In Vivo CAR-T: Where the Field Stands and What It Means for Biomarkers

In Vivo CAR-T: Where the Field Stands and What It Means for Biomarkers

Autologous CAR-T changed the outlook for patients with B-cell cancers and multiple myeloma, but it is still made one patient at a time. Cells are collected by apheresis, shipped, engineered and expanded, then returned after lymphodepleting chemotherapy. In vivo CAR-T aims to remove most of that chain: a single intravenous dose of a targeted vector programmes the patient’s own T cells inside the body.

Eighteen months ago this was a preclinical idea with a handful of first-in-human trials. As of October 2026, there are peer-reviewed clinical reports in multiple myeloma, lymphoma and autoimmune disease, and big pharma has made a series of acquisitions in the space. In vivo CAR-T cell therapy is now a real clinical modality.

For translational and biomarker teams, it also changes the questions you need to answer. With no manufactured product to release, there is no certificate of analysis telling you how many CAR-positive cells the patient received. Everything you need to know about dose, expansion, specificity and safety has to be measured in the patient. This article covers where the field stands and what that means for biomarker strategy.

Key takeaways

• In vivo CAR-T delivers the CAR gene directly to T cells in the patient, using targeted lentiviral vectors (stable integration) or lipid nanoparticles carrying mRNA (transient expression).
• Early clinical data in multiple myeloma and B-cell lymphoma show deep responses without apheresis, ex vivo manufacturing or lymphodepletion, but cohorts are very small and follow-up is short.
• AstraZeneca, AbbVie, Gilead/Kite, Bristol Myers Squibb and Eli Lilly have all bought in vivo CAR-T companies since March 2025.
• Because the “product” is made inside the patient, pharmacokinetics, CAR expression, B-cell aplasia and off-target transduction become the primary evidence of dose and activity.
• Monitoring needs differ sharply between lentiviral and LNP-mRNA platforms, particularly for vector copy number, integration and long-term follow-up.

What is in vivo CAR-T?

Conventional (ex vivo) CAR-T engineers T cells outside the body. In vivo CAR-T instead gives the patient a vector that finds T cells (or a T-cell subset) in the blood and lymphoid tissue and delivers the CAR gene to them there. The patient’s body does the manufacturing.

The practical advantages are clear. There is no apheresis, no vein-to-vein wait, no individual manufacturing run and, in most programmes so far, no lymphodepleting chemotherapy. That could cut cost of goods, open the modality to centres without cell therapy infrastructure, and make it realistic for larger populations such as patients with autoimmune disease. AstraZeneca described the goal as allowing cell therapy to be given as a simple IV infusion without immune cell depletion when it announced its acquisition of EsoBiotec.

Lentiviral vs LNP in vivo CAR-T

Targeted lentiviral vectors

Lentiviral programmes re-target the vector envelope so that it enters T cells preferentially, often through CD3 or another T-cell surface protein, and use T-cell-restricted promoters as a second layer of specificity. EsoBiotec’s ESO-T01 is described in its phase 1 publication as a nanobody-directed, immune-shielded lentiviral vector encoding a humanised anti-BCMA CAR. Because the CAR integrates into the genome, a single dose can in principle give durable, self-renewing CAR-T cells, much as with ex vivo products.

The trade-offs come from the same property: integration means a theoretical risk of insertional events, the need to show that non-target cells were not transduced, and long-term follow-up obligations.

Targeted LNP-mRNA and circular RNA

LNP programmes carry CAR mRNA (or circular RNA) in a lipid nanoparticle decorated with a T-cell-targeting antibody. Capstan’s platform, now part of AbbVie, is an example: in a 2025 Science paper, its targeted LNPs reprogrammed CD8+ T cells, depleted B cells in cynomolgus monkeys, and the B cells that came back were mainly naive, suggesting an immune “reset”.

Expression is transient. That is attractive in autoimmune disease, where the aim is a deep but time-limited B-cell depletion, and it removes integration risk. It also means efficacy may depend on repeat dosing, and that immunogenicity against the LNP or the CAR becomes a practical question.

Where the field stands in October 2026

The deals

In vivo CAR-T has become one of the most active areas for cell therapy dealmaking. The acquisitions below are confirmed by company announcements or filings.

AstraZeneca agreed to pay up to $1 billion for EsoBiotec and completed the deal on 20 May 2025. AbbVie agreed to pay up to $2.1 billion for Capstan and completed it on 19 August 2025. Kite announced a $350 million cash deal for Interius on 21 August 2025 and closed it in October 2025. Bristol Myers Squibb agreed to pay $1.5 billion in cash for Orbital and completed the deal in December 2025. Lilly’s agreement for Kelonia, announced on 20 April 2026, is worth up to $7 billion including $3.25 billion upfront, and closed on 25 June 2026.

