Chapter 26: Peptide Vaccines — Cancer Immunotherapy and Infectious Disease Prevention

"In the fields of observation, chance favors only the prepared mind." — Louis Pasteur, lecture at the University of Lille (1854)

Overview

Every chapter before this one has followed a peptide into the body and asked what it binds. This chapter asks a different question, and the difference is the whole point: here the peptide is not the drug. It is the message.

The immune system does not recognize a virus, or a tumor, as such. What a T cell recognizes is a short fragment of a protein — a peptide — displayed on the surface of a cell. That is not a teaching simplification; it is the physical mechanism, and it means that at the level where adaptive immunity actually operates, the message is a peptide. A vaccine's job is to put the right peptide in front of the right cells, in a context alarming enough that the system bothers to remember it.

Which reframes vaccination as peptide pharmacology, and is why this chapter is in this book. It is also why the chapter will feel strange after twenty-five chapters of receptors and half-lives. A vaccine has a half-life, and the half-life is irrelevant: the molecule is gone in hours and the effect is meant to last years. You need a different mental model, and building it is most of the work here.

This chapter covers cancer treatment. Some of you are reading it as a patient, or as the person who drives someone to their appointments, and I want to be exactly as accurate as I can manage in both directions. There is real progress here — individualized neoantigen vaccination is one of the more genuinely interesting ideas to reach oncology in thirty years, and it is now in properly randomized trials. And as of this writing in 2026, no individualized cancer vaccine has general approval anywhere in the world. Both sentences are load-bearing. Overstating the first would be a cruelty; understating it would be a different kind of dishonesty.

Chapter 25 covered antimicrobial peptides — innate immunity: hard-coded, fast, forgetful. This chapter is the other half: slow, learned, specific, durable. Adaptive immunity's language is peptides.

In this chapter, you will learn to:

• Explain what a vaccine teaches, and why "memory" and not "occupancy" is the right unit of effect
• Define an epitope, and say why T cells read peptides while antibodies read shapes
• Describe how MHC/HLA molecules display peptides, and why the enormous variability of HLA between people is the central obstacle for peptide vaccines
• Explain why an adjuvant is not an optional extra but frequently the entire difference between a response and nothing
• Trace an individualized neoantigen vaccine from tumor biopsy to injection, and name the failure point at every step
• Distinguish an mRNA vaccine from a peptide drug without losing sight of the fact that its message is still a peptide sequence
• State honestly where the evidence stands in 2026, and name the specific results that would move it

Learning Paths

Written for the 🔬 Science and 🏥 Clinical paths, but §26.1–26.3 are the most useful immunology in the book for everybody.

💊 GLP-1 — optional. If you read one section, read §26.4: it explains why "the same molecule" delivered differently is not the same intervention. 🏋️ Performance — optional. §26.4 and §26.8 together explain why the immune system's response to an injected peptide is not always the response you wanted (Chapter 19). 🔬 Science — read in full. §26.2 and §26.3 are the mechanistic core; §26.5 is the payoff. 💄 Cosmetic — skip, except §26.2 if you are curious why "peptide signaling" means something completely different in immunology than in Chapter 30. 🏥 Clinical — read in full, and read §26.7 and §26.8 twice. Patients ask about cancer vaccines, and the honest answer requires holding two things at once.

26.1 What a vaccine actually teaches the immune system

Start with a fact that sounds like a technicality and is not: a vaccine is not a drug in the sense this book has used the word for twenty-five chapters.

A peptide drug works by occupancy. Semaglutide sits on the GLP-1 receptor; while it is there, the receptor signals; when it is cleared, the effect ends. That is why Chapter 3 spent so long on half-life and why the toolkit in Chapter 33 is aimed at keeping molecules around longer. Exposure produces effect; remove exposure and you remove effect.

A vaccine inverts that relationship. The injected material is cleared within days, and the intended effect begins after it is gone and is supposed to persist for years. The vaccine is a rehearsal. Your own immune system does the work, on its own schedule, possibly years later, in response to something the vaccine never touched.

So what is actually changing?

Two arms, two kinds of memory

The adaptive immune system runs two parallel programs.

B cells make antibodies. An antibody binds its target directly, out in the open — in blood, in mucus, in tissue fluid. That is how you neutralize a virus particle before it enters a cell, or tag a bacterium for destruction. What an antibody binds is a surface: a three-dimensional patch on an intact molecule.

T cells do something different and, for this chapter, more important. A T cell cannot see anything free-floating. Its receptor engages only a peptide being held up by a specialized display molecule on another cell's surface. Cytotoxic T cells (CD8) kill cells displaying something wrong — a viral fragment, a mutated fragment. Helper T cells (CD4) do not kill; they license and coordinate, supplying the signals B cells need to make good antibodies and CD8 cells need to become durable memory rather than a brief flare.

Before you are vaccinated against anything you already carry a vast, mostly idle repertoire — millions of distinct T-cell and B-cell clones, each generated by randomized genetic recombination, each specific for something you will probably never meet. Some tiny number happen to be specific for what is in the vaccine. Vaccination finds them, wakes them, multiplies them, and leaves behind a memory population larger, faster, and better positioned than what you started with.

TWO WAYS A MOLECULE CAN MATTER

  A PEPTIDE DRUG — occupancy                A VACCINE — instruction

  effect                                    effect
   │  ╭──────╮                               │                    ╭──────────────
   │ ╱        ╲                              │                   ╱
   │╱          ╲___                          │                  ╱
   └──────────────────► time                 └─────────────────╱──────────► time
    dose            cleared                   dose      cleared      challenge

  Effect tracks exposure.                   Effect begins after clearance and is
  Stop dosing, effect stops.                triggered by something else entirely,
  Half-life is the design problem.          possibly years later.
                                            DURABILITY OF MEMORY is the design
                                            problem. Half-life is nearly irrelevant.

  Unit of effect: receptor occupancy.       Unit of effect: a changed repertoire.
  Measured in concentration.                Measured in cells, clones, and time.

Almost every confusion about vaccines comes from applying the left-hand model to a right-hand object. "How long does the vaccine stay in your body?" is a left-panel question. It has an answer (days), and the answer tells you nothing you wanted to know. How long the memory lasts is the question, and that is a property of your cells, not of the injected material.

What a vaccine cannot do by wanting to

Vaccination can only expand clones that already exist. If your repertoire contains no T cell capable of recognizing a given peptide, no amount of vaccinating with that peptide will produce one. And the repertoire is not random with respect to your own body: during development, T cells that recognize your own proteins too strongly are deleted or restrained. This is tolerance, and it is why you do not spend your life attacking yourself.

It is also why therapeutic cancer vaccines failed for thirty years. A tumor is made of your cells, and most of what it displays is your proteins. Vaccinating against them asks the immune system to do the one thing it was carefully built not to do — using a repertoire from which the best-qualified clones were already removed. Hold that thought; §26.5 is the escape hatch.

🔍 Check Your Understanding

1. A peptide drug and a peptide vaccine may both be cleared from the body within a day. Why does that fact matter for one and not the other?
2. What does it mean to say a vaccine's unit of effect is "a changed repertoire" rather than "receptor occupancy"?
3. Chapter 25 described antimicrobial peptides as fast, broad, and without memory. Name the two properties adaptive immunity trades away to get memory and specificity.

26.2 Epitopes: the peptide is the message

Now the central mechanical fact of the chapter.

An antigen is anything the immune system can recognize — usually a whole protein. An epitope is the specific small part that a receptor actually contacts. A protein of 400 amino acids is one antigen and contains many potential epitopes, most of which will never be used.

The two arms of the system define epitopes differently, and the difference is not trivia.

