The concept of oncolytic virotherapy originates from clinical reports of cancer regression that coincides with natural viral infections1. Virotherapy is currently being developed by genetically modifying viruses for the selective infection and destruction of cancer cells2,3. Many viruses have specific tissue tropisms that can be exploited as a starting point for preferential infection and replication within the tumour microenvironment, killing cancer cells while replicating and spreading within disease foci. Clinical trials of oncolysis have been performed for decades4, but virus engineering strategies have only recently been developed to closely monitor virus replication and to address clinically relevant challenges, such as efficient systemic delivery, tight tumour specificity and improved efficacy in combination with current cancer therapies. By exploiting our ever greater understanding of tumour biology (Box 1), these advances support the clinical translation of many new and diverse viruses that have been rationally designed to have greater safety and efficacy in the clinic3,5.

Excellent reviews have thoroughly covered the results of current clinical trials of oncolytic virotherapy3,6. In this Review, as an introduction to the field, we summarize the most advanced clinical trials for viruses from nine different families that are currently being tested as anticancer therapies3. Table 1 lists these selected examples and the modifications of the engineered viruses, the routes of administration and the use of combination therapies for each trial. We focus on three points: first, the increasing diversity of viral families that are being developed for oncolysis; second, the notable safety of currently used viruses, which has, in many cases, been shown at the highest doses achievable by today's manufacturing processes; and third, the successes that have been achieved using oncolytic viruses that express immunostimulatory transgenes. As a transition to next-generation preclinical viruses, we also highlight the continued clinical need for improved delivery to and replication within systemic tumours, as well as therapeutic synergy with both the immune system and currently available cancer therapeutics.

Table 1 Selected examples of current clinical trials with viruses from nine families

Virus families have evolved specificities for different cell types, and this natural diversity is being used for therapeutic development. Currently, there are viruses from nine different families in clinical trials: Adenoviridae7, Picornaviridae8, Herpesviridae9,10, Paramyxoviridae11,12, Parvoviridae13, Reoviridae14,15, Poxviridae16,17,18, Retroviridae19 and Rhabdoviridae20,21 (Table 1). Viruses from all of these families (except Reoviridae and Parvoviridae) have been engineered to have greater tumour specificity and/or efficacy than their parental strains. A reverse genetics system has become available for reovirus22, which will enable further development of this virus family in the future. Broadened virus availability is exemplified by the newly discovered picornavirus Seneca Valley virus, which was found to preferentially infect neuroendocrine tumours, such as small cell lung cancer23. Indeed, robust replication was specifically observed in patients who had small cell lung cancer8, and this has led to the initiation of a Phase II clinical trial.

Generally, oncolytic virotherapy has been a well-tolerated experimental clinical therapy after both localized and systemic administration5. The most common adverse effects are fever and general flu-like symptoms, but more serious toxicities have been documented in rare cases5,6. In addition, no transmission of an oncolytic virus from treated patients to carers or other contacts has been noted, although shedding of virus has been documented in the urinary and respiratory tracts, especially after systemic administration. Continuing technological advances in virus production should enable more aggressive dosing in future trials3, placing greater onus on the tumour specificity of future viruses to maintain current safety profiles.

An important result from clinical trials is the success of therapeutic protocols that are based on the expression of the immunostimulatory cytokine granulocyte–macrophage colony-stimulating factor (GM-CSF). Replicating viruses from three different families (Table 1), talimogene laherparepvec (Amgen; Herpesviridae), JX-594 (Jennerex Biotherapeutics; Poxviridae) and CG0070 (Cell Genesys; Adenoviridae), combine replicative oncolysis with GM-CSF-mediated stimulation of granulocytes and monocytes to induce inflammation and adaptive immunity against tumour antigens. All three viruses have shown efficacy in advanced clinical trials: the results of a Phase II trial for talimogene laherparepvec in melanoma have been reported9 and a Phase III trial in melanoma has recently been completed; JX-594 has been tested in a randomized Phase II trial in hepatocellular carcinoma18; and CG0070 is currently being tested in a Phase II trial for bladder cancer24.

