Loading
Some viruses kill cancer cells and leave their neighbours alone, because a cell that broke its own virus alarm on the way to becoming a cancer cannot switch it back on. Four such viruses have an approval somewhere in the world. Not one of them has shown a survival benefit in a randomised trial. This page is the biology, the hundred-year history, every approved product with what its evidence really shows, the engineering, the programmes that failed and why, and the 2024 case of a virologist who injected her own tumour.
A healthy cell that finds a virus inside it has a set of things it can do. It makes interferon, a chemical alarm that warns itself and its neighbours. It slows its own protein factory so the virus has nothing to build with. It calls the immune system. If none of that is enough, it kills itself before the virus can finish copying.
Becoming a cancer means switching off brakes, and the interferon response is a brake. It stops cells dividing and it tells them to die. A cell that has broken it can grow more freely, and many cancers have broken some part of it on the way to becoming cancers.
That leaves them with a hole where the alarm used to be. A virus that a healthy cell would shut down in hours can get in, take over and keep copying until the cell bursts. The virus is not clever and it is not seeking out cancer. It is simply the one kind of cell in the body that cannot stop it.
When the infected cancer cells burst they spill two things at once: pieces of the tumour, and the unmistakable molecular signature of a virus. The immune system arrives for the virus and finds the tumour. That second effect, not the bursting, is what the field is actually trying to produce.
Tumours were seen shrinking during ordinary virus infections before anyone could grow a virus. The field was tried, abandoned, and restarted twice. Nothing in it is new except the ability to design the virus rather than hunt for one.
From the turn of the nineteenth century, when viruses were first recognised, case reports described cancers regressing while a patient had an ordinary virus infection. These were observations, not experiments, and they were the whole evidence base for decades.
Clinicians used body fluids containing human or animal viruses to infect patients with cancer. Most often the immune system arrested the virus and the tumour was unaffected. In immunosuppressed patients the infection persisted and tumours did regress, but the damage to normal tissues was unacceptable. Southam and Moore's 1952 report of Egypt 101 virus is one of the documented series.
With rodent models available, researchers tried to force the evolution of tumour-specific viruses by repeated passage. Success was limited, tumour specificity could not be made reliable, and most researchers left the field.
Martuza and colleagues deleted thymidine kinase from herpes simplex virus type 1, making it dependent on the host cell's nucleotide pool. It killed human glioma cells in culture, inhibited implanted human gliomas in mice, and prolonged survival when injected into intracranial tumours. Reverse genetics turned the field from selection into design.
An adenovirus lacking the E1B 55-kilodalton protein appeared to replicate only in p53-deficient tumour cells. It went into large trials, including a randomised study with cisplatin and fluorouracil in recurrent head and neck cancer. In 2004 O'Shea and colleagues showed the selectivity tracks late viral RNA export, not p53, which is why clinical response never sorted by p53 status.
Coffey and colleagues showed reovirus needs an activated Ras pathway. Stojdl and colleagues showed vesicular stomatitis virus exploits the broken interferon response. Together these gave the field a mechanistic argument that did not depend on any single gene.
A 160-patient randomised trial added intratumoural H101, an E1B-55 kilodalton-deleted adenovirus of essentially the ONYX-015 design, to cisplatin-based chemotherapy in squamous cell cancer of the head and neck or oesophagus. Response rate was 78.8 per cent against 39.6 per cent with chemotherapy alone. No survival result was reported. H101 was approved in China in November 2005 and marketed as Oncorine.
Two measles-seronegative patients with drug-refractory myeloma received one intravenous infusion of 10^11 infectious units of an engineered measles virus carrying the sodium iodide symporter gene. Both responded; one had a durable complete remission at all sites, with radioiodine imaging showing the virus replicating inside the tumours. Both patients were chosen for having no pre-existing measles antibodies, which excludes almost everyone.
The first approval outside China, on a phase 3 trial in which the durable response rate was 16.3 per cent against 2.1 per cent for granulocyte-macrophage colony-stimulating factor. Median overall survival was 23.3 against 18.9 months, which did not reach statistical significance.