B cells and antibodies read shapes. An antibody binds the intact, folded surface of a molecule. The residues it contacts are usually scattered along the sequence and brought together by folding — a conformational epitope. Unfold the protein and the epitope disappears, because it was never a stretch of sequence; it was a patch of three-dimensional surface. The great majority of antibody epitopes on natural proteins are conformational.

T cells read sequences. A T-cell receptor engages a linear epitope: a short contiguous run of amino acids, detached from the rest of the protein and held in a groove on a display molecule. Folding is irrelevant, because by the time a T cell sees it the protein has been chopped up.

That asymmetry organizes the entire field. A short synthetic peptide is a natural way to deliver a T-cell epitope and an unnatural way to deliver a B-cell epitope. If your goal is neutralizing antibodies against a virus, a short peptide is usually the wrong tool: the antibodies it raises recognize the floppy peptide and often fail to recognize the folded protein on the actual virus. If your goal is a cytotoxic T-cell response against an infected or malignant cell, a peptide is exactly right, because a peptide is precisely what the T cell was built to see. Remember that at §26.8, when you wonder why peptide vaccines have made more headway in cancer therapy — a T-cell problem — than in infectious disease prevention, which is mostly an antibody problem.

How a peptide gets onto a cell surface

Inside every nucleated cell, a fraction of newly made and defective proteins is fed into the proteasome, a barrel-shaped shredder that cuts them into short fragments. A transporter pumps some of those into the endoplasmic reticulum, where they are loaded into the groove of a class I display molecule, and the loaded complex travels to the cell surface. There it sits, holding up an 8-to-10-residue sample of the cell's internal protein production, waiting for a passing T cell.

A healthy cell displays thousands of such complexes showing normal self peptides, and CD8 T cells walk past. An infected cell displays viral peptides among them. A cancer cell may display peptides from mutated proteins. That display is the only way a killer T cell can know what is happening inside a cell it cannot enter, and it is why the mechanism exists.

The second route handles external material. Dendritic cells, macrophages, and B cells engulf things from outside, degrade them in acidified compartments, and load the resulting fragments — longer, 13 to 25 residues, because this groove is open at both ends — onto class II molecules for CD4 helper T cells.

THE PEPTIDE IS THE MESSAGE — two display routes

  CLASS I: "here is what I contain"          CLASS II: "here is what I found"

   protein made inside the cell               protein taken in from outside
            │                                          │
        proteasome  (shred)                     endosome (acidify, digest)
            │                                          │
     8–10 residue peptide                     13–25 residue peptide
            │                                          │
     loaded in the ER                          loaded in the vesicle
            │                                          │
   ┌────────▼────────┐                       ┌─────────▼────────┐
   │  class I + pep  │  ───► CD8 T cell      │ class II + pep   │ ───► CD4 T cell
   └─────────────────┘       (kills the      └──────────────────┘      (helps, licenses,
     on nearly EVERY          displaying       only on professional     coordinates,
     nucleated cell           cell)            presenting cells         sustains memory)

  Either way, the thing the T-cell receptor actually touches is a PEPTIDE.
  There is no route by which a T cell perceives a whole pathogen, or a tumor,
  or an organ. It perceives fragments, in grooves, on surfaces.

The chapter's thesis follows without further argument. If the only thing a T cell can perceive is a peptide in a groove, then every T-cell-directed vaccine ever made — live, attenuated, inactivated, subunit, viral-vectored, mRNA — is ultimately a peptide-delivery system. They differ in how the peptide arrives and what comes with it, not in what the immune system ends up reading.

🧬 The Molecule — what an epitope looks like up close

A class I epitope is typically eight to ten residues — smaller than every therapeutic peptide in this book: shorter than oxytocin, a fraction of BPC-157's fifteen, a seventh of semaglutide.

It sits in a closed groove, pinned at both ends. Two or three of its side chains — the anchor residues, usually near position 2 and at the C-terminus — drop into pockets in the groove and hold the peptide in place. The remaining side chains point upward, and those are the ones the T-cell receptor reads.

This is Chapter 1 §1.2 doing real work. Whether a peptide can be displayed at all is decided by whether its anchor side chains fit the pockets of your particular display molecule — greasy into a greasy pocket, charged into a complementary one. Different display molecules have differently shaped pockets and therefore prefer different anchors.

Two consequences follow. First, a single amino acid substitution can abolish presentation by destroying an anchor, or create it by supplying one. Second — §26.3's whole subject — which peptides you can display is determined by which display molecules you inherited. The groove does the selecting, and no two people have quite the same set of grooves.

📊 Evidence Rating

Claim: Adaptive immune recognition operates by T-cell receptors engaging short peptide epitopes presented on MHC/HLA molecules. Rating: ✅ (mechanistic; settled immunology, 2026) Reason: Established across five decades of converging structural, genetic, and functional work, including solved crystal structures of peptide-loaded MHC molecules and receptor complexes, and recognized with a Nobel Prize in Physiology or Medicine in 1996 for the discovery of MHC-restricted recognition; no serious contemporary dispute exists about the mechanism. What would change it: essentially nothing short of a revolution in immunology; this is as settled as anything in this book.

And now the discipline. Chapter 2 §2.9 made a rule this chapter leans on heavily: a mechanistic ✅ licenses nothing clinical. It is fully established that T cells recognize peptides. It does not follow that injecting a peptide produces a useful T-cell response, or that such a response controls a tumor, or that controlling a tumor extends anyone's life. Each is a separate empirical claim needing its own evidence, and this chapter rates them separately and much less generously.

26.3 MHC and HLA: why a peptide vaccine can be personal, and why it has to be

The display molecules have a name: the major histocompatibility complex (MHC), called human leukocyte antigens (HLA) in humans for historical reasons involving transplant rejection — they were discovered as what makes one person's tissue foreign to another, long before anyone knew their job.

The nomenclature is worth fixing once. HLA-A, HLA-B, and HLA-C are the classical class I genes: their products appear on essentially every nucleated cell and present internally derived peptides to CD8 T cells. HLA-DR, HLA-DQ, and HLA-DP are the classical class II genes, appearing mainly on professional antigen-presenting cells and presenting externally derived peptides to CD4 T cells. You inherit one set from each parent, so you express up to six classical class I molecules and a comparable handful of class II. That handful is your entire vocabulary. Any peptide that does not fit one of those grooves is, to your T cells, invisible.

The most variable genes you have

The HLA genes are the most polymorphic in the human genome. Not somewhat variable — extraordinarily variable, with tens of thousands of distinct alleles cataloged across human populations and more described every year. Many of the differences sit precisely in the groove, changing the shape and chemistry of the pockets that grip the anchor residues.

Which means two people, given the same protein, will display different peptides from it. Not different amounts of the same peptides. Different peptides. Your immune system and mine, shown an identical virus, read different excerpts of the same book.

There is a good evolutionary reason: a population whose members all displayed the same peptides could be swept by a single pathogen that learned to avoid them, and HLA diversity is a species-level hedge against exactly that. It is magnificent design. It is also, for anyone making a peptide vaccine, an enormous problem.

WHY THE SAME PEPTIDE IS A VACCINE IN ONE PERSON AND NOTHING IN ANOTHER

  the same 9-residue peptide, offered to two people

  PERSON A                                 PERSON B
  inherited HLA groove: deep hydrophobic   inherited HLA groove: shallow, charged
  pocket at position 2                     pocket at position 2

     peptide anchors fit ──► displayed        peptide anchors DO NOT fit
     on the cell surface                      ──► never displayed
              │                                        │
     CD8 T cells can see it                   No display. No recognition.
              │                                No response. Not a weak response —
     response possible                         no response, from a mechanism that
     (still not guaranteed)                    never engaged at all.