Independent developers remain active too. Umoja Biopharma’s UB-VV111, a CD19-directed in vivo CAR-T, was the first in vivo CAR-T to have an IND cleared by the FDA, and it is in a phase 1 study in large B-cell lymphoma and CLL, given with and without rapamycin. It received FDA Fast Track designation in September 2025, and AbbVie holds an option to license Umoja’s CD19 in vivo candidates.

The clinical data

ESO-T01 (BCMA, lentiviral). The first clinical report came in The Lancet in July 2025: four patients with relapsed or refractory multiple myeloma treated with a single infusion, without leukapheresis or lymphodepletion. A fuller phase 1 report followed in Nature Medicine in March 2026. Five heavily pretreated patients received one dose of 0.2 × 10⁹ transduction units. Four of five responded, three with stringent complete remission, and all four evaluable responders were MRD-negative at 10^-5 by day 60. Toxicity was real: all patients had grade 3 or higher adverse events, four had cytokine release syndrome (three grade 3), one had grade 1 neurotoxicity, and one patient died of spinal cord compression from an extramedullary lesion. The trial was stopped early in 2025.

KLN-1010 (BCMA, lentiviral). Kelonia presented first-in-human data from the inMMyCAR study (NCT07075185) at ASH 2025. According to the company, all four patients treated were MRD-negative, CAR-T cells reached up to 85% of circulating T cells without lymphodepletion, and there was no grade 3 or higher CRS and no ICANS. These are company-reported data from four patients with up to five months of follow-up.

LB2501 (CD19/CD20, lentiviral). At EHA 2026, investigators from Jiangsu Province Hospital reported 12 patients with relapsed or refractory non-Hodgkin lymphoma (NCT07002112). There were no responses at dose level 1; at dose level 2, all six patients responded and five had a complete response, with a median follow-up of only 2.3 months. CRS was grade 1–2 and no ICANS was reported (OncLive report).

Autoimmune disease. The New England Journal of Medicine has published two correspondence reports of in vivo CD19 CAR-T in autoimmunity: one in refractory systemic lupus erythematosus (2025) and one using a lentiviral vector in neurologic autoimmune disorders (September 2026). Capstan’s CPTX2309 is in a phase 1 study in healthy participants and patients with rheumatoid arthritis or lupus.

The pattern is consistent: deep responses and fast expansion, achieved without lymphodepletion, in very small cohorts with short follow-up. Durability, dose selection and the safety profile at scale are all still open.

Indications: blood cancers and autoimmunity first, solid tumours next

Almost every clinical in vivo programme targets CD19, CD20 or BCMA. These antigens are well validated by ex vivo CAR-T, and B-cell aplasia gives a clean, measurable readout of activity.

Autoimmune disease is where the access argument bites hardest. Ex vivo CD19 CAR-T has produced drug-free remissions in lupus and other B-cell-driven diseases, but apheresis, bespoke manufacturing and lymphodepleting chemotherapy are difficult to justify outside refractory cases. An off-the-shelf intravenous dose changes that calculation, which is why Capstan, Orna and Orbital were all built around autoimmunity rather than oncology.

Solid tumours remain the harder problem, and it is the same problem ex vivo cell therapy faces: antigen heterogeneity, poor trafficking into tumour and an immunosuppressive microenvironment. In vivo delivery solves manufacturing, not biology. What it may add is the ability to redose, and the option of programming cell types other than T cells. Early solid tumour work has tended to use mRNA-LNP to generate transient CAR-expressing effector cells, and NK cell therapy and in vivo CAR-NK approaches are being explored on the same logic. Our wider view of where these modalities are heading is in our analysis of immuno-oncology biomarker trends.

Biomarker implications: what changes when the product is made in the patient

With ex vivo CAR-T, the infused product is characterised before it reaches the patient. You know the CAR-positive fraction, the vector copy number per cell, the phenotype and the potency. With in vivo CAR-T none of that exists. The vector dose is known; the cell product is not. Everything that would normally be a release specification becomes a clinical measurement.

1. In vivo transduction and CAR expression

This is the first question any in vivo programme has to answer: did the vector reach T cells, and how many are now CAR-positive? Two orthogonal assays carry the load.