Although Phase I and II clinical trials are not designed to directly measure efficacy, most current-generation viruses in clinical trials have fallen short of the efficacy expectations that were set by preclinical models2,3. Improving efficacy is a multifactorial challenge, and in this Review, we summarize the preclinical engineering strategies for the virus families that have been most extensively studied so far. We describe engineering strategies that focus on overcoming continued clinical challenges: resisting antibody neutralization using genetic and chemical virus shielding; improving tumour specificity by targeting tumour-associated receptors and controlling post-entry viral replication; and improving therapeutic synergy with the immune system, chemotherapy and radiotherapy. By carefully pairing diverse virus families with engineering strategies and combination therapies, next-generation viruses can be rationally designed for greater efficacy against diseases that have unmet clinical needs.

Avoiding virus neutralization

Systemic administration of oncolytic viruses via the vasculature gives the virus access to all perfused regions of primary and metastatic tumours, making it the preferred administration route for the treatment of metastatic disease. However, systemic delivery also makes the virus susceptible to inactivation by pre-existing antibodies in the blood, which can arise in patient populations via natural contagion, scheduled immunization or prior administration of a therapeutic virus. Pre-existing and induced antiviral immune responses can be pharmacologically tempered to limit the neutralization of therapeutic viruses25,26,27,28,29. In this section, we discuss virus engineering strategies that can shield therapeutic viruses from pre-existing neutralizing antibodies by changing or physically masking the epitopes that are recognized by the antibodies. Multiple engineering strategies that differ in the type and magnitude of modification have been applied to different virus families, but the end result is always a chimeric virus with an engineered serotype that is not recognized by the pre-existing antibodies present in the target patient population (Fig. 1). We use vesicular stomatitis virus (VSV), adenovirus and measles virus to illustrate these different strategies, highlighting both the type and the extent of each modification, as well as discussing the applicability of these strategies to other virus families and combination regimens with pharmacological immunosuppression.

Figure 1: Vector-shielding strategies.
figure 1

a | Interchange of different serotypes from the same virus species is shown for vesicular stomatitis virus (VSV), which has only one envelope glycoprotein. b | Replacement of multiple immunogenic epitopes in different proteins is shown for adenovirus; exchange can be achieved using full domains (such as the fibre knob) or individual motifs (such as hexon hypervariable loops). c | Generation of a new serotype is shown for measles virus. The two glycoproteins of this monotypic virus are substituted by the glycoproteins of an animal virus of the same genus. d | Chemical shielding of viral epitopes is shown for adenovirus; small polymers can be added to purified viral particles.

PowerPoint slide

Serotype exchange. The most basic method of genetic shielding is serotype exchange, whereby a different serotype of the same virus species is engineered onto the core of an established virus, generating a chimeric virus that has the donor serotype but the original core (Fig. 1a). This strategy requires the availability, for a given virus, of multiple serotypes, which typically have less than 60% identity between their surface-exposed glycoproteins or capsid proteins30. For example, all neutralizing epitopes that are present on VSV particles are found in the VSV-G glycoprotein, therefore serotype switching can be accomplished by simply replacing the entire VSV-G glycoprotein gene sequence with that of another serotype (Fig. 1a). Serotype switching for adenovirus is more complicated because multiple surface-exposed capsid proteins, such as the hexon or fibre knob proteins, contain neutralizing epitopes, as recently reviewed in Ref. 31. The hexon capsid protein encodes seven distinct hypervariable regions that all contribute to determining the serotype of the virus32; by modifying all seven hypervariable regions, the human adenovirus 5 (HAdV-5) serotype, against which most of the population has neutralizing immunity, can be substituted for the much rarer HAdV-48 serotype33 (Fig. 1b). Replacing the HAdV-5 fibre knob with that from HAdV-3 also generated a less immunogenic chimeric virus that was resistant to neutralization by serum from HAdV-5-immunized mice during gene transfer in vivo and ex vivo34 (Fig. 1b). However, in addition to changing serotypes, knob modifications can change receptor specificity: replacing the fibre of HAdV-5 with the fibre from HAdV-16 or HAdV-50 changes receptor specificity from coxsackie–adenovirus receptor (CAR) to the ubiquitous CD46 molecule35. Notwithstanding tropism modifications, which are discussed below, these genetic exchanges can create viruses that have unique serotypes built around a common viral backbone.