403 patients having resection for recurrent high-grade glioma were randomised to a replicating retroviral vector injected into the resection cavity plus a prodrug, or to standard care. Median overall survival was 11.10 against 12.22 months, hazard ratio 1.06.
In June 2021 Japan's Ministry of Health, Labour and Welfare approved Delytact Injection as a regenerative medical product, on a single-arm investigator-initiated trial of 19 patients in which one-year survival after starting treatment was 84.2 per cent. The approval is conditional and time-limited, with a use-results comparison survey and resubmission required within seven years.
A replication-deficient adenovirus that delivers the interferon alfa-2b gene to the bladder lining. 53.4 per cent of patients with carcinoma in situ had a complete response within three months and 45.5 per cent of those responders still had it at twelve. It is gene delivery, and it is regularly counted as a win for oncolytic virotherapy.
692 patients with advanced melanoma were randomised to talimogene laherparepvec plus pembrolizumab or placebo plus pembrolizumab. Neither progression-free survival nor overall survival improved.
A 50-year-old virologist with locally recurrent, muscle-invasive breast cancer injected her own tumour with an Edmonston-Zagreb measles vaccine strain and then a vesicular stomatitis virus, both grown in her own laboratory, before any other treatment for the recurrence. The tumour shrank enough for simple excision. The report drew a bioethics literature about whether journals should publish self-experimentation.
In 140 patients with melanoma that had progressed on anti-PD-1 therapy, an engineered herpes virus with nivolumab produced a 32.9 per cent confirmed response rate, and responses in uninjected lesions, including visceral ones, matched injected lesions in frequency, depth and duration. That is the strongest clinical evidence to date that the mechanism is systemic immunity rather than local lysis. A randomised confirmatory trial is required.
Two distinctions decide how to read this table. The first is whether the virus replicates in the tumour at all: nadofaragene firadenovec and aglatimagene besadenovec are replication-defective vectors that deliver a gene, and are routinely counted as oncolytic virus successes. The second is whether the trial measured a response in the treated lesion or a survival benefit. Every replicating oncolytic virus in the table was approved on a response endpoint. The only randomised survival result belongs to a product that does not replicate.
| Product | How it is given | Approval | Evidence | What it achieves | |||
|---|---|---|---|---|---|---|---|
H101 H101, Oncorine | Adenovirus type 5 | Oncolytic, replicates | Injected into the tumour, 5.0 x 10^11 to 1.5 x 10^12 viral particles daily for five days every three weeks. | China, November 2005. Never approved outside China. | One randomised phase 3 trial, 160 patients recruited and 123 analysed, adding H101 to cisplatin-based chemotherapy in squamous cell cancer of the head and neck or oesophagus. | Response rate 78.8 per cent (41 of 52) versus 39.6 per cent (21 of 53) with cisplatin and fluorouracil alone. No overall or progression-free survival result was reported. | Local response only |
Talimogene laherparepvec T-VEC, Imlygic | Herpes simplex virus type 1 | Oncolytic, replicates | Injected into lesions, 10^6 plaque-forming units per mL first, then 10^8 per mL three weeks later and every two weeks. | United States 2015, European Union 2015, for unresectable melanoma with injectable lesions. | OPTiM, an open-label phase 3 trial randomising 436 patients 2:1 against subcutaneous granulocyte-macrophage colony-stimulating factor. | Durable response rate 16.3 per cent versus 2.1 per cent, odds ratio 8.9. Median overall survival 23.3 versus 18.9 months, hazard ratio 0.79, p = 0.051, not significant. The comparator was not a modern melanoma treatment. | Local response only |