• Flow cytometry for CAR surface expression on T cells, ideally with subset resolution (CD4/CD8, memory phenotype), because the therapeutic cell is generated rather than selected. Programmes targeting a specific subset, as Capstan’s does for CD8+ T cells, need to show that the subset restriction held in humans.
• qPCR or ddPCR for vector copies in genomic DNA from peripheral blood, reported as copies per microgram of genomic DNA or copies per cell. In the LB2501 lymphoma study, CAR sequences were detectable by qPCR in all dose level 2 patients, with a median peak of around 109,000 copies per microgram of genomic DNA reached at a median of 15 days.

The two readouts answer different questions. Vector copy number tells you how much gene transfer occurred; flow tells you how many functional CAR-expressing cells resulted. For LNP-mRNA products there is no integrated copy to count, so CAR protein by flow, plus mRNA quantification in the relevant window, becomes the primary measure, and sampling has to be earlier and denser because expression is transient.

2. Pharmacokinetics of CAR-T expansion

In vivo CAR-T turns a cell product into something that behaves like a drug with a PK curve. Teams are reporting the familiar parameters, derived from vector copy number or CAR+ cell counts over time: C_max, T_max, AUC and persistence. The ESO-T01 phase 1 listed pharmacokinetics and pharmacodynamics as secondary endpoints alongside efficacy.

Three points matter for study design:

• Expansion kinetics resemble ex vivo products despite no lymphodepletion. Peak expansion in the lentiviral studies has landed around two weeks, and Kelonia reported CAR-T cells reaching a large fraction of circulating T cells. Sampling schedules built for a lymphodepleted ex vivo product are roughly the right shape, but that should not be assumed.
• Exposure, not dose, is what correlates with outcome. The vector dose administered is several steps removed from the number of CAR-T cells generated. Dose-exposure-response modelling is therefore central rather than exploratory, and it is the basis for dose selection.
• The dose level 1 result in the LB2501 study is instructive. No responses at the lower dose, with peak vector copies roughly a hundred-fold lower than at dose level 2, which is exactly the kind of exposure threshold these assays exist to find.

3. B-cell aplasia as a pharmacodynamic marker

For every CD19, CD20 and BCMA programme, depletion of the target cell compartment is the most direct evidence that the CAR worked. Peripheral B-cell counts by flow cytometry, with immunoglobulin levels, are cheap, standardised and available in any clinical laboratory, which makes them the natural primary pharmacodynamic endpoint, especially in autoimmune trials where tumour response is not the readout.

Interpretation differs by platform. With an integrating lentiviral vector, durable aplasia is expected and its resolution signals loss of functional CAR-T cells. With transient LNP-mRNA, B-cell return is anticipated, and the interesting question is what comes back: the Science data in non-human primates showed repopulation with predominantly naive B cells, which is the mechanistic basis for the immune reset argument. Teams should therefore characterise the returning B-cell repertoire, not just count cells. For BCMA programmes the equivalent markers are plasma cell depletion, serum free light chains, paraprotein and MRD by flow or next-generation sequencing.

4. Cytokine release and ICANS biomarkers

In vivo CAR-T introduces two overlapping sources of acute inflammation: CAR-T activation, and the innate response to the vector or the lipid nanoparticle itself. Infusion-related reactions were common in the LB2501 study, which is a vector effect rather than a CAR effect, and distinguishing the two matters clinically.

The core panel is unchanged from ex vivo practice: serial IL-6, IFN-gamma, TNF-alpha, IL-10, CRP and ferritin, with CRS and ICANS graded by ASTCT criteria. What changes is timing and attribution. Reactions in the first hours after infusion point to the delivery vehicle; the classic CRS window tracks peak CAR-T expansion at one to two weeks. Correlating cytokine peaks with the vector copy number curve in the same patients is the only way to tell those apart, and it is also how a programme builds the evidence that a lower dose can preserve efficacy while reducing toxicity. Given the class-level risk of immune effector cell-associated haemophagocytic syndrome seen with ex vivo products, ferritin, triglycerides and liver enzymes belong in the routine panel. Our review of CAR-T cell therapy biomarkers covers the evidence behind these markers in more detail.

5. Off-target transduction monitoring

This is the biomarker question that is genuinely new. An intravenous vector circulates, and nothing guarantees it only enters the intended cells. Programmes need data on transduction of:

• other immune cells (B cells, NK cells, monocytes) and haematopoietic stem and progenitor cells;
• liver, the default destination for both lipid nanoparticles and systemic viral vectors;
• the malignant cells themselves. CAR transduction of a leukaemic B cell has been shown to mask CD19 and cause resistance in an ex vivo setting (Ruella et al., Nature Medicine, 2018), and that risk is structurally higher when the vector meets tumour cells inside the patient;
• gonadal tissue, for the germline transmission assessment regulators expect of integrating vectors.