Novel serotypes. A variation of the serotype exchange strategy has been implemented for monotypic viruses, such as measles virus, which have only one serotype. The measles virus glycoproteins were replaced with those of canine distemper virus (CDV), which is also a Morbillivirus, to generate an infectious virus that is not neutralized by anti-measles virus antibodies36 (Fig. 1c). Importantly, CDV and measles virus, and possibly all members of the Morbillivirus genus, enter cells using the same primary receptors — signalling lymphocytic activation molecule (SLAM) and nectin 4 — although there are differences in cross-species receptor recognition37,38. The applicability of glycoprotein exchange is limited by the availability of suitable envelope donors that are compatible with measles virus; the glycoproteins from a closely related virus of another genus did not sustain efficient assembly of chimeric particles39. Much like serotype exchange, glycoprotein or capsid exchange between viruses requires a balance between protein diversity that is sufficient for avoiding cross-neutralization and sequence and structural similarities that are necessary to support particle formation.

Chemical modifications. Shielding strategies can be used to generate a repertoire of viruses that are built around a common core but that have unique serotypes. Such a repertoire can then be used for sequential rounds of therapy that maintain efficacy even as patients develop immunity to previously used viruses. This sequential administration strategy using engineered serotypes can enhance efficacy by expanding the therapeutic window for virotherapy and, at the same time, maintaining common targeting and arming strategies.

Chemical modifications of virus particles help to overcome some of the challenges that are associated with genetic shielding. Polymer deposition on particles can physically shield epitopes from antibody neutralization (Fig. 1d); polyethylene glycol (PEG) and poly-N-(2-hydroxypropyl) methacrylamide (poly-HPMA) polymers are two examples of polymers that are used to shield oncolytic viruses40,41. Chemical shielding has been most extensively applied to Adenoviridae to protect particles from inactivation in the blood and decrease off-target liver transduction42,43,44,45, but other virus families, such as Rhabdoviridae46 and Poxviridae47, have also been chemically modified. The utility of chemical shielding depends on viruses retaining their ability to enter cells after polymer deposition, which has been successful with the icosahedral adenovirus and enveloped viruses, such as VSV and vaccinia virus. Alternatively, polymer shields that are linked to specific ligands can be used to restore and target virus entry41,48, as discussed below. Importantly, chemical shielding strategies simplify virus production, but in vivo virus replication will leave progeny particles unprotected, potentially limiting efficacy in applications that depend on virus spread.

Tumour targeting

In this Review, we use the term targeting to describe virus modifications that confer greater specificity for tumour cells by improving infection of diseased tissues and decreasing infection of healthy tissues. This specificity can be enhanced either at the stage of virus entry into target cells or post-entry during replication. Entry targeting can be achieved by fusing or conjugating specificity domains that modify receptor usage of virus particles. Post-entry, tumour-specific replication targeting, using promoters and engineered microRNA (miRNA) target sequences, can restrict virus replication in off-target tissues. These targeting strategies differ fundamentally from those that were used in the first-generation oncolytic viruses, which were often based on the removal of virulence factors that are redundant for replication in tumours49. Although this enhances safety in normal tissues, it often limits replicative fitness in target tissues50. Instead, these novel retargeting strategies can be applied to viruses that have more wild-type characteristics to minimize off-target toxicity without compromising virus replication within disease foci.

