isotope or DNA barcodes or nanoparticle functionalization have vastly expanded throughput and sensitivity. These methods also support simultaneous interrogation of multiple candidate neoantigens with a single precious clinical sample^59–61. Fully formed HLA molecules in which peptides can be inserted at will have also been reported^62–64. These new technologies should enable rapid testing of candidate neoantigens to all common HLA alleles. Positive results would not prove that a candidate neoantigen is properly processed and presented, but negative results would exclude neoantigen–HLA combinations from further testing in cell-based assays.

Third, advances in assays measuring functional T cell responses to neoantigen-presenting cells will continue to be important. For example, tandem minigene constructs that allow for unbiased and simultaneous screening of both HLA class I- and class II-restricted neoantigens from multiple mutation sites have aided the identification of candidate public neoantigens in numerous studies^4,5,33,46.

Fourth, combinations of recurrent mutations (Fig. 3) and common HLA alleles (Table 1) could be screened (for example, by expressing the top 25 genes harboring hotspot mutations in cells expressing common HLA alleles, followed by mass spectrometry of purified pHLA complexes). Testing various cancer cell lines representing the most common cancer types and HLA types would help to mitigate differences in presentation capabilities. Though costly, this project is feasible and would have immediate clinical ramifications. Such experiments on cells engineered to express these mutations could be a starting point for interrogating variability between cancer types and heterogeneity among the tumors of individual patients.

Fifth, it is important to determine whether candidate neoantigens demonstrated to be processed and presented on the cell surface in an experimental system are presented on a particular individual’s cancer cells. Genetic or epigenetic differences in protein processing or presentation create interpatient and intratumor heterogeneity. For example, if cancer cells do not express HLA complex components, they cannot present public neoantigens. Inactivation of HLA genes or other genes involved in HLA class I and II antigen processing and presentation is known to occur in some patients, making tumors resistant to any immunotherapy relying on particular pHLA complexes, including immune checkpoint inhibitors^9,33,65. Assessment of functional HLA molecules on cancer cells is likely to become the standard of care for assessing the value of selected immunotherapeutic approaches in the future. Because neoantigen cross-presentation can be clinically relevant, neoantigen presentation by cell types other than the cancer cells themselves should be measured in tumors^66,67—an issue that is more pertinent to the development of vaccines against public neoantigens than for antibodies or cells that react with specific pHLA complexes.

At present, the approach used to judge an individual patient’s likelihood of responding to an immunotherapy targeting a public neoantigen is to evaluate the tumor DNA sequence and cell surface HLA expression. However, this does not definitively determine presentation of the public neoantigen by the patient’s tumor cells. In the future, it is hoped that mass spectrometry methods will have advanced sufficiently to be applied directly to patient samples. The identification of neoantigens in pHLA complexes by mass spectrometry currently requires hundreds of millions of cells²⁴, but sensitivity for both HLA class I and class II alleles continues to be improved^{24,25,67–69}. The application of such techniques to hundreds of thousands rather than millions of neoplastic cells would enable demonstration that an individual patient’s cancer cells do present the candidate public neoantigens. Documentation that T cells from the patient with cancer can react with candidate public neoantigens could provide additional evidence of presentation and would further suggest that stimulation of this immune response, particularly in combination with immune checkpoint inhibitors, could be therapeutic.

Although these emerging and aspirational advances are not essential for moving public neoantigen-targeting therapy candidates forward in the short term, they will be essential for expanding the number of patients public neoantigen-targeting therapies can routinely reach in the long term. Advancing technologies will also help to increase the number of validated public neoantigens, making them relevant to more patients. To take advantage of off-the-shelf therapies, companion diagnostics to identify suitable patients must similarly be available in an off-the-shelf fashion and at scale. Some of these approaches, such as adding HLA typing to next-generation sequencing panels or implementing HLA expression assays, should be straightforward to integrate into clinical workflows. Other assays, such as individualized mass spectrometry or T cell reactivity assays, would be more difficult to implement and standardize in a clinical setting. Ultimately, applying them to every patient may not be essential if a large proportion of patients with the mutation and HLA type can be shown to present the relevant public neoantigen and respond to treatment.