Practically this means vector copy number measured in sorted cell populations rather than whole blood, plus integration site analysis for integrating vectors and testing for replication-competent lentivirus. Programmes should also assume questions about clonality over time, which is a sequencing exercise rather than a flow exercise.

6. Immunogenicity

Immunogenicity is a bigger issue for in vivo CAR-T than for ex vivo products, because the vector itself is administered to the patient. Three assays are worth building early:

• Anti-vector antibodies, including neutralising antibodies against the envelope or targeting domain. Several lentiviral programmes describe immune-shielded vectors precisely to manage this, and that is a claim which needs clinical confirmation.
• Anti-PEG and anti-LNP antibodies for lipid nanoparticle products, which matter most where redosing is part of the regimen.
• Anti-CAR responses, particularly against non-human or partially humanised binders, which can curtail persistence.

Pre-existing immunity also becomes a screening consideration. If a patient already has neutralising antibodies that block transduction, exposure will be low and the dose effectively wasted, which is both a safety and a trial-integrity problem.

7. Patient selection biomarkers

Patient selection for in vivo CAR-T has one requirement that ex vivo products do not: the patient must have enough fit T cells, in circulation, at the time of dosing, for the vector to find. That shifts attention to baseline measures of the T-cell compartment:

• absolute lymphocyte count and T-cell subset composition, including memory and exhaustion phenotype;
• target antigen expression on the tumour (BCMA, CD19, CD20), which is the usual predictive question;
• pre-existing anti-vector immunity;
• disease burden and tumour kinetics, which drive both response and toxicity risk.

Heavily pretreated patients, the population in most first-in-human studies, often have the most compromised T-cell compartment. A baseline T-cell signature that predicts in vivo transduction efficiency would be one of the more valuable biomarkers in this field, and no validated version of it exists yet. The methodological standards such a marker would have to meet are set out in our guide on validating an IO biomarker in a clinical trial.

8. How monitoring differs from ex vivo products

The operational consequence is that in vivo CAR-T trials are assay-heavy and time-sensitive. Dose escalation decisions depend on PK and pharmacodynamic data arriving quickly, which raises the bar for central laboratory partners, assay validation and sample logistics. For diagnostics, tools and CRO companies, that is the commercial opening: validated, transferable ddPCR and flow assays for in vivo CAR detection are infrastructure this field currently lacks.

Safety considerations

The safety profile of in vivo CAR-T has three layers.

Shared with ex vivo CAR-T. CRS, ICANS, cytopenias, infection and hypogammaglobulinaemia all occur. The ESO-T01 phase 1 is a useful corrective to the idea that skipping lymphodepletion removes toxicity: every patient had a grade 3 or higher adverse event, and CRS was mostly grade 3. The FDA also requires a boxed warning for T cell malignancies across approved BCMA- and CD19-directed autologous CAR-T products, with lifelong monitoring for secondary malignancy. There is no reason to assume an in vivo route lowers that risk.

Specific to systemic vector delivery. Infusion-related and innate inflammatory reactions, hepatic exposure and transaminase elevations (seen in the ESO-T01 cohort), and off-target transduction. For integrating vectors there is also the insertional mutagenesis question and the need to exclude replication-competent virus.

Regulatory consequences. An integrating in vivo product is a gene therapy given systemically. FDA guidance on long term follow-up after administration of human gene therapy products points to extended observation, up to around 15 years, where a vector integrates into the genome, and FDA has held a dedicated workshop on integration site analysis during that follow-up. Non-integrating RNA products should be able to justify a shorter period. Either way, the biomarker plan is also a regulatory commitment: sponsors should expect to defend their choice of assays, their validation status and their sampling schedule, and to show how transduction specificity was demonstrated in humans rather than only in animals.

One more piece of context: in late August 2026 Novartis paused its autoimmune trials of rapcabtagene autoleucel, an autologous CD19 CAR-T, after three deaths linked to immune effector cell-associated haemophagocytic syndrome, and Bristol Myers Squibb paused its autoimmune programme for zolacabtagene autoleucel as a precaution after transient inflammatory events. Those are ex vivo products, but they set the safety bar that in vivo programmes in autoimmune disease, where patients are not facing imminent death from cancer, will be measured against.