All viruses require interactions with surface molecules on host cells to start infection. Viruses can be specifically retargeted to recognize molecules that are preferentially or exclusively expressed on tumour cells (Box 1). This strategy requires modifications to be made to the receptor-binding proteins that are present on viral particles. These modifications are either genetic and generate chimeric proteins, or they use chemical adaptors to link specificity domains to virus particles51. Targeting of enveloped viruses from the Paramyxoviridae and Herpesviridae families has rapidly progressed owing to the plasticity of their glycoproteins and the separation of receptor-binding and membrane-fusion functions, which are mediated by different proteins52. By contrast, non-enveloped icosahedral viruses, such as the Adenoviridae, have very stringent structural constraints on particle assembly, which limit viable modifications to short peptides or the chemical retargeting of assembled particles44,53 (Fig. 2).

Figure 2: Principles of tumour targeting — illustrated for four virus families.
figure 2

From top to bottom: targeting cell entry (detargeting from natural receptors and retargeting to tumour surface markers) and post-entry targeting (targeting of transcription, replication or microRNAs (miRNAs)). CAR, coxsackie–adenovirus receptor; HVEM, herpesvirus entry mediator; SLAM, signalling lymphocytic activation molecule.

PowerPoint slide

Receptor targeting: Paramyxoviridae. Among enveloped viruses, Paramyxoviridae remain the preferred platform for the development of new targeting strategies because receptor binding and membrane fusion are mediated by two different proteins, the attachment protein (haemagglutinin, haemagglutinin-neuraminidase or glycoprotein) and the fusion protein, respectively. Adding specificity domains, such as single-chain antibodies, to the carboxyl terminus of the attachment protein sustains binding to designated receptors and subsequent membrane fusion is achieved by the unmodified fusion protein (Fig. 1). Recombinant measles virus particles that express a retargeted haemagglutinin in place of the standard haemagglutinin can be reliably generated and stably passaged54. In addition, entry via the natural receptors can be ablated by mutating specific haemagglutinin protein residues that are necessary for binding and/or entry55. This targeting principle has recently been extended to designed ankyrin repeat proteins (DARPins), which are engineered repeat-motif proteins that are smaller than single-chain antibodies and can be combined to achieve multiple specificities56; for example, it was shown that DARPins can simultaneously retarget measles virus to two different tumour markers57. This dual retargeting strategy could theoretically be used to target both tumour parenchyma (for debulking) and CD133+ cancer-initiating cells (for prolonged growth inhibition58) using a single virus, as well as to safeguard against tumour resistance due to heterogeneous or downregulated receptor expression.

Receptor targeting: Herpesviridae. The herpes simplex virus (HSV) entry mechanism is more complex than that of the Paramyxoviridae. HSV relies on five proteins, glycoprotein C, glycoprotein B, glycoprotein D and the glycoprotein H–L dimer, for receptor binding and membrane fusion, which can occur at the plasma membrane or in endocytic vesicles59,60. Glycoprotein D is responsible for binding three different receptors: the herpesvirus entry mediator (HVEM); the cell adhesion molecule nectin 1; and 3-o-sulphotransferase-modified heparan sulphate59,60 (Fig. 2). Despite this complexity, HSV retargeting shares common themes with measles virus: glycoprotein D can be engineered to express ligands, such as interleukin-13 (IL-13)61 or urokinase plasminogen activator62,63, or single-chain antibodies against human epithelial growth factor receptor 2 (HER2; also known as ERBB2)64 near its amino terminus, which retargets HSV to antigens expressed on gliomas and breast tumours, respectively. Furthermore, the natural receptor tropism can be ablated by sterically blocking the receptor-binding interfaces of glycoprotein D with a single-chain antibody64 or by using a single chain antibody to replace the entire imunoglobulin core of the glycoprotein65, which results in simultaneous retargeting of the virus to HER2 and detargeting from its natural receptors. In addition, glycoprotein C functions with glycoprotein B during initial cell binding and can be used to retarget entry using small deletions and appended single-chain antibodies66. Modifications to glycoprotein B and glycoprotein D have also been combined: entry-accelerating mutants of glycoprotein B have been paired with single-chain variable fragment (sc-Fv)-engineered envelope glycoprotein D to improve the efficiency of HSV retargeting to epidermal growth factor receptor, which is expressed on glioblastomas67. Although the entry mechanism of HSV is more complex than that of measles virus, both viruses can be genetically engineered for targeted entry using similar strategies.