  A whole-protein or whole-organism vaccine sidesteps this: it supplies dozens of
  candidate peptides and lets each person's own grooves choose whichever ones fit.
  A short-peptide vaccine has made that choice in advance, at the factory, for
  everyone.

This is HLA restriction, and it is not a subtlety at the margins. It is why a great many peptide cancer vaccine trials in the 1990s and 2000s enrolled only patients carrying one particular common class I allele. That decision had two consequences, both still with us.

First, it caps the addressable population. Even the most common class I alleles are carried by a minority of people worldwide. A vaccine built around one is, by construction, a therapy for a fraction of patients — and the fraction shrinks again once the tumor must also express the target.

Second, it skewed who got studied. The alleles chosen were typically those most common in the populations where the trials ran and the reference data were richest, which in practice meant patients of European ancestry were far more likely to be eligible. That is an equity problem in its own right and a scientific one: a therapy validated in a narrow HLA background has an unproven claim on everyone else. Individualized approaches (§26.5) improve on this in principle, selecting epitopes against your HLA type rather than a designated common one — though whether the underlying reference data and prediction algorithms perform equally well across all HLA backgrounds is a separate question, and they were trained on unevenly distributed data.

Presentation is necessary, not sufficient

One more thing, because it is the most common overclaim in the subfield. Predicting that a peptide will bind your HLA is not predicting that it will produce an immune response. Between binding and response sit at least four additional requirements:

• The peptide must actually be generated and transported — plenty of well-binding peptides are never produced from the parent protein at all.
• It must reach sufficient density on the cell surface to be noticed.
• Your repertoire must contain a T cell that recognizes that peptide-HLA combination, and that clone must not have been deleted by tolerance.
• That T cell must be activated rather than tolerized — §26.4's entire subject.

Contemporary binding-prediction algorithms are genuinely good. Immunogenicity prediction — will this raise a response in this person — is substantially worse, and in practice a minority of predicted epitopes elicit a measurable T-cell response when tested. That gap between "predicted to bind" and "raises a response" is where a great deal of optimistic arithmetic in this field goes to die. Be alert to any account that quietly treats the two as the same thing.

26.4 Adjuvants: the necessary accomplice

Put a short synthetic peptide in saline, inject it, and in most cases nothing happens.

Not "a weak response happens." Frequently nothing, and sometimes worse than nothing.

A short peptide is, from the immune system's point of view, deeply unimpressive. It is small. It is cleared in minutes to hours by exactly the proteases Chapter 1 §1.3 introduced. It carries no feature marking it as dangerous. And the immune system does not respond to foreignness alone — if it did, you would be attacking your food and your gut bacteria continuously. It responds to foreignness accompanied by evidence of danger.

Antigen-presenting cells carry pattern recognition receptors — Toll-like receptors and their relatives — that detect molecular signatures of pathogens and tissue damage: bacterial cell wall components, double-stranded RNA, unmethylated bacterial DNA motifs. When those receptors fire, the presenting cell matures: it migrates to the draining lymph node, upregulates costimulatory molecules, and produces the cytokines that shape what kind of response follows.

A T cell that meets its peptide with costimulation becomes activated and eventually forms memory. A T cell that meets its peptide without costimulation typically becomes anergic — switched off for that antigen — or is deleted, or becomes a regulatory cell that actively suppresses responses to it. You can vaccinate someone into unresponsiveness, and in the cancer setting that is not hypothetical; it is one plausible explanation for trials in which vaccinated patients, if anything, did slightly worse.

So: the adjuvant is not an additive. For a peptide vaccine it is frequently the entire difference between a response and nothing, and adjuvant selection is a major determinant of whether a program succeeds. An immunologist once called adjuvants the field's dirty little secret — the unglamorous ingredient that made everything else work while nobody could explain how. The mechanism is much better understood now. The dependence has not diminished.

What adjuvants do, in categories

You do not need to memorize adjuvants. You do need to know they are not interchangeable, because different ones produce different kinds of immunity.

Two design points deserve emphasis because they get skipped.

Location matters more than dose. The response is generated in the lymph node draining the injection site, not in the muscle or the fat. Antigen and danger signal have to arrive there together and reach the same presenting cells. Much contemporary vaccine engineering is really lymph-node delivery engineering — molecules that hitch a ride on serum albumin into the node, or antigen packaged in particles the right size to drain there. Most of this work is preclinical or early-phase.

A depot can backfire. The intuition that keeping antigen around longer produces more immunity is unreliable. In animal studies, persistent water-in-oil depots have retained the very T cells they induced at the injection site, where those cells were deleted rather than dispatched to the tumor. That is a preclinical finding, in mice, and I will not overstate it — but it is a well-characterized mechanism, it plausibly contributed to human trial failures, and it illustrates that with adjuvants, "more" and "longer" are not synonyms for "better."

🩺 Safety and Risk — the adjuvant is where much of the risk lives

A sore arm, a low fever, and a day of feeling wrung out is usually not a side effect in the incidental sense. That is the danger signal working — a local innate immune response, which is the thing that converts an injected antigen into durable memory. Reactogenicity and immunogenicity are linked, which is why the most effective adjuvants are often the least pleasant.

Three real cautions.

The safety bar depends entirely on the setting. A prophylactic vaccine given to millions of healthy people can tolerate almost no serious risk. A therapeutic vaccine given to a person with metastatic disease is weighed against a very different alternative. A level of reactogenicity that would end a prophylactic program may be acceptable in an oncology one. §26.8 develops this.

Rare serious events are real and should be described accurately. The clearest recent example is myocarditis following mRNA vaccination against COVID-19: low rates, concentrated in adolescent and young adult males, usually mild and self-limited, and detected precisely because prophylactic vaccine safety surveillance is intensive. Both denying it and inflating it are failures of the same kind.

Targeting self-antigens can produce autoimmunity, and sometimes does. Immunotherapies aimed at pigment-cell antigens in melanoma have produced vitiligo — loss of skin pigment — because the response cannot distinguish melanocytes in a tumor from those in healthy skin. That on-target, off-tumor toxicity is one of the arguments for the neoantigen strategy in §26.5, which targets sequences existing nowhere in healthy tissue.

None of this is a basis for a personal decision. Vaccine decisions, and especially decisions about entering an oncology trial, belong with a clinician who knows your history (Chapter 39).

26.5 Neoantigens and individualized cancer vaccines

This is the most striking material in the chapter, and it starts from the problem §26.1 left hanging.

The tolerance trap

For most of the history of cancer immunotherapy, the targets were tumor-associated antigens: proteins tumors express abundantly, or express when they should not — antigens normally restricted to germ cells, differentiation antigens shared with the healthy tissue of origin, enzymes reactivated in malignancy, overexpressed growth factor receptors.

Every one of those is your own protein, and your immune system spent your entire development learning not to attack your own proteins. T cells with high-affinity receptors for self peptides are deleted in the thymus — central tolerance — and those that escape are held in check peripherally. So a vaccine against a tumor-associated antigen is trying to raise an army from the recruits who were specifically rejected, using the ones that got through because they were not very good at the job.

That is the structural reason therapeutic cancer vaccines have such a bad track record. The trials were not failing because the immunologists were careless. They were failing because they were fighting the immune system's most carefully engineered feature.

The escape hatch

Tumors accumulate mutations; that is what makes them tumors. Most are meaningless passengers, but some change a protein's amino acid sequence, and when that protein is processed and displayed, some of the resulting peptides contain the altered residue.

Those peptides exist nowhere in a healthy body. They were not in the thymus during development. No tolerance was ever established against them. They are, in the most literal immunological sense, foreign — as foreign as a viral peptide, and generated by your own cell.

These are neoantigens, and their significance is hard to overstate. The high-affinity repertoire against them is intact, because no central tolerance was ever imposed. They are tumor-specific, so a response against one cannot damage healthy tissue lacking the mutation — a fundamentally better safety proposition than targeting a shared self protein. And they are, in most cases, unique to one patient's tumor, which is the whole difficulty and the whole idea.