Teserpaturev G47 delta, Delytact | Herpes simplex virus type 1 | Oncolytic, replicates | Injected into the tumour, repeatedly, for up to six doses. | Japan, June 2021, as a regenerative medical product for malignant glioma. Conditional and time-limited: a use-results comparison survey and resubmission are required within seven years. | A single-arm, investigator-initiated phase 2 trial in 19 adults with residual or recurrent glioblastoma after radiotherapy and temozolomide, stopped early when its endpoint was met. | One-year survival after starting treatment 84.2 per cent (16 of 19); median overall survival 20.2 months from starting treatment. Best response over two years was one partial response and 18 stable disease, because the lesion characteristically enlarges while contrast enhancement clears. | Single arm, no comparison |
Vusolimogene oderparepvec RP1, Tudriqev | Herpes simplex virus type 1 | Oncolytic, replicates | Injected into lesions, up to eight doses of up to 10 mL, with nivolumab. | United States, accelerated approval on 6 August 2026, with nivolumab, for melanoma that has progressed on anti-PD-1 therapy. A randomised confirmatory trial is required. | IGNYTE, a single-arm registrational cohort of 140 patients with confirmed progression on anti-PD-1 therapy, assessed by independent central review. | Confirmed response rate 32.9 per cent with 15.0 per cent complete responses; median duration of response 33.7 months. Uninjected lesions, including visceral ones, responded with the same frequency, depth, duration and timing as injected ones. | Uninjected lesions responded |
| Adenovirus type 5 | Oncolytic, replicates | Into the bladder, 1 x 10^12 viral particles per 0.8 mL weekly for six weeks, then maintenance. | Not approved. Under rolling review in the United States. | BOND-003 Cohort C, a single-arm international phase 3 trial in 112 treated patients with BCG-unresponsive non-muscle-invasive bladder cancer with carcinoma in situ. | Complete response at any time in 83 of 110 patients, 75 per cent (95% CI 66.3 to 83.2), at a median follow-up of 25.8 months. No grade 3 or 4 treatment-related adverse events. Single-arm, so no comparison with an alternative. | Local response only | |
Nadofaragene firadenovec Adstiladrin | Adenovirus, replication-deficient | Gene delivery, does not replicate | Into the bladder, a single 75 mL dose of 3 x 10^11 viral particles per mL, repeated at months 3, 6 and 9. | United States, December 2022, for BCG-unresponsive non-muscle-invasive bladder cancer with carcinoma in situ. European Union conditional authorisation 28 May 2026. | A single-arm, open-label, repeat-dose phase 3 trial, 157 treated and 151 analysed. | Complete response within three months in 55 of 103 patients with carcinoma in situ, 53.4 per cent; maintained at twelve months in 25 of those 55, 45.5 per cent, which is about a quarter of the cohort. Single-arm. | Local response only |
Aglatimagene besadenovec CAN-2409 | Adenovirus, replication-defective | Gene delivery, does not replicate | Three courses injected into the prostate, 5 x 10^11 viral particles, with valaciclovir, alongside external beam radiotherapy. | Not approved. Phase 3 reported in 2026. | A randomised, double-blind, placebo-controlled phase 3 trial in 745 men with intermediate or high-risk localised prostate cancer, randomised 2:1. | Median disease-free survival not reached versus 86.1 months with placebo, hazard ratio 0.70 (95% CI 0.52 to 0.94), p = 0.016, at a median follow-up of 50.3 months. Grade 3 or worse treatment-emergent adverse events in 8 per cent versus 7 per cent. | Randomised survival benefit |
E1B 55-kilodalton gene deleted, essentially the ONYX-015 design; also a partial E3 deletion.
ICP34.5 deleted so the virus cannot block the host shutdown of protein synthesis in a normal cell; ICP47 deleted, which also moves US11 to an immediate-early position and restores replication; human granulocyte-macrophage colony-stimulating factor inserted.
Triple-mutated third generation: the two mutations of the second-generation backbone plus a further deletion, engineered both for tumour selectivity and for better antigen presentation.
Granulocyte-macrophage colony-stimulating factor plus a fusogenic glycoprotein, so infected cells fuse with their neighbours and die in a way that presents antigen well.
E1A driven by the E2F-1 promoter, so replication depends on a defective retinoblastoma-E2F pathway; granulocyte-macrophage colony-stimulating factor inserted.
Not an oncolytic virus. A replication-deficient recombinant adenovirus carrying human interferon alfa-2b complementary DNA, delivered with an excipient that lets the bladder lining take it up. It transduces the urothelium, which then makes interferon; the virus does not replicate or lyse.