What this means for biomarker and diagnostics teams

A few practical conclusions for teams planning work in this area:

• Write the PK and pharmacodynamic plan first. In the absence of a characterised product, PK and pharmacodynamic data are the dose-finding evidence. They should not be exploratory endpoints.
• Build two orthogonal transduction assays. ddPCR or qPCR for vector copies, and flow cytometry for CAR protein on defined T-cell subsets, validated and fit to run on short turnaround.
• Measure specificity in sorted populations. Whole-blood vector copy number cannot answer the off-target transduction question that regulators and reviewers will ask.
• Treat immunogenicity as a platform risk. Anti-vector, anti-PEG and anti-CAR assays affect both dosing strategy and patient selection.
• Plan the long tail. Integrating vectors bring long-term follow-up, integration site analysis and secondary malignancy surveillance, which are commitments measured in years.
• Expect companion diagnostic questions later. If a baseline T-cell or antigen expression marker starts to predict benefit, it will need the same regulatory path as any other predictive test. Our FDA companion diagnostic approval guide sets out that process.

Where these conversations continue in 2027

In vivo CAR-T sits exactly where immuno-oncology and biomarker science meet, which is the gap ImmunoMark Summit San Diego is built to close. On 10–11 February 2027, 300+ senior pharma, biotech and diagnostics leaders meet across two parallel streams, with Stream 1: Next-Gen Immuno-Oncology covering cell therapy and the next generation of IO modalities, and Stream 2 covering biomarkers and companion diagnostics. The format is keynotes, panels, case studies, pre-arranged 1:1 meetings and a Day 1 drinks reception.

The series continues at the immuno-oncology conference in London on 3–4 March 2027 and in Boston in June 2027. For the wider picture, see our list of the best immuno-oncology conferences to attend.

Register for ImmunoMark San Diego 2027

If your company wants to meet this audience directly, whether you supply cell therapy tools, immune monitoring assays or CDMO capacity, see the San Diego 2027 sponsorship packages.

FAQs

What is in vivo CAR-T and how does it differ from conventional CAR-T?

Conventional CAR-T collects a patient’s T cells by apheresis, engineers them in a manufacturing facility and reinfuses them after lymphodepleting chemotherapy. In vivo CAR-T gives the patient a single intravenous dose of a targeted vector, either a lentiviral vector or a lipid nanoparticle carrying CAR mRNA, which delivers the CAR gene to T cells inside the body. There is no apheresis, no individual manufacturing run and, in most trials so far, no lymphodepletion.

Has in vivo CAR-T worked in patients?

Yes, in small early studies. A phase 1 of the anti-BCMA lentiviral product ESO-T01 in Nature Medicine in 2026 reported objective responses in four of five patients with relapsed or refractory multiple myeloma, including three stringent complete remissions, with all evaluable responders MRD-negative by day 60. Kelonia reported MRD-negative responses in four patients treated with KLN-1010 at ASH 2025, and in vivo CD19 CAR-T has been reported in lupus and in neurologic autoimmune disorders in the New England Journal of Medicine. Cohorts are very small and follow-up is short.

What is the difference between lentiviral and LNP in vivo CAR-T for monitoring?

Lentiviral vectors integrate the CAR gene, so CAR expression is stable and vector copy number by qPCR or ddPCR is the central pharmacokinetic measure, alongside integration site analysis, replication-competent virus testing and long-term follow-up. LNP-mRNA products express the CAR transiently with no integration, so monitoring relies on CAR protein by flow cytometry and mRNA quantification in a much narrower window, with immunogenicity against the nanoparticle mattering more because redosing is likely.

Which biomarkers matter most for in vivo CAR-T programmes?

In vivo transduction and CAR expression (flow cytometry plus ddPCR or qPCR for vector copies), the pharmacokinetics of CAR-T expansion, B-cell aplasia and immunoglobulin levels as pharmacodynamic markers, cytokine panels for CRS and ICANS, vector copy number in sorted non-target cells for off-target transduction, immunogenicity against the vector and the CAR, and baseline T-cell fitness and target antigen expression for patient selection.

Will in vivo CAR-T work in solid tumours?

It is not proven. Clinical programmes so far target CD19, CD20 and BCMA in blood cancers and autoimmune disease. In vivo delivery removes the manufacturing barrier but not the biological ones that limit cell therapy in solid tumours: antigen heterogeneity, poor trafficking into tumour and an immunosuppressive microenvironment. The ability to redose, and to programme cell types other than T cells, may help, but solid tumour data remain early.