There is strong indirect human evidence that neoantigen recognition already does real work in people. Across many tumor types, response to checkpoint inhibitor therapy is associated with high tumor mutational burden and with mismatch-repair-deficient tumors carrying very large numbers of mutations. The natural reading is that checkpoint blockade works partly by releasing brakes on pre-existing responses against neoantigens — the immune system had already found the mutations and been suppressed. That is an association rather than proof, and mutational burden is an imperfect predictor for any individual patient. But it is real and reproducible, and it is the strongest argument that this idea is not merely elegant.

The pipeline

Here is the actual sequence of operations, remarkable as engineering regardless of how the clinical story ends.

INDIVIDUALIZED NEOANTIGEN VACCINE — from biopsy to injection

  1  SAMPLE          tumor tissue + a normal sample (usually blood)
                     └─ failure point: one biopsy may not represent a
                        heterogeneous tumor

  2  SEQUENCE        exome (and usually RNA) from both; compare
                     └─ the normal sample is what makes a mutation
                        provably somatic rather than inherited

  3  CALL MUTATIONS  identify sequence changes present only in the tumor
                     └─ failure point: subclonal mutations present in only
                        part of the tumor are escape routes

  4  TYPE THE HLA    determine this patient's own class I and II alleles
                     └─ from the same sequencing data

  5  PREDICT         for each mutation, generate candidate peptides and
                     score: will it be processed? will it bind THIS
                     patient's HLA? is the gene expressed? is the mutation
                     clonal? is it likely immunogenic?
                     └─ failure point: binding prediction is good;
                        immunogenicity prediction is much weaker

  6  SELECT          rank and choose a set (programs have used roughly
                     ten to a few dozen targets per patient)

  7  MANUFACTURE     synthesize the peptides, OR synthesize one mRNA
                     encoding them end-to-end as a single construct
                     └─ failure point: weeks of turnaround, against a
                        disease that does not wait

  8  RELEASE         identity, purity, sterility, potency testing —
                     for a product with a batch size of one patient

  9  ADMINISTER      with an adjuvant, and usually alongside a checkpoint
                     inhibitor to release the brakes on whatever responds

 10  MEASURE         immune readouts (are there new T cells specific for
                     the vaccinated peptides?) AND clinical endpoints
                     └─ these two do not always agree, which is §26.10

Steps 1 through 8 constitute a bespoke drug development program, for one person, compressed into weeks. That is precision medicine at the level of the individual tumor, and no other therapeutic modality is quite like it: the manufacturing process is the product, because no two doses are the same molecule.

Design details that turn out to matter

Long peptides beat minimal epitopes. Early trials often used minimal 9-residue epitopes, which load directly onto the class I molecules of any cell they meet — including cells with no costimulatory capacity, producing precisely the tolerizing presentation §26.4 warned about. Peptides of twenty to thirty residues cannot load directly; they must be taken up and processed by professional presenting cells, which is where you wanted the antigen anyway. They also carry both class I and class II epitopes, recruiting CD4 help. One of the field's genuine methodological improvements.

Clonality matters. A clonal mutation, present in every tumor cell, is a target the tumor cannot escape by losing a subpopulation. A subclonal mutation selects for the cells that lack it.

Some neoantigens are shared. A minority of mutations recur across patients — hotspot substitutions in commonly mutated oncogenes and tumor suppressors. Where one produces a peptide presentable by a common HLA allele, an off-the-shelf vaccine becomes possible for patients carrying both mutation and allele: cheaper, faster, testable in a conventional trial design, and in clinical development.

The tumor fights back. Even a perfect vaccine faces a microenvironment built to suppress T cells, and tumors escape antigen-specific attack by documented routes: losing expression of the target, mutating the machinery required to display any peptide at all, or losing one parental HLA haplotype. A vaccine that raises a beautiful response against a target the tumor then discards has produced an immune success and a clinical nothing.

🔬 Read the Study — the shape of the neoantigen evidence, not its numbers

I will describe the design of this literature rather than quoting results, because in a field moving this fast misquoted figures do real damage — and because design is more informative than any single number.

Generation one (published from 2017). Small, single-arm, first-in-human studies in melanoma — each with fewer than twenty patients — using either synthetic long peptides with a TLR agonist adjuvant, or an RNA-based construct. What they could show: that the pipeline is feasible in a clinically relevant timeframe, and that vaccinated patients develop measurable T-cell responses against neoepitopes not detectable beforehand. What they could not show: anything about efficacy. A single-arm study of a handful of patients with heterogeneous disease cannot separate a drug effect from selection, natural history, or concurrent therapy. Read any single-arm oncology result with the question Chapter 5 taught you: compared to what?

Generation two (randomized). Every patient receives a checkpoint inhibitor; patients are randomized to receive the individualized vaccine in addition. This is the correct design — the only way to isolate the vaccine's contribution — and it is hard: you are looking for an increment on top of an already active therapy, which demands more patients and longer follow-up than a comparison against nothing. The melanoma work in this generation has been encouraging enough to support larger confirmatory trials, which are underway.

What to watch for in any report here:

• Randomized or single-arm? If single-arm, no efficacy conclusion is available, whatever the press release says.
• What endpoint — recurrence-free survival, overall survival, or an immune measurement? An immune measurement is a process check, not a patient outcome.
• Planned primary analysis, subgroup, or interim look?
• What was the comparator arm receiving?
• How many patients, and how long was follow-up? Recurrence takes years to count.

📊 Evidence Rating

Claim: Individualized neoantigen vaccines improve clinical outcomes (recurrence-free or overall survival) in patients with solid tumors. Rating: 🔬 (frontier, as of 2026) Reason: The mechanism is established and the approach is proceeding properly through randomized trials with encouraging early signals, but no confirmatory randomized trial has established a durable clinical benefit, no such vaccine has general approval anywhere, and because these products are given alongside checkpoint inhibitors, attributing benefit to the vaccine component specifically is genuinely difficult. What would change it: completed, adequately powered randomized trials reporting survival or durable recurrence-free survival benefit, in defined tumor types and disease settings, with the vaccine arm separable from the checkpoint inhibitor's contribution by design. A single positive confirmatory trial in one setting would move this to ⚠️ for that setting; consistent results across settings plus approval would move it toward ✅ — for those populations and endpoints only.

⚠️ Hype Check — "the personalized cancer vaccine is here"

The headline, in its usual form:

"Scientists have created a personalized vaccine that trains the immune system to hunt down each patient's individual cancer. Trials show it works."

What is true in it, and it is a lot. The pipeline is real and runs on a clinically usable timescale. Vaccinated patients really do develop T-cell responses against their own tumor's mutations. Randomized trials are running, and early melanoma results were positive enough that sponsors and regulators agreed the idea deserves large confirmatory trials — which is not how most oncology ideas end up.

Where it fails. Four places.

"Trials show it works" compresses "an early randomized trial in one tumor type, in one disease setting, on a recurrence endpoint, with follow-up still accruing" into a claim about cancer in general. The compression is the entire problem.

"Is here" is not true in the sense a reader will take it. As of 2026 these products are available through clinical trials and expanded access arrangements, not by prescription, anywhere.

The combination confound vanishes from popular accounts. These vaccines are given with checkpoint inhibitors, which are themselves effective. The randomized design separates the two contributions, and the design is exactly what headlines omit.

And the framing quietly implies generality. The approach is most plausible where tumor burden is low and the immune system relatively intact — after surgery, in minimal residual disease — and in tumors with enough mutations to yield good targets. Tumors with few mutations offer little raw material, and that is structural rather than temporary.