Receptor retargeting: Adenoviridae. Adenovirus entry targeting is more demanding than that of Paramyxoviridae or Herpesviridae owing to the constraints of the icosahedral particle structure. Nevertheless, short heterologous peptides and specificity domains that recognize tumour-associated antigens have been inserted into the HI loop and the C terminus of the receptor-binding trimeric fibre protein, which is located at the vertices of the capsid53,68,69,70,71. The HI loop, in particular, is flexible and the inserted short peptides have minimal negative effects on virus fitness, even in combination with mutations that detarget viruses from natural adenovirus receptors68. In addition, an adenoviral minor capsid protein, the hexon-interlacing protein IX (also known as 'cement' protein IX), can be used to successfully retarget entry using specificity domains, such as single-domain antibodies that do not require oxidation for folding, but not domains that are folded in the endoplasmic reticulum, such as single-chain antibodies72. A second strategy for adenovirus retargeting uses adaptors to non-covalently link particles to larger specificity domains44,53 (Fig. 1): X-ray crystallography structural data were recently used to develop universal adaptors that have high affinity for adenovirus fibre. These adaptors were then linked to DARPin specificity domains to retarget the cell entry of coated particles73. Such a universal adaptor makes it possible to target many different tumour markers using a single starting virus but, as discussed above for shielding, chemical virus retargeting is limited to a single round of replication.

Post-entry targeting. In this section we consider two post-entry targeting principles: positive targeting — which selectively promotes the expression of viral genes or engineered transgenes in target cells using tumour-specific transcriptional control74,75 — and negative targeting — which restricts infection in non-target cells using tissue-specific miRNAs that recognize target sequences that have been engineered into oncolytic virus genomes (Fig. 3). These strategies are complementary and are being applied, sometimes in combination, to multiple virus families (Fig. 2).

Figure 3: Post-entry targeting.
figure 3

a | Positive transcription targeting relies on promoters that are highly expressed in cancer cells to stimulate the preferential expression of viral genes or transgenes in tumours. b | Negative targeting depends on microRNA (miRNA) expression in normal cells to restrict the replication of vectors that express miRNA-recognition sequences within their genomes. Tumours have decreased expression of certain miRNAs, which renders them unable to restrict vector replication.

PowerPoint slide

Tumorigenesis is, in part, driven by aberrantly high transcription levels of genes that are not expressed in normal tissue (Box 1). Positive transcriptional targeting can control the expression of virus genes as they have been determined to be dependent on overexpressed, tumour-specific promoters76 (Fig. 3a). Positive transcriptional targeting has been most extensively applied to adenovirus using differentially expressed tumour-associated promoters, such as telomerase (reviewed in Refs 77, 78, 79). Novel tumour-associated promoters can be identified for individual diseases using gene expression profiling, for example, by comparing transcriptional profiles between hepatocellular carcinoma (HCC) and normal liver76. This strategy has been used to identify and validate HCC-specific promoter expression in vitro and in HCC xenografts in mice76, and then to exploit those promoters to activate the expression of wild-type virulence factors, such as HSV-infected cell protein 27 (ICP27) and ICP34.5, specifically in diseased tissues80,81. Positive replication targeting can be applied to any virus family that relies on the cellular machinery for transcription but not to viruses that use virally encoded polymerases for replication in the cytoplasm, such as measles virus and vaccinia virus (Fig. 2).

