Why Cell And Gene Therapies Break After Early Success

By Wei Zhu, Ph.D.

Cell and gene therapy has entered a new phase. The field has largely solved how to trigger biology. We can activate immune cells, deliver genes into target tissues, and induce measurable biological responses in patients. And yet, many programs still fail. Not because nothing happens, but because what happens cannot be sustained, controlled, or directed over time. Across modalities, the limiting factor is no longer whether we can generate activity. It is whether we can govern that activity once it begins. Without that layer of control, early success becomes unstable — and failure becomes a recurring pattern rather than an exception.

The Pattern No One Wants To Name

Across different companies, targets, and technologies, a familiar trajectory keeps appearing: strong preclinical data and clear early clinical signals followed by unexpected toxicity, relapse after initial response, loss of persistence, or failure to meet primary endpoints. This pattern has appeared in CAR-T therapies in solid tumors, AAV-based gene therapies in late-stage trials, and cell-based therapies with early responses that do not hold.

For example, multiple CAR-T programs targeting antigens, such as mesothelin or GD2, in solid tumors have shown early biological activity but limited durable efficacy in clinical studies. Additionally, late-stage AAV gene therapy programs, including Duchenne muscular dystrophy efforts, have struggled to translate expression into consistent clinical benefit across patients.

These are not isolated failures. They are recurrent behaviors of the system.

The Industry’s Default Explanation — And Its Limits

When programs fail, explanations tend to cluster around familiar categories:

• The target was wrong.
• The vector wasn’t efficient enough.
• Potency was insufficient.
• Cells became exhausted.

These explanations are not incorrect, but they are incomplete. They describe what happened, but they do not explain why similar failure patterns recur across entirely different biological systems. If different modalities, targets, and delivery systems all converge toward similar outcomes, the problem may not lie in any single component. It may lie in how the system behaves over time.

What Is Actually Breaking

When we step back and look across programs, three recurring failure behaviors emerge.

Activation Without Context

Cells or genes act where they should not. Engineered immune cells recognize low-level antigen expression in healthy tissues and gene expression occurs outside intended spatial or physiological boundaries.

This results in toxicity, off-target effects, and narrow therapeutic windows.

Action Without Preparation

The therapy enters a biological environment that is not permissive, which causes immune-suppressive tumor microenvironments, diseased or stressed cells unable to utilize delivered genes, and lack of supporting signals for function or expansion, resulting in a lack of efficacy despite successful delivery.

Response Without Stability

Initial success cannot be maintained. For example, CAR-T cells lose function or persistence, systems revert to prior biological states, or adaptive resistance emerges. This can cause relapse, transient benefit, or diminishing returns.

These are often treated as independent issues. They are not. They reflect a common underlying problem: We can initiate biological action, but we cannot reliably control its trajectory.

The Shift: From Building Tools To Governing Behavior

For the past decade, the field has focused on improving tools, such as more efficient viral vectors, stronger promoters, more potent effector cells, and more precise gene editing.

These advances have been transformative. But they share an implicit assumption: If we improve the tool, the outcome will improve. In many cases, that assumption no longer holds because biology does not fail primarily due to insufficient power. It fails because power is applied without sufficient control. The central question is shifting from “Can we make this intervention work?” to “Under what conditions should this intervention occur and how is that controlled?”

A Different Way To Think About Therapy Design

A useful reframing is to stop viewing therapies as static interventions and instead view them as dynamic processes unfolding over time.

This introduces a different set of design questions:

• What state is the system in before intervention?
• What transition is being induced?
• What conditions must be met for that transition to succeed?
• What stabilizes the new state once achieved?
• What prevents the system from reverting?

This shift changes how decisions are made.

Traditional framing asks:

• Does the therapy activate?
• Does it kill target cells?
• Does it persist?
• Is delivery efficient?

Alternative framing asks:

• Under what conditions does it activate?
• Does it act only in the correct context?
• Does it stabilize a beneficial state?
• Is the system ready to respond to delivery?

This is not a theoretical distinction. It directly impacts trial design, patient selection, combination strategies, and safety management.

Why This Matters Now (Not Later)

This shift is no longer optional. Several converging signals suggest the field is reaching a structural limit: increasing safety concerns as more potent therapies enter the clinic, plateauing efficacy in solid tumors despite improved engineering, high inter-patient variability even with similar interventions, and costly late-stage failures, particularly in gene therapy programs.

The suspension of CAR-T trials due to severe neurotoxicity highlighted how powerful immune activation can exceed controllable limits. In gene therapy, variability in patient response — despite similar vector delivery — underscores that delivery alone does not determine outcome.

These signals all point in the same direction: Intervention capability is scaling faster than control capability. And that gap is becoming the dominant source of risk.

Implications For Drug Developers And Investors

If this framing is correct, it has several practical implications.

For drug developers:

• Early design decisions should consider biological context, not just target and modality.
• Preclinical models should assess state dependence, not just activity.
• Combination strategies should focus on preparing systems, not just amplifying effect.

For clinical strategy:

• Patient selection may need to shift from static biomarkers to state-dependent criteria.
• Timing of intervention may become as important as the intervention itself.
• Durable outcomes may depend more on stabilization mechanisms than initial potency.

For investors:

• Programs should be evaluated not only on what they do, but on how their effects are governed over time.
• Apparent early efficacy may not translate unless the system can sustain it.
• Competitive advantage may emerge from control architectures, not just new tools.

Closing

The industry has spent a decade learning how to trigger biology. That was necessary, but it is no longer sufficient. The next phase will be defined by whether we can coordinate biological actions across time, constrain them to the correct context, and stabilize the outcomes they create.

Without that, success will remain fragile, failures will continue to repeat, and progress will appear real but not durable. The question is no longer whether we can intervene, but rather whether we can govern what happens after we do.

About The Author

Wei Zhu, Ph.D., works on the operating systems behind how next-generation cell and gene therapy (CGT) will actually function. Her focus is at the architecture layer, where immune states, logic layers, and multimodal interventions are orchestrated into stable therapeutic attractors. Zhu helps organizations move beyond pipeline and platform thinking and design the systems that will define the next three to five years of CGT innovation. As an independent strategist, she works at the intersection of upstream innovation, system architecture, and multi-modality design in CGT. Her recent work centers on building the State-Logic Operating System (SLOS) — a next-generation framework that upgrades CGT development.