Verdict. A ⚠️-shaped story reported as a ✅-shaped one, about a claim this book rates 🔬. If you or someone you love has cancer, the useful version is neither "there is a vaccine" nor "it is all hype." It is: this is a serious approach in active randomized testing, availability is through trials, eligibility depends on tumor type and disease stage, and your oncologist is the person who knows whether a trial is reasonable in your situation. Chapter 39 covers how to ask that well.

26.6 The mRNA relationship: instructions for a peptide, not the peptide

A reasonable objection: if the most advanced individualized cancer vaccines use mRNA, why are they in a book about peptides?

An mRNA vaccine does not deliver a peptide. It delivers instructions for your own cells to make one. The lipid nanoparticle fuses with cell membranes, the mRNA reaches the cytoplasm, ribosomes translate it into protein, and the protein enters exactly the processing and presentation machinery §26.2 described. The peptide a T cell eventually reads was manufactured inside your own cell, from a sequence someone designed.

So an mRNA vaccine is not a peptide drug — nothing peptide-shaped was administered. But it is unambiguously a peptide-antigen vaccine: the therapeutic message is a peptide sequence, and only the delivery differs. The designer chose amino acids. The vial contained nucleotides. The immune system read amino acids.

This is the same distinction Chapter 22 made about CGRP. There, migraine drugs target a peptide signaling system without being peptides — the peptide is the target, not the drug. Here the peptide is the message, not the drug. In both cases understanding the medicine requires understanding a peptide, and in neither case is the medicine itself one.

Why individualized programs reached for mRNA

The reason is manufacturing. To make an individualized peptide vaccine targeting, say, twenty neoepitopes, you must synthesize twenty distinct peptides to pharmaceutical standard, each with its own synthesis, purification, analysis, and release testing — for one patient. Some of those sequences will be difficult to make; peptide synthesis does not treat all sequences equally (Chapter 32).

For the mRNA equivalent you design one nucleic acid encoding all twenty epitopes end-to-end and run one synthesis. The process is the same regardless of which epitopes the algorithm chose, which is exactly what you need when the product differs for every patient. Faster, cheaper, and far more standardizable — and standardizability decides whether this modality can ever be delivered at scale rather than in a few academic centers.

The trade-offs belong in any honest account: the antigen is expressed transiently and you control the sequence rather than the displayed peptides; the platform needs cold chain and specialized formulation; the lipid nanoparticle supplies both useful adjuvant activity and most of the reactogenicity; and long constructs must be designed so that joining epitopes end-to-end does not create junctional sequences that behave unpredictably.

🧬 The Molecule — what is actually in the vial, across platforms

Five ways to deliver the same informational payload. Notice that the immunological endpoint is identical in every row, and that only one of them contains a peptide.

Platform	What is administered	How the epitope reaches the groove
Synthetic long peptides	20–30-residue peptides plus adjuvant	taken up and processed by presenting cells
Minimal epitope peptides	8–10-residue peptides plus adjuvant	load directly onto surface class I — including on cells that cannot costimulate, the §26.4 tolerance risk
mRNA in lipid nanoparticles	nucleic acid; the LNP is partly its own adjuvant	translated inside the cell, then processed normally
DNA plasmid	nucleic acid, often with electroporation	transcribed, translated, processed
Autologous cell products	the patient's own presenting cells, loaded outside the body	the presenting step happens in a facility, not in the patient

Every row ends in the same place: a peptide, in a groove, on a surface, read by a T cell. The platform is a delivery argument. The peptide is the therapy.

26.7 The trials in progress, date-stamped

Everything here is a statement about status as of 2026, written to be checked and to go out of date. That is the point of date-stamping (Chapter 5, rule 5): a rating that cannot expire is not a rating, it is an opinion.

Where individualized cancer vaccines stand

The field has moved from single-arm feasibility studies into randomized trials. That transition matters more than any individual result: it is the move from "we can do this" to "does it help anyone."

The most advanced work is in melanoma, after surgical removal of high-risk disease, where an individualized vaccine is added to checkpoint inhibitor therapy and compared against checkpoint inhibitor therapy alone. Randomized results in this setting have been encouraging enough to support larger confirmatory trials, which are running. I am deliberately not quoting effect sizes: interim results from mid-size randomized trials are among the most commonly misreported statistics in oncology journalism, and if you need the number you need it from the primary publication with its confidence interval attached, not from a textbook written a year earlier.

Trials are ongoing in other tumor types, including lung, kidney, bladder, colorectal, pancreatic, and head and neck cancers. Several are in the adjuvant or minimal residual disease setting — after surgery, when imaging shows no evident disease but recurrence risk is high, sometimes selected by circulating tumor DNA. There are good reasons to expect that setting to be the most favorable: tumor burden is lowest, the immunosuppressive microenvironment left with the tumor, the immune system is in better shape than it will be later, and the weeks required to manufacture are weeks in which the patient is not losing ground.

One instructive small study. In resected pancreatic cancer — historically dismal outcomes, relatively few mutations — a small single-arm study of an individualized mRNA vaccine reported that the subset of patients who developed a vaccine-induced T-cell response appeared to remain recurrence-free longer than those who did not. That is striking in a disease that badly needs something, and it is a comparison between responders and non-responders, which is not a randomized comparison at all. Patients able to mount a strong immune response may differ systematically from those who cannot — in immune fitness, in disease biology, in how much treatment they completed. This confound has misled oncology repeatedly. It is a reason to run a randomized trial, not evidence of benefit.

As of this writing, no individualized neoantigen cancer vaccine has general approval. Access is through clinical trials.

The attribution problem, stated precisely

Because these vaccines are given with checkpoint inhibitors, someone always asks how we will know the vaccine did anything. The answer is better than people expect: randomization solves it in principle. If every patient in both arms receives the checkpoint inhibitor and only one arm also receives the vaccine, any difference between arms is attributable to the vaccine. That is the design being used.

The difficulty is practical rather than logical. Detecting an increment on top of an active therapy requires more patients and longer follow-up than detecting an effect against nothing. If benefit is concentrated in a subgroup — high mutational burden, particular HLA types, particular settings — an overall trial may dilute it into non-significance, while a subgroup analysis that finds it will be hard to trust. And the endpoints available early are recurrence-based; translating a recurrence benefit into a survival benefit is an inference, not an observation. These are the ordinary difficulties of good oncology trials, and they are reasons to wait for the primary analysis rather than reasons for cynicism.

Peptide vaccines in infectious disease

The infectious disease side of this chapter's title is shorter, and the brevity is informative.

No purely peptide-based prophylactic vaccine is in routine widespread use. The vaccines that prevent disease at population scale are live attenuated organisms, inactivated organisms, recombinant proteins, polysaccharide conjugates, virus-like particles, viral vectors, and mRNA. Not short peptides.

The reason is §26.2. Protection against most pathogens depends substantially on neutralizing antibodies, antibodies read conformational surfaces, and short peptides present those poorly. Peptide immunogens routinely raise antibodies that bind the peptide beautifully and the actual pathogen badly. Add HLA restriction (§26.3) — a serious constraint when your target population is everyone on earth — and a very high safety bar for healthy recipients, and the arithmetic goes against short peptides for most prophylactic uses.

Peptide approaches remain genuinely interesting where the goal is T-cell immunity rather than antibodies, or where one region of a pathogen must be targeted while carefully avoiding another: candidates aiming at conserved internal proteins for broader influenza coverage, T-cell-directed constructs for tuberculosis and HIV, and bacterial vaccine candidates built around defined epitope regions specifically to avoid provoking cross-reactive responses against human tissue. Legitimate research programs, all of them — and all unlicensed as of 2026.