An alternative strategy for regulating replication is the insertion of miRNA target sequences within the untranslated regions (UTRs) of virus transcripts to provide negative post-transcriptional regulation in non-target tissues (Fig. 3b). Proof-of-principle for this targeting strategy was first shown using an oncolytic picornavirus that had miRNA target sequences for muscle-specific miRNA in its genome. Indeed, virus replication in muscles was minimal owing to the cellular miRNA machinery recognizing and degrading viral transcripts, thereby eliminating toxic myositis without negatively affecting viral oncolysis82. Negative replication targeting is versatile and has been applied to Adenoviridae83,84,85,86,87,88, Herpesviridae89,90, Paramyxoviridae91, Poxviridae92 and Rhabdoviridae93. The common principle is to express perfect-match miRNA target sequences, often multiple sequences in tandem, in the UTRs of essential viral genes. These target sequences are chosen on the basis of abnormally low expression of specific miRNAs in tumours. With increasing knowledge about miRNA expression in tumours94,95 (Box 1) and the relative ease with which these target sequences can be incorporated into virus genomes without negatively affecting replication, this tumour-targeting strategy has broad applicability (Fig. 2).

Combined post-entry targeting. Positive and negative replication targeting have recently been combined to create an HSV virus that depends on the liver-specific apolipoprotein E–α-1-antitrypsin (AAT) promoter for the expression of envelope glycoprotein H and is restricted in normal liver (but not in tumours) by three differentially expressed miRNAs90. This combination of positive and negative targeting effectively blocks replication of this virus in tissues other than hepatic tumours without the need to remove the virulence factors that are necessary for optimal oncolytic efficacy. Oncolytic adenovirus has also been dually targeted using positive and negative replication engineering to restrict virus infection in normal liver but not in multiple tumour types85. When combined, these targeting strategies effectively limit virus replication to tumour cells, even if promoter and miRNA expression are insufficient for tight restriction alone. As next-generation viruses that have more wild-type characteristics are developed, it will be possible to counteract their potential for greater toxicity using combinations of entry and post-entry retargeting strategies that are applicable to many different virus families.


Oncolytic viruses must infect and kill tumour cells to achieve efficacy, and despite intratumoural replication, accessing and infecting 100% of tumour cells remains a major clinical challenge for therapeutic viruses. Therefore, the therapeutic efficacy of oncolytic viruses can be enhanced using strategies that induce 'bystander cell killing', whereby a protein that is expressed by the oncolytic virus sensitizes both the infected cell and surrounding uninfected cells to subsequent combination therapies or immune destruction. Prodrug convertases that are expressed from oncolytic viruses can enhance the efficacy of chemotherapy by activating prodrugs, ion transport proteins can promote radiation poisoning of tumours owing to the concentration of radioisotopes, and immunostimulatory factors can induce innate and adaptive immune responses to tumour-associated antigens (Fig. 4). We discuss these three types of virus arming below.

Figure 4: Arming strategies that induce bystander cell killing.
figure 4

a | Convertase enzymes that are expressed in infected cells metabolize prodrugs into toxic metabolites that diffuse and kill uninfected tumour cells. b | The sodium–iodide symporter (NIS) concentrates radioactive ions in infected cells, which induces radiation poisoning of uninfected bystander tumour cells. c | Immunostimulatory transgenes that are expressed in infected cells prime responses against tumour antigens, which causes the systemic destruction of tumour cells.