Not an oncolytic virus. A replication-defective adenovirus delivering the herpes simplex virus thymidine kinase gene into the prostate, followed by oral valaciclovir, which the enzyme converts into a cytotoxic nucleotide inside the transduced cells.
This is the part that makes oncolytic virotherapy a technology rather than an anecdote. Selectivity is built by removing a viral gene whose job a tumour cell already does, or by putting replication behind a promoter only a tumour fires. Potency is built by inserting something that turns lysis into an immunisation. Then there is the part nobody has solved, which is getting the virus there and keeping antibody off it.
| The move | What is changed | Why it works | Where it is used | |
|---|---|---|---|---|
| Delete thymidine kinase | deletion | Remove the viral thymidine kinase gene, so the virus cannot make its own nucleotide precursors. | A dividing tumour cell has a large nucleotide pool and supplies the missing function; a resting neurone does not. This was the first selectivity deletion ever made, and it also removes the virus's sensitivity to aciclovir and ganciclovir, which is the safety net if infection runs away, so later designs moved to other genes. | dlsptk (Martuza 1991), the ancestor of the herpes lineage. |
| Delete ICP34.5 | deletion | Remove the herpes neurovirulence gene whose product reverses the host shutdown of protein synthesis. | In a normal cell, protein kinase R shuts translation down when it senses viral RNA and the virus cannot proceed. In a tumour cell with a defective protein kinase R or Ras response, translation continues and the virus replicates. This is the deletion that carries selectivity in every approved herpes product. | Talimogene laherparepvec; teserpaturev; vusolimogene oderparepvec. |
| Delete ICP47 | deletion | Remove the herpes gene that blocks the transporter associated with antigen processing, which loads peptides onto class I molecules. | An infected cell that cannot present antigen is invisible. Deleting ICP47 restores presentation, so the infected tumour cell displays both viral and tumour peptides. In herpes simplex virus type 1 the deletion also moves US11 into an immediate-early position, which partly restores the replication lost by deleting ICP34.5. | Talimogene laherparepvec; teserpaturev. |
| Delete adenovirus E1B 55K | deletion | Remove the adenoviral protein that inactivates p53. | The original claim was that the virus then needs the cell's p53 to be already lost, making it selective for p53-mutant tumours. That explanation did not survive: selectivity tracks late viral RNA export instead, and clinical response never sorted by p53 status. The deletion still produces some selectivity, for a reason other than the one it was designed on. | ONYX-015; H101 (Oncorine). |
| Put replication under a tumour-active promoter | targeting | Drive the essential early gene from a promoter that only fires in the tumour. | Instead of removing a viral function, this removes the virus's permission to start. Replication depends on a transcription factor the tumour has in excess, so a normal cell never licenses the first step. | Cretostimogene grenadenorepvec, whose E1A is driven by the E2F-1 promoter and so depends on a defective retinoblastoma-E2F pathway. |
| Insert granulocyte-macrophage colony-stimulating factor | insertion | Arm the virus with a cytokine that recruits and matures antigen-presenting cells. | Lysis alone releases antigen into a tumour that is bad at presenting it. Producing the cytokine inside the tumour turns lysis into an immunisation and concentrates a drug that is toxic when given systemically. | Talimogene laherparepvec; vusolimogene oderparepvec; cretostimogene grenadenorepvec. |
| Insert a fusogenic glycoprotein | insertion | Make infected cells fuse with their neighbours into a dying multinucleated mass. | Fusion spreads the effect beyond the cells the virus can reach and produces a highly immunogenic form of cell death, which is the mechanism behind the uninjected-lesion responses seen with vusolimogene oderparepvec. | Vusolimogene oderparepvec. |
| Insert a reporter gene you can image | insertion | Add the sodium iodide symporter, so infected cells take up radioiodine and can be imaged. | Without it, nobody knows whether an intravenous dose reached the tumour at all, which is the question that decides whether a negative trial means the virus does not work or never arrived. It also allows the same gene to be used therapeutically with a beta-emitting isotope. | MV-NIS, the engineered measles virus used in the 2014 myeloma report. |
| Give up on systemic delivery and inject the tumour | delivery | Deliver the virus directly into the lesion, or into a body cavity that contains it. | It works, and it is the reason every approval is in a disease with an accessible target: skin and nodal melanoma, the bladder lining, the prostate, a glioma resection cavity. It also caps the addressable population at patients whose disease can be reached with a needle. | All seven products in the table above. |
| Work around pre-existing and rising antibody | delivery | Select seronegative patients, use a virus most people have not met, switch viruses part-way, or hide the virus inside carrier cells. | This is the hardest problem in the field. Most adults have neutralising antibody to measles and to common adenovirus serotypes, and titres rise a hundredfold within weeks of the first dose, so the second and third doses face an immune system that has already learned the virus. The 2014 myeloma remission required patients selected for being measles-seronegative and a dose of 10^11 infectious units. The 2024 self-experiment switched from measles virus to vesicular stomatitis virus after three weeks for exactly this reason. | MV-NIS in seronegative patients; sequential measles then vesicular stomatitis virus in the 2024 case report. |
Five programmes, each tested at a scale that should have settled the question, and what each one taught. A field is best judged by what it did when it was given a fair test.