💊 In the Clinic — what a cancer vaccine trial actually involves

No protocols and no dosing; this is orientation only, so the vocabulary is familiar if you ever sit in a room where it is used.

It starts with tissue. An individualized vaccine requires tumor material of sufficient quantity and quality plus a normal comparison sample, usually blood. Archival tissue is sometimes insufficient — one of the more common reasons a patient turns out to be ineligible.

Then a wait. Sequencing, prediction, manufacture, and release testing take weeks, during which patients receive standard-of-care therapy; the vaccine is added afterward. This is why the adjuvant setting is attractive: it is a period where a wait is tolerable.

Then randomization. You do not choose your arm, and neither does your oncologist. That is uncomfortable, and it is the only reason the trial can answer anything. A trial is run because nobody knows which arm is better; if someone did, running it would be unethical.

Monitoring runs on two tracks. Blood samples ask whether new T cells specific for the vaccinated peptides have appeared; scans ask whether the cancer has returned. They do not always agree.

On participation. A trial is not a last resort and not a guaranteed benefit. It is a treatment option with a specific, disclosed uncertainty. Your oncologist and the trial team judge whether one fits your situation; the questions worth asking are about eligibility, what the comparator arm receives, what is required of you logistically, and what happens if the disease progresses during the study. Chapter 39 is about that conversation.

26.8 Therapeutic versus prophylactic: two different problems wearing the same word

The word vaccine covers two enterprises that share a mechanism and almost nothing else.

A prophylactic vaccine is given to healthy people to prevent a disease they do not have. The recipient's immune system is intact and unhurried, and there is no time pressure — the rehearsal can happen years before the performance. The population is enormous, so tolerable risk is extraordinarily low: a serious adverse event at one in a hundred thousand is program-ending when you are treating millions of healthy people. Efficacy is measured by counting cases that did not happen.

A therapeutic vaccine is given to a person who already has the disease. Every one of those parameters inverts.

These are different problems, and the long history of therapeutic cancer vaccine failure is largely a history of underestimating how different. A generation of programs reasoned from prophylactic success — vaccines work, here is an antigen, therefore vaccinate — without reckoning with tolerance (§26.1, §26.5), adjuvant requirements (§26.4), HLA restriction (§26.3), or a tumor microenvironment evolved specifically to shut down the response they were trying to induce.

The record, honestly

Large randomized therapeutic cancer vaccine trials have failed repeatedly across decades, in multiple tumor types, with multiple antigen classes and platforms. Programs targeting a cancer-testis antigen in lung cancer and melanoma, a mucin-derived antigen in lung cancer, a telomerase-derived peptide in pancreatic cancer, and a mutant growth-factor-receptor peptide in glioblastoma all reached large randomized trials and all failed their primary endpoints. In several cases the vaccines demonstrably raised the immune response they were designed to raise. That is the part worth internalizing: immunological success and clinical failure occurred in the same trial, repeatedly.

The successes are few and instructive. One autologous cellular immunotherapy for metastatic prostate cancer was approved on an overall survival benefit without a corresponding improvement in progression measures — a discordance still debated, and a demonstration of how oddly immunotherapies behave against conventional endpoints. Its logistical complexity and cost also became a cautionary tale about individualized manufacturing, directly relevant to §26.5. An engineered oncolytic virus injected into lesions has approval in melanoma. And a bacterial preparation instilled into the bladder has been standard therapy for early bladder cancer for decades — an immunostimulant rather than a vaccine, but the oldest evidence that provoking immunity locally can control a tumor.

Here is the fact that ought to lead any honest discussion of cancer vaccines: the most successful cancer vaccines in existence are prophylactic antiviral vaccines. Vaccination against hepatitis B prevents a large fraction of liver cancer, and vaccination against human papillomavirus prevents cervical and other cancers. Both work by preventing the chronic viral infection that causes the cancer, both are given to healthy people long before any tumor exists, and neither is a peptide vaccine — one is a recombinant surface protein, the other a virus-like particle. They are the largest cancer-prevention successes immunology has produced, and they succeeded by solving the easier problem.

Why the failure record is not a verdict on neoantigens

It would be lazy to conclude from that history that individualized neoantigen vaccines will fail too. Three things changed, and each addresses a specific diagnosed cause of the old failures:

1. The antigens are genuinely foreign. Tolerance was arguably the deepest problem, and neoantigens are the one class of tumor antigen that sidesteps it rather than fighting it.
2. The adjuvants and formats are better. Long peptides instead of minimal epitopes, defined TLR-targeting adjuvants instead of depot emulsions, nucleic acid platforms with built-in innate stimulation.
3. Checkpoint blockade exists. Previous vaccines asked T cells to work inside a tumor that switched them off on arrival. Checkpoint inhibitors partially release that brake, and the combination is not merely additive in concept — a vaccine that generates T cells and a drug that keeps them functional address two halves of the same failure.

Those are real, specific, mechanistically motivated differences. They are also, at present, an argument rather than a result. The history says the prior probability for this class of intervention is low; the mechanistic changes say this attempt differs in ways that plausibly matter; the randomized trials are how we find out which consideration dominates. That is exactly the situation the 🔬 rating describes.

📊 Evidence Rating

Claim: Therapeutic cancer vaccines, as a general historical class, improve survival in patients with established solid tumors. Rating: ❌ to ⚠️ (as of 2026) Reason: A long record of large randomized trials failing to meet primary endpoints across multiple tumor types, antigens, and platforms — often while demonstrably inducing the intended immune response — against a small number of genuine successes, including one approved autologous cellular immunotherapy with a survival benefit in metastatic prostate cancer. What would change it: it is already changing; each confirmatory randomized success in a defined population moves the corresponding claim upward for that population.

Read the scope of this rating carefully. It is a statement about the historical class, not about current neoantigen approaches, which differ in the specific and potentially decisive ways listed above. Rule 6 of the rating system: one molecule, many ratings — and here, one category, many ratings. Using this ❌ to dismiss neoantigen vaccines would be exactly the error rule 4 forbids: downgrading a claim by association rather than by evidence. Using neoantigen optimism to retroactively rescue the historical class would be the mirror error.

26.9 Peptide immunotherapy for allergy: teaching the immune system to stand down

Everything so far has been about teaching the immune system to attack. The same machinery can, in principle, be run in reverse.

An allergy is a misdirected immune response. In an allergic person, exposure to an otherwise harmless protein — cat dander, grass pollen, dust mite, peanut — produces IgE antibodies against that allergen, which sit on the surface of mast cells and basophils. When the allergen arrives it binds two adjacent IgE molecules and cross-links them, triggering release of histamine and other mediators within seconds: the itching, the swelling, the wheezing, and at worst anaphylaxis.

Allergen immunotherapy — repeated controlled exposure, by injection or under the tongue, escalating over years — is an established treatment for several allergic conditions and can produce lasting benefit after the course ends. It has two well-known drawbacks: it takes a long time, and because it uses the whole allergen it can itself cross-link IgE and cause reactions, occasionally severe. That is the tension the peptide approach was designed to resolve.

The idea

Cross-linking requires the allergen to engage two IgE molecules at once, which requires two spatially separated B-cell epitopes on one molecule. A short peptide is too small to do that. It carries a T-cell epitope and cannot bridge two antibodies.

So: identify the dominant T-cell epitopes of the allergen, synthesize just those, and administer them. The T cells driving the allergic response meet their epitope repeatedly in a setting deliberately lacking danger signals — which, per §26.4, pushes them toward anergy, deletion, or conversion into regulatory cells rather than activation. Meanwhile the mast cells never fire, because nothing present can cross-link their IgE. In principle: the tolerance benefit without the anaphylaxis risk, and faster. The intended downstream changes are those of successful conventional immunotherapy — suppression of the type 2 response, regulatory T cells and IL-10, and a shift toward IgG4 antibodies that intercept the allergen before it reaches IgE.