PowerPoint slide

Prodrug convertases. Prodrug convertases include the thymidine kinase96,97, the cytosine deaminase97 and the purine nucleoside phosphorylase (PNP)98 systems. The corresponding genes have all been expressed by oncolytic viruses to activate non-toxic precursors, which generate highly toxic metabolites in the tumour microenvironment99. The HSV thymidine kinase phosphorylates ganciclovir to generate ganciclovir triphosphate, cytosine deaminase converts chemotherapeutic 5-fluorocytosine (5-FC) to 5-fluorouracil (5-FU), and PNP converts fludarabine phosphate into 2-fluoroadenine99. These nucleoside analogues are incorporated into the DNA of replicating cells, which halts replication and ultimately results in cell death99. The common goal of these strategies is to reduce systemic toxicity by giving lower doses of minimally toxic chemotherapeutic drugs that are only converted to highly toxic metabolites within the tumour microenvironment (Fig. 4a). Ganciclovir triphosphate cannot easily diffuse between cells after activation100, making it a good option for the selective elimination of the toxicity that is induced by retrovirally transduced cells101,102. However, the activated metabolites 5-FU and 2-fluoroadenine can both diffuse out of the infected cell and into surrounding cells98,103 to induce chemotherapeutic bystander killing (Fig. 4a), making them clinically relevant arming strategies for virotherapy. We focus in this section on the most clinically advanced viruses that are armed with the cytosine deaminase and PNP transgenes, as well as on promising preclinical viruses that are ready to enter future clinical trials.

Cytosine deaminase and PNP have both been incorporated into several virus classes that have been preclinically tested, including viruses that are based on Herpesviridae104, Adenoviridae105, Poxviridae106,107, Paramyxoviridae108,109 and Rhabdoviridae110. The most advanced cytosine deaminase virus is a replication-competent retrovirus known as Toca 511, which integrates the cytosine deaminase transgene into the genome of infected cells to establish permanent reservoirs of tumour cells that are sensitive to subsequent rounds of chemotherapy using 5-FC19. Toca 511 is currently being tested in combination with 5-FC in Phase I and II clinical trials using intratumoural administration in patients with grade 4 glioblastoma multiforme (Table 1). HSV and VSV viruses that express cytosine deaminase are also being preclinically developed for combination therapies using 5-FC104,110.

Intratumoural administration of an adenovirus with the PNP transgene in combination with intravenous fludarabine has been used to treat patients with head and neck tumours111 ( identifier: NCT01310179). In general, cytosine deaminase and PNP transgenes can be applied to many virus families because their small sizes and low cellular toxicities incur minimal negative effects on in vitro virus fitness or production. However, the timing of prodrug dosing in vivo must be optimized to ensure that virus replication and spread is sufficient for maximal synergistic effects with the chemotherapeutic prodrug104,109,112. In this respect, a PNP-expressing measles virus that has been retargeted to CD20 has been extensively tested in combination with fludarabine in preclinical models of lymphoma108,109.

Radiosensitization. The normal physiological function of the human sodium–iodide symporter (NIS) is to transport iodide ions into cells, which occurs predominantly in the thyroid but also in the stomach, salivary glands and mammary glands113,114. When NIS is expressed from the genome of an oncolytic virus, infected cells concentrate iodide or similar isotopes intracellularly. During virotherapy, γ-emitting isotopes, such as 123I and pertechnetate, can be administered to visualize virus replication using single-photon emission computed tomography (SPECT; Box 2), whereas β-emitting isotopes, such as 131I and 188Re, can be administered to specifically induce radiation poisoning within the tumour microenvironment (Fig. 4b), in analogy to the clinically well-established radiotherapy that is used for metastatic thyroid cancer.

The NIS transgene system has undergone extensive preclinical development in multiple virus families, and radiovirotherapy has consistently achieved synergistic tumour destruction in several radiosensitive preclinical disease models115,116,117. Measles virus is an especially efficacious virus for NIS-mediated imaging (Box 2) and radiovirotherapy. Preclinical studies of lymphoma118, ovarian cancer119, myeloma120 and mesothelioma121, in addition to many other disease models117, have used NIS expression and SPECT–computed tomography (SPECT–CT) imaging to visualize and quantify virus replication and enhance disease regression using combination radiovirotherapy. Phase I clinical trials using NIS-expressing measles virus have been initiated for ovarian cancer, myeloma, mesothelioma and head and neck cancer ( identifiers: NCT00408590, NCT00450814, NCT01503177 and NCT01846091, respectively), with all four studies using SPECT–CT intervention to image virus replication in patients. The results of these clinical trials and the continued translation of diverse NIS-expressing viruses in different tumour types will inform the applicability of SPECT–CT imaging and radiovirotherapy interventions to future trials.