Therapeutic targeting of public neoantigens

The validation of public neoantigens opens opportunities to exploit them therapeutically. In the future, sequencing cancer driver genes from patients’ tumor DNA is likely to be routine, to determine the applicability of targeted therapy. The same DNA could be readily assessed for HLA type and the patient matched to an off-the-shelf therapy corresponding to that public neoantigen and HLA type combination.

Four general treatment modalities targeting public neoantigens have been explored to date (Fig. 2). The first is vaccination. The treatment of patients with cancer with vaccines engendering an immune response against public neoantigens derived from KRAS mutations has a near three-decade history^34,41,70. Vaccination strategies against TP53 and BCR/ABL fusion public neoantigens have also been conducted^70,71. Although such vaccines have succeeded in generating neoantigen-responsive T cells⁷⁰, measures of clinical benefits have been modest and anecdotal, with no vaccinations yet demonstrating statistically significant patient benefit in randomized trials⁷². This suggests that the vaccines did not stimulate a sufficiently strong immune response to eliminate cancer cells in an immunosuppressive tumor environment. Better adjuvants, improved delivery formats and combinations with immune checkpoint inhibitors could improve efficacy. Much of the current intense effort to improve the efficacy of immune checkpoint inhibitors will also be applicable to vaccines, and in particular to vaccine–immune checkpoint combinations. Trials might also benefit from focusing on patients with tumors that have been validated to present the neoantigen target. One illustrative study designed a vaccine targeting public neoantigens derived from 20 recurrent mutations that were predicted to cover a median of 11% of nine types of common solid tumors³⁴. Similar vaccines targeting multiple public neoantigens could simplify regulatory certification and production. For example, the need to prepare individual vaccines for each patient (as will always be the case for private neoantigens) could be alleviated by producing vaccines specific for each HLA type, with each one including the most common recurrent mutant peptides presented by that HLA.

The second immunotherapeutic modality that can be employed to target public neoantigens is the adoptive transfer of specifically reactive tumor-infiltrating lymphocytes (TILs) or cytotoxic T cells derived from autologous or allogeneic sources. One study of a patient with metastatic colorectal cancer reported tumor regression after adoptive transfer of autologous TILs exhibiting reactivity to a KRAS p.Gly12Asp neoantigen³³. In a second study, adoptive transfer of autologous TILs exhibiting reactivity for a BRAF p.Val600Glu neoantigen was associated with melanoma regression³⁵. Autologous T cells specific for a BCR/ABL translocation public neoantigen, transferred in three patients with acute lymphocytic leukemia, led to remission when combined with tyrosine kinase inhibitors⁷³. This work suggests that adoptive cell transfer of T cells specific for public neoantigens can produce durable clinical responses. This suggestion was supported by another study in which adoptive transfer of T cells with reactivity to a TP53 p.Arg175His neoantigen was evaluated⁴⁶. However, only some lesions responded in this study and an immune checkpoint inhibitor was used, complicating interpretation and raising the possibility of alternative explanations for the observed efficacy.

Adoptive transfer therapies using patient-specific, public neoantigen-reactive TILs are limited in their scalability, but they have heralded the third modality of public neoantigen targeting: validated TCRs cloned from the naturally occurring T cells. TCRs from TILs and peripheral T cells that recognize a variety of public neoantigens have been described^{37,38,42,43,74}. Thus, it is theoretically possible to genetically engineer the relevant TCR into the T cells of other patients to furnish reactivity against a specific neoantigen. Transposon or CRISPR–Cas9 systems can streamline the rapid and safe engineering of TCRs into T cells^75,76. Notably, trials in which patients receive autologous T cells engineered to express TCRs targeting KRAS p.Gly12Val or p.Gly12Asp altered peptides presented by HLA-A*11:01 are already underway⁷². Because these trials employ a murine TCR, it will be important to analyze whether host reactivity against the TCR may decrease the treatment efficacy or contribute to adverse effects⁴⁵. Alternatively, soluble constructs directed by monoclonal TCR moieties could be used as an off-the-shelf therapy for matched public neoantigens. One example of this format is ImmTACs (immune mobilizing monoclonal T cell receptors against cancer), a class of bispecific molecules comprising a TCR component connected to an anti-CD3 antibody component, which together crosslink target cells to T cells, thereby activating them^77,78. Improving pipelines to identify relevant TCR sequences from patients with cancer, healthy donors or engineered mice could accelerate the implementation of these TCR-based strategies^44,79–82.