An elegant idea, and it has been one for a long time.

What happened in the clinic

Peptide immunotherapy has been tested in randomized, placebo-controlled trials in cat, grass pollen, house dust mite, and bee venom allergy. Early and mid-stage trials produced encouraging results, including symptom improvements and the expected immunological changes.

Then the largest programs did not replicate at phase 3. A prominent cat allergy program advanced into large late-stage trials and failed to separate from placebo — and the widely discussed reason recurs across allergy research: placebo responses in allergy trials are unusually large. Symptom scores improve substantially in placebo arms, driven by expectation, regression to the mean, seasonal variation in exposure, and the subjectivity of the endpoints. A treatment can produce a real biological effect and still fail to beat placebo on a symptom score. Related programs in other allergens were subsequently discontinued.

Work continues with revised peptide selection, altered schedules, and different allergens, and the mechanistic rationale has not been refuted. But the honest statement is: real randomized human data exist, results are mixed, the largest confirmatory efforts did not succeed, and the question is open.

📊 Evidence Rating

Claim: Peptide-based allergy immunotherapy reduces allergic symptoms in patients with IgE-mediated allergy. Rating: ⚠️ (as of 2026) Reason: Multiple randomized placebo-controlled human trials have been conducted with mixed results — encouraging earlier-phase findings and expected immunological changes, but the most prominent late-stage program failed to separate from placebo, in a therapeutic area where placebo responses are notoriously large. What would change it: a well-powered phase 3 trial with objective or validated composite endpoints showing separation from placebo would support ✅ for that allergen and population; continued late-stage failures would move this toward ❌ for the approach as formulated.

Notice why this is ⚠️ and not 🔬; the contrast with §26.5 is the most useful thing here. Peptide allergy immunotherapy has been through multiple randomized, placebo-controlled trials including large late-stage ones — real randomized human data that does not settle the question, which is the definition of ⚠️. Individualized neoantigen vaccines are earlier: the randomized evidence is younger and thinner, and confirmatory trials are still running. Same rating system, different position on the same axis.

The allergy story also carries a warning for the cancer story. A mechanistically beautiful intervention produced the expected immunological changes and then failed on the endpoint that mattered to patients. §26.8 documented the same pattern in oncology. Immune readouts are not clinical endpoints.

26.10 Why this is 🔬, and what would move it

Chapter 5 built the rating system; this section is where its hardest distinction earns its keep.

⚠️ requires real randomized human data that does not settle the question. 🔬 is early-stage science proceeding properly, where translation is unproven. The line is not about how promising something feels, how much money is behind it, or how many press releases it has generated. It is about what has been demonstrated in randomized human trials.

Individualized neoantigen cancer vaccines are 🔬, and because randomized data do exist, the reasoning has to be precise. Five considerations:

1. The randomized evidence is early and narrow. Mid-size randomized trials in one tumor type and one disease setting, reading out on recurrence, with confirmatory trials still accruing. That is the beginning of an evidence base, not the settling of one.

2. The intervention is a process, not a molecule. Every claim in this book so far concerned a defined chemical entity — the same semaglutide for everyone. An individualized vaccine is a different product for every patient, generated by a pipeline of sequencing, prediction, selection, and manufacture. What a trial tests is that pipeline. Whether it performs equivalently for a different tumor type, mutational landscape, HLA background, or manufacturing site is genuinely open in a way it is not for a fixed drug. This is the most underappreciated reason for caution and it is specific to this modality.

3. Attribution against an active comparator is unresolved. These products are given with checkpoint inhibitors that work on their own. Randomization isolates the increment in principle (§26.7); the trials that will do so definitively are ongoing.

4. Nothing is approved. No regulator has judged the evidence sufficient for general use — a fact about the evidence base, not a formality.

5. Mechanism is not the question. Rule 3: never upgrade with mechanism. Peptide presentation is ✅ (§26.2), and neoantigens are real, immunogenic, and relevant to how existing immunotherapy works. None of that is evidence that vaccinating against them extends anyone's life, which is the claim being rated.

This sits near the ⚠️/🔬 boundary and a reasonable person could place it on the other side. This book puts it at 🔬 because the modality — individualized manufacturing as a therapeutic platform — has not yet produced a completed confirmatory trial demonstrating durable clinical benefit. Under rule 5 that judgment is date-stamped and falsifiable, and may well be wrong by the time you read it. That is a feature. A rating you cannot imagine changing is not a rating.

What would move it up

• A completed, adequately powered, randomized confirmatory trial reporting a durable recurrence-free or overall survival benefit in a defined tumor type and setting, with the vaccine arm separable from the checkpoint inhibitor's contribution by design.
• Replication in a second tumor type or setting — what distinguishes a platform from a single lucky application.
• Regulatory approval, which imports independent expert review of the full dataset.
• Evidence that benefit is not confined to high-mutational-burden tumors and favorable HLA backgrounds — or clear characterization of exactly which patients benefit, which is nearly as useful.
• Manufacturing timelines and quality proving reproducible outside a few specialized centers.

Any one of those moves the claim for that setting to ⚠️. Several together, plus approval, move it to ✅ for those populations and endpoints — and no further, because that is how rule 1 works.

What would move it down

• Confirmatory trials failing to reproduce the early randomized signal.
• Benefit turning out to be attributable to the checkpoint component or to comparator-arm differences.
• Benefit confined so narrowly that the addressable population becomes negligible.
• Manufacturing turnaround proving incompatible with the pace of the diseases that most need it.

Writing both lists is the discipline. A rating with only an upgrade path is optimism with a symbol on it.

🔍 Check Your Understanding

1. Peptide allergy immunotherapy is rated ⚠️ and individualized neoantigen vaccines are rated 🔬, even though the cancer work arguably has more momentum behind it. Explain the distinction using the definitions rather than the subject matter.
2. §26.2 rates peptide-epitope presentation ✅. Why does that rating not support any clinical claim in this chapter?
3. Why does the fact that an individualized vaccine is "a process, not a molecule" create an evidentiary problem that does not arise for a conventional drug?
4. Name one result that would move the neoantigen rating up and one that would move it down.

📋 Your Evidence Dossier

Field 3 — Mechanism — for immune interventions.

Every previous chapter filled Field 3 the same way: name the receptor, the tissue, the downstream signal, and describe the effect as occupancy over time. That template breaks completely here.

A vaccine has no receptor in the pharmacological sense. Its "target" is a T-cell repertoire and its "effect" is memory rather than occupancy. The molecule you administered is gone long before anything you care about happens. Fill in Field 3 the usual way for a vaccine and you will write something technically true and completely uninformative.

So this chapter adds a variant of Field 3 for immune interventions. Use it for anything that trains a system rather than binding a target: vaccines, allergen immunotherapy, and — with modification — cell therapies.