Immunostimulation. Advanced clinical trials of oncolytic viruses from three different virus families have used the combination of lytic infection and immunostimulatory transgene expression to induce antitumour immunity (Fig. 4c). The most successful strategy that has been used so far is the expression of GM-CSF to stimulate the production of granulocytes and monocytes, which in turn stimulate adaptive immunity against tumour-associated antigens122. HSV, adenoviruses and vaccinia viruses that express GM-CSF are currently used in clinical trials and have repeatedly been shown to have clinical efficacy122,123, especially in diseases that are amenable to immunotherapy, such as melanoma (Box 1). These advanced clinical studies have shown that these viruses work via immunotherapy rather than via virus-mediated tumour lysis: only moderate replication of vaccinia JX-594 was documented in tumour biopsies or measured by the detection of GM-CSF in the blood, but lytic replication or recovery of infectious virus from harvested tumours were not reported18,124. The HSV virus talimogene laherparepvec9,125 is currently being tested against melanoma in a Phase III clinical trial that will directly compare the virus expressing GM-CSF with GM-CSF administration alone. This study will greatly improve our understanding of the relative contributions of virus replication and transgene expression in stimulating antitumour immunity. The successes of current oncolytic viruses that express immunostimulatory transgenes, even in the absence of robust intratumoural replication, highlights the potential for new viruses that have greater replication and transgene expression to stimulate improved, and potentially curative, immune responses against the tumour microenvironment.

Future directions

Our increasing knowledge of the determinants of virus tropism, and of the proteins and gene expression pathways that are altered in tumour tissues, are important resources for developing next-generation viruses. With few exceptions, such as vaccinia virus and VSV, virus families have adapted to tissue niches, which are often defined by specific receptors; for example, measles virus initially uses SLAM to establish systemic infection in lymphatic organs, and then nectin 4 to infect epithelia126,127,128,129. Oncolysis, similarly to wild-type virus spread, depends not only on efficient cell entry via specific receptors, but also on efficient replication; in particular, viruses often activate or exploit certain gene expression pathways that facilitate their replication, and similar events may be required for efficient oncolysis. Consequently, efficient cell entry might not always result in efficient oncolysis; for example, measles virus-based oncolytic viruses are likely to replicate most efficiently in SLAM-positive haematological malignancies, such as lymphoma, and in nectin 4-positive epithelial malignancies, such as breast130, ovarian131 and lung132 tumours. As there is an increasingly diverse pool of oncolytic viruses and our knowledge of tumour biology is improving, choosing virus classes to target specific tumour types, and then stratifying individual patients on the basis of disease susceptibility to virotherapy, should soon become standard practice.

Oncolytic virotherapy has, to date, proven to be very safe in humans. By contrast, the therapeutic efficacy of oncolytic viruses in humans has been less than expected from preclinical studies. Early protocols were based on highly attenuated viruses, and generalized attenuation interfered with clinical efficacy. Next-generation viruses will benefit from retaining wild-type replicative potential in disease tissue, combined with engineered entry and post-entry restriction mechanisms that maintain current safety profiles in off-target tissues. In addition, arming strategies that combine chemo-, radio- and immunotherapies will be potentiated by greater virus replication, and advances in in vivo imaging will enable real-time tracking of virus spread (Box 2).

New knowledge and new viruses can now be used to address existing clinical needs. Although many challenges remain for oncolytic virotherapy, such as the production and validation of an increasing number of therapeutic viruses, there are strategies to overcome them. There is no single best virotherapy approach for all tumours, but creative engineering strategies that use viruses from diverse families are yielding new viruses that are likely to be highly efficacious in specific applications. Understanding the susceptibility and resistance of different tumour types and progress in stratifying individual patients will facilitate the recruitment of those patients who are most likely to benefit from a new virus. In synergy with approved cancer therapies and antitumour immunity, these new viruses are expected to achieve the clinical promise of oncolytic virotherapy.