The argument is clean. Checkpoint inhibitors work best where a T-cell response already exists; most tumours have none; an oncolytic virus kills tumour cells in a way that is loud to the innate immune system and releases tumour antigen alongside a viral danger signal. In principle it makes a cold tumour hot and a checkpoint inhibitor effective. The biology supporting it is real: in the 2024 self-experiment the excised tumour showed lymphocyte infiltration rising from 10 to 45 per cent and PD-L1 appearing in a tumour that had been PD-L1 negative, and in IGNYTE response was associated with increased CD8-positive T-cell infiltration and PD-L1 expression.
Encouraging complete response rate, with evidence of increased T-cell infiltration in treated lesions. Uncontrolled, and the trigger for the phase 3.
Randomised, double-blind, placebo-controlled, 692 patients. No improvement in progression-free survival (hazard ratio 0.86) or overall survival (hazard ratio 0.96). Response rate rose from 41.3 to 48.6 per cent and did not translate into either endpoint.
32.9 per cent confirmed response rate in 140 patients whose melanoma had progressed on anti-PD-1 therapy, with uninjected lesions responding as often and as deeply as injected ones. Single-arm, so the contribution of nivolumab alone cannot be separated; the randomised confirmatory trial has not reported.
In mice, giving the virus early and then resecting sensitised otherwise refractory tumours to checkpoint blockade and prevented relapse in most animals. The authors proposed testing this in the window between diagnosis and surgery rather than in late metastatic disease.
The only randomised test of the combination hypothesis in a treatment-naive population was negative. What survives is a narrower claim, supported by a single-arm cohort: after anti-PD-1 therapy has already failed, adding an engineered virus produces responses in lesions that were never injected. That is worth confirming and has not been confirmed. Anyone citing synergy between oncolytic viruses and checkpoint blockade as established is citing the phase 1b and skipping the phase 3.
This is one person, and it is a case report. It is not evidence that oncolytic virotherapy works, and it is not evidence that anyone should treat themselves. The authors say so in the paper: self-medicating with oncolytic viruses should not be the first approach to a diagnosed cancer. The patient was a professional virologist who grew and titred the viruses in her own laboratory and had oncologists monitoring her and ready to intervene. If you are facing a cancer diagnosis, the useful thing in this section is the argument for testing virotherapy before surgery in a trial, not the protocol.
A 50-year-old virologist had been treated for multifocal invasive ductal triple-negative breast cancer in 2016 with mastectomy and adjuvant chemotherapy. A local recurrence was excised in 2018, leaving a small seroma that was monitored. By 2020 that had become a hard, inflamed 2 cm nodule invading the pectoral muscle and infiltrating the skin. Rather than a second round of chemotherapy she told her oncologists she would first inject the tumour with viruses of the kind then in clinical development, and they agreed to monitor her and to intervene with conventional treatment if she came to harm or the tumour grew.