In contrast with the three modalities described above that share reliance on neoantigen-reactive TCRs, a fourth approach employs antibody moieties to drive specific activity (for example, through full-length antibodies, antibody–drug conjugates, bispecific antibodies or chimeric antigen receptor T cell (CAR-T) therapies)^83,84. Two representative studies used phage display to select antibody fragments against KRAS, EGFR and CTNNB1 public neoantigens, but endogenous processing and presentation of these neoantigens remains to be validated^85,86. A third study used yeast display to select antibody fragments against an NPM1 public neoantigen and demonstrated conversion into a CAR format³⁹. In two other recent studies, bispecific antibodies directed to TP53 p.Arg175His, KRAS p.Gly12Val or RAS p.Gln61His/Leu/Arg public neoantigens were shown to be somewhat efficacious in both in vitro and in vivo models^10,40. Such antibodies against pHLA molecules are sometimes referred to as TCR mimics and tend to have much higher affinity than natural TCRs^87–90. Thus, they must be carefully screened to avoid cross-reactivity or binding of the HLA component independent of the presented peptide. As with engineered TCRs, negative selection against off-target peptides during directed evolution is an important method for avoiding cross-reactivity^39,85,86. Synthetic reagents with less cross-reactivity than analogous natural receptors have been created in at least one instance⁹¹. Combinatorial library screening approaches can provide determinations of specificity on a large scale^59,92,93. Structural analysis of a reagent in combination with its cognate pHLA can also help to predict cross-reactivity and guide rational designs to improve specificity^10,94–98.

Comparisons and optimizations of therapeutic formats

Vaccine therapies provide practical advantages over cell-based therapies because they are easier to manufacture and more scalable. Neoantigen vaccine testing in patients has also proven to be generally safe^99,100. In contrast, data on objective clinical responses to cancer vaccines are sparse, as illustrated by the lack of responses to vaccines against KRAS hotspot neoantigens^101,102. The limited efficacy so far may be a result of inhibitory factors within the tumor microenvironment that ultimately provide barriers that are too high for vaccine-induced T cells to overcome. Innovating vaccine design, composition and delivery to maximize immune stimulation might bolster efficacy^103–106. Rather than being used against advanced disease, vaccines may be more effective in early disease settings or adjuvant settings to prevent relapse^103,107,108. The efficacy of an experimental vaccine against human papilloma virus antigens in premalignant lesions serves as a model, although clinical benefit awaits the results of an ongoing randomized phase 3 clinical trial¹⁰⁹. As a more radical but potentially effective strategy, public neoantigen-targeting vaccines could be used prophylactically^110–112. In theory, healthy individuals could ultimately be vaccinated against all validated public neoantigens matched to their HLA alleles.

Cell therapies with CARs or TCRs have been more potent than vaccines, offering hope that these platforms can be adapted for targeting public neoantigens¹¹³. Although they are more scalable than approaches that seek to grow natural neoantigen-specific T cells from individual patients, important challenges still persist in the manufacturing, scalability and safety of these therapies¹¹⁴. The personalized manufacturing component for autologous cell products limits their potential for widespread application of public neoantigen-targeted therapies. In contrast, universal T cells, in which allogeneic cells are used for therapy, would in theory overcome this hurdle¹¹⁵. The major challenges for the treatment of patients with an allogeneic product are graft-versus-host reactivity and rejection by the host’s immune system. Approaches to overcome these limitations include genetically ablating TCR genes (the source of graft reactivity against the host), genetically ablating HLA genes (the source of host recognition of transferred allogeneic cells), adoptive transfer into lympho-ablated patients, engineering cells with synthetic modules to make them resistant to rejection, or using alternative cell types such as natural killer cells^116,117. Methods to program immune cells in patients in situ without an ex vivo manufacturing process could streamline delivery for any type of cellular therapy^118,119.

Soluble antibody or TCR construct-based therapies, such as bispecific antibodies or ImmTACs, represent a bridge between the power of specifically redirecting T cells and the logistical advantages of non-cell-based therapy. Combining soluble or cell-based therapeutics with vaccines or with immune checkpoint inhibition therapies may enable targeting of a patient’s full public and private neoantigen repertoire to induce maximum responses while preserving T cell functionality. T cells can be genetically engineered to overcome immunosuppression and to function effectively in the hostile microenvironment of solid tumors, offering tremendous opportunities for future innovation^120–123. As the efficacies of various engineered T cell and T cell-engaging therapies targeting public neoantigens evolve, each therapy’s unique and relative toxicities will also require attention to fine tune anti-tumor immunity while reducing treatment-associated morbidity and mortality.

Emerging insights from biology

Emerging knowledge of the biological differences between redirecting T cells with CARs, TCRs or bispecific antibodies will further illuminate the optimal properties for platforms targeting public neoantigens. The immunological synapses formed with TCRs, CARs and bispecific antibodies are distinct in their structure, affinity and binding kinetics^124–126, leading to differing downstream signaling between CARs and TCRs^127–129. A consequence of these differences is the distinct sensitivity to antigen density levels conferred by various therapeutic formats^130–132—a clinically important point given that low antigen expression is a mechanism of resistance to and relapse following CAR-T therapies¹³³. Peptide–HLA complexes, including those of public neoantigens, are typically present on the cell surface at copy numbers of less than 100 per cell^{10,24,40,134,135}. TCRs can trigger a response to one or a few antigens¹³⁶. CARs, with some exceptions, seem to require an antigen density threshold on the order of hundreds to thousands per cell and are less sensitive than TCRs when compared using synthetic constructs^{130,131,137–139}. Bispecific antibodies generally require hundreds to thousands of antigens per cell¹⁴⁰, but recent studies have also shown reactivities of special formats of bispecific antibodies against antigens at densities of single digits per cell^10,40. Similarly, ImmTACs are active against cells with one to tens of copies per cell⁷⁷. Within each format, optimization of architecture can potentially decrease neoantigen density requirements¹³⁹. Beyond antigen density sensitivity, in-depth analyses relating cell and treatment characteristics with patient outcomes also hold potential for informing additional engineering efforts for different therapeutic platforms^141–143.

A final issue is how best to approach cancers that carry HLA class II-restricted public neoantigens. Immunologic responses have been achieved using adoptive transfer of CD4⁺ T cells that recognize class II-restricted neoantigens^144,145. Additionally, neoantigen vaccines that have demonstrated enhanced anti-tumor immunity have been associated predominantly with CD4⁺ T cell responses^146–148. The precise mechanisms of these CD4⁺ T cells in mediating anti-cancer activity remain to be determined^100,149,150, although augmenting activation of cytotoxic T cells and inducing tumoricidal inflammatory macrophages are two established functions of tumor antigen-specific CD4⁺ cells. Directly cytotoxic HLA class II-dependent CD4⁺ T cells have also been observed¹⁵¹. Further study of CD4⁺ T cell roles and of which cell types within the tumor microenvironment present class II-restricted public neoantigens is important for future progress.

Conclusions and outlook

Public neoantigens are unquestionably a compelling therapeutic target. Initial work has validated their existence and is translating that knowledge into therapies. Indeed, the first clinical trials of therapeutic agents targeting public neoantigens are underway. Progress in identifying more public neoantigens will be critical for expanding the applicability of these approaches to more patients. Moreover, the widespread adoption of cancer genome sequencing will facilitate the pairing of patients with therapies that target the public neoantigens in their tumors. In the future, an armamentarium of public neoantigen-targeted therapeutics could be developed to power precision immunotherapies against the most common genetic changes that drive cancer^3,10,40,152.

Data availability

Data for The Cancer Genome Atlas mutation frequencies used in the analyses presented in Fig. 3 and Table 2 are available from the National Cancer Institute Genomics Data Commons (https://gdc.cancer.gov/). Data for the HLA frequencies used in the analyses presented in Tables 1 and 2 and Supplementary Table 1 are available from the Allele Frequency Net Database (http://www.allelefrequencies.net/) and National Marrow Donor Program (https://bioinformatics.bethematchclinical.org/hla-resources/haplotype-frequencies/high-resolution-hla-alleles-and-haplotypes-in-the-us-population/). Data for cancer incidence used in the analyses presented in Fig. 3 and Table 2 are available from the National Cancer Institute Surveillance, Epidemiology, and End Results Program (https://seer.cancer.gov/statfacts/html/common.html). Data for ethnicity representation in the United States used in the analyses presented in Tables 1 and 2 and Supplementary Table 1 are available from the United States Census Bureau (https://www.census.gov/quickfacts/fact/table/US/PST045219).