FIELD 3 — MECHANISM (IMMUNE VARIANT)
  Administered       the physical contents of the vial
  The message        the peptide sequence(s) the immune system is meant to learn
  Who presents it    class I, class II, or both — restricted to which HLA?
  Cells taught       CD8, CD4, B cells, or regulatory cells
  Direction          immunity (attack) or tolerance (stand down)
  Danger signal      the adjuvant, and which innate receptor it engages
  Where it acts      the draining lymph node — NOT the site of disease
  Unit of effect     repertoire change and memory; NOT receptor occupancy
  Immune readout     how a response is measured in blood
  Clinical endpoint  what actually happened to patients — kept separate
  Duration logic     half-life is irrelevant; durability of memory is the question
  What defeats it    tolerance, HLA mismatch, antigen loss, suppressive environment

Worked demonstration — an individualized neoantigen cancer vaccine

FIELD 3 — INDIVIDUALIZED NEOANTIGEN VACCINE     [worked demonstration]

  Administered       Synthetic long peptides with an adjuvant, OR an mRNA construct in
                     a lipid nanoparticle. Different physical product per patient.
  The message        Peptide sequences carrying amino acid changes from somatic
                     mutations in THIS tumor, absent from the normal proteome.
  Who presents it    Class I (CD8 killing) and class II (CD4 help), selected
                     computationally against this patient's own HLA alleles.
  Cells taught       CD8 cytotoxic T cells primarily; CD4 help deliberately included
                     because durable memory requires it.
  Direction          Immunity. Opposite of §26.9's goal, same machinery.
  Danger signal      A TLR-targeting adjuvant (peptide) or the LNP itself (mRNA).
  Where it acts      The lymph node draining the injection site. The T cells must then
                     find the tumor — a separate problem the vaccine does not solve.
  Unit of effect     New or expanded neoantigen-specific clones, and memory.
                     NOT drug concentration. NOT receptor occupancy.
  Immune readout     T cells responding to vaccinated peptides that were undetectable
                     before — measured in blood.
  Clinical endpoint  Recurrence-free survival; overall survival. RECORD SEPARATELY.
  Duration logic     The construct clears in days. The claim is about years.
  What defeats it    Non-immunogenic predicted epitopes; the tumor losing the mutation
                     or its ability to present anything; a suppressive microenvironment
                     disabling the T cells on arrival; manufacturing slower than the
                     disease.
  Rating (2026)      🔬 for clinical outcomes in solid tumors — see §26.5, §26.10.

The one thing this field is for

Look at the two lines kept deliberately apart: immune readout and clinical endpoint.

For a receptor drug, "does it engage the target" is nearly the same question as "does it work." For an immune intervention it is not even close. §26.8 documented large randomized cancer vaccine trials that induced exactly the immune response they were designed to induce and produced no clinical benefit. §26.9 documented an allergy program that produced the expected immunological changes and did not beat placebo.

So the rule for this variant of Field 3 is: never let an immune readout occupy the line reserved for a clinical endpoint. If a report says patients developed robust T-cell responses, write it under immune readout and leave the clinical line empty. An empty clinical line is information — the single most valuable thing your dossier can record about a frontier therapy, and exactly what press coverage of this field obscures.

If you added a cancer vaccine to your dossier for personal reasons, fill in both lines with equal care and date the entry. Dating matters more here than anywhere else in the book, because this is the fastest-moving material in it.

Conclusion

The immune system cannot see a virus or a tumor. It sees fragments — short peptides, held in grooves, on cell surfaces — and decides what to do based on what those fragments are and what else was happening when it met them. That mechanism is as well established as anything in biology, and it makes vaccination peptide pharmacology practiced on an immune system rather than a receptor.

Everything difficult in this chapter follows from it. Because the display molecules are the most variable genes we have, a peptide that works in one person may not be displayed at all in another, and no amount of clever chemistry removes that constraint. Because a naked peptide carries no evidence of danger, the adjuvant is frequently the whole difference between a response and nothing — and occasionally between a response and active tolerance. Because tumors are made of self, decades of therapeutic cancer vaccines asked the immune system to violate its most carefully constructed rule, and it declined.

Neoantigens are the escape hatch, and a genuinely beautiful idea: mutations that make a patient's own tumor foreign, found by sequencing that patient's tumor and normal tissue, filtered by prediction against that patient's own HLA, manufactured to order, and given alongside a drug that keeps the resulting T cells from being switched off on arrival. Delivered as peptides or as mRNA encoding them, the therapeutic message is a peptide sequence. It is the most personal medicine anyone has attempted.

And it is 🔬. Randomized trials are running; early melanoma results were encouraging enough to justify confirmatory trials; nothing is approved; and the field's history is full of vaccines that generated the intended immune response and changed nothing that mattered. Both halves are true at once, and holding both is not fence-sitting. It is the only honest position available in 2026, and it has an expiration date — the confirmatory trials will report, and the rating will move, one way or the other.

If you are reading this because cancer is not an abstraction in your life, the takeaway is neither "a vaccine is coming" nor "it is all hype." It is that a serious idea is being tested seriously, that access today is through clinical trials, and that your oncologist is the person who can tell you whether that is relevant to your situation.

Chapter 27 turns to a different oncology use of peptides — not teaching the immune system, but using peptide targeting to deliver something directly to a tumor cell. Peptides again; entirely different logic.

Key Terms

Antigen — anything the adaptive immune system can recognize; usually a whole protein containing many potential epitopes.

Epitope — the portion of an antigen an immune receptor actually contacts. T-cell epitopes are short linear peptides; B-cell epitopes are usually three-dimensional surface patches.

Linear epitope — a contiguous stretch of sequence, recognized by T cells after the protein has been fragmented.

Conformational epitope — a surface patch assembled by folding from residues distant in sequence; the usual antibody target, and the reason short peptides make poor antibody-directed vaccines.

Immunogen — an antigen that in practice provokes a response. Not all antigens are immunogens; an adjuvant is often the difference.

MHC (major histocompatibility complex) — the family of cell-surface molecules that display peptides to T cells.

HLA (human leukocyte antigen) — the human MHC. Classical class I: HLA-A, -B, -C. Classical class II: HLA-DR, -DQ, -DP.

Class I presentation — display of 8–10-residue peptides from proteins made inside the cell, on nearly every nucleated cell, to CD8 T cells.

Class II presentation — display of 13–25-residue peptides from material taken in from outside, on professional antigen-presenting cells, to CD4 helper T cells.

HLA restriction — a T cell recognizes its peptide only when presented by a particular HLA molecule; the reason a peptide vaccine may work in one person and be invisible to another.

Anchor residues — the peptide side chains that fit pockets in the HLA groove; they determine whether a peptide can be presented by a given allele.

CD8 T cell (cytotoxic T lymphocyte) — kills cells displaying a recognized peptide on class I.

CD4 T cell (helper T cell) — recognizes peptides on class II and coordinates the response; its help is required for durable CD8 memory and high-quality antibodies.

Immunological memory — the persistent population of antigen-experienced cells left after a response; the actual product of vaccination.

Adjuvant — the component supplying the innate danger signal that converts antigen exposure into a durable response. For peptide vaccines, frequently the difference between a response and nothing.

Pattern recognition receptor — an innate receptor, such as a Toll-like receptor, detecting signatures of pathogens or damage; the target of most modern adjuvants.

Costimulation — the second signal a T cell requires alongside peptide recognition. Without it, peptide typically produces anergy or tolerance rather than immunity.

Anergy — functional unresponsiveness in a lymphocyte that met its antigen without costimulation.

Central tolerance — developmental deletion of T cells recognizing self peptides too strongly; the reason vaccines against self antigens perform poorly. Peripheral tolerance restrains the escapees.

Tumor-associated antigen — a self protein overexpressed or aberrantly expressed by tumors; subject to tolerance, and the target class behind most historical cancer vaccine failures.

Neoantigen — a peptide containing an amino acid change produced by a somatic mutation, absent from the normal proteome and therefore not subject to central tolerance.

Tumor mutational burden — the number of somatic mutations in a tumor; associated with checkpoint inhibitor response and with the availability of neoantigen targets.

Clonal mutation — present in every tumor cell; a target the tumor cannot escape by losing a subpopulation. A subclonal mutation is present in only some cells and is an escape route.

HLA loss of heterozygosity — a tumor's loss of one inherited set of HLA genes; a documented escape route from antigen-specific attack.

Checkpoint inhibitor — a drug blocking inhibitory signals that restrain T cells. Not a peptide, but the near-universal partner of cancer vaccines in current trials, and the source of §26.7's attribution problem.

Synthetic long peptide — a