Magnetic resonance imaging, positron emission tomography with computed tomography and two independent ultrasound estimates all gave a tumour volume of 2.47 plus or minus 0.06 cm3, with invasion into the pectoral muscle and infiltration of the skin, and no metastatic or nodal disease. A core needle biopsy showed the tumour had changed from triple-negative at first diagnosis to HER2 3+.
An Edmonston-Zagreb measles vaccine strain, given into the tumour at three to four day intervals, seven times, totalling 7.89 log CCID50. Grown in MRC-5 and then Vero cells as clarified culture supernatant, not purified from host-cell nucleic acid and protein.
The tumour reached its largest volume, 4.28 cm3, the worst clinical and ultrasound picture of the course. A similar transient increase has been described in trials of oncolytic virotherapy in liver cancer.
A vesicular stomatitis virus Indiana strain, given into the tumour three times, separated by two weeks and then one week, totalling 9.07 log CCID50. The switch of virus was deliberate: antibody raised against the first virus would neutralise it, so a virus the immune system had not yet learned was used instead.
Two weeks after the first vesicular stomatitis virus dose the tumour measured 2.17 cm3 and lymph nodes enlarged in both axillae, which the authors attribute to infiltration by virus-specific lymphocytes.
The tumour was excised. Pathological volume was 0.91 cm3, and it was confined to the subcutis with no infiltration of skin or pectoral muscle, in contrast to the baseline imaging. The resection was simple and non-invasive, which it would not have been at baseline.
Two months after excision, one subcutaneous dose of measles virus around the surgical suture. Because the excised tumour was HER2 3+, one year of adjuvant trastuzumab followed, in line with guidelines.
Antibody titres rose a hundredfold during the course. Switching from measles virus to vesicular stomatitis virus after three weeks put a virus in front of an immune system that had not yet learned it. Sequential virotherapy has been proposed and shown in preclinical models and is not part of any approved regimen.
Ten injections in under seven weeks, to hold the concentration of infectious virus in the tumour high rather than to give a dose and wait. Approved regimens dose every two or three weeks.
Oncolytic viruses are normally tested last, in people with widely metastatic disease and exhausted immune systems. The preclinical case for testing them first, in the window between diagnosis and surgery, was published in 2018 and this case is consistent with it. That is the hypothesis a trial should test.
Four of these six were named by the field itself in 2012 and none has been solved.
Every approval is in a disease with an accessible target. Intravenous delivery has to survive complement, antibody and sequestration in the liver, and then exceed a threshold concentration in the blood before any virus reaches the tumour at all. The 2014 myeloma report is the proof it can work and the demonstration of what it costs: two patients selected for having no measles antibodies, and a dose of 10^11 infectious units.
Most adults have antibody to measles and to the common adenovirus serotypes before the first dose. Titres rise a hundredfold within weeks of starting, so the doses that matter face an immune system that has already learned the virus. Selecting seronegative patients, switching viruses part-way and hiding virus inside carrier cells are all attempts to work around a problem nobody has solved.
The virus needs to spread, which means the antiviral response must be held off. The therapeutic effect needs an immune response, which means it must then be provoked. These requirements conflict and the field has no principled way to sequence them. It was named as an unsolved problem in 2012 and it is still unsolved.
H101, talimogene laherparepvec, teserpaturev, cretostimogene and vusolimogene oderparepvec were all supported by response or complete response rates. Not one of them has a randomised overall survival benefit. In a field where the treatment is injected into the tumour being measured, that endpoint is the easiest to obtain and the least informative.
Intravenous dosing needs orders of magnitude more virus per patient than intratumoural injection. Yields have to rise by a similar factor for systemic delivery to be practical, and this constraint sits upstream of every clinical question about the route.
The selectivity argument rests on interferon-pathway defects, Ras pathway activation and receptor density, none of which is measured before treatment. There is no companion diagnostic for any oncolytic virus, so trials enrol patients who cannot be expected to respond alongside those who can.
Records for the scientists whose work this page rests on. Public professional information only.
Every claim on this page is sourced to the paper linked beside it. The full set of records, with the numbers, the caveats and what each paper does and does not show, sits on the oncolytic viruses technology page and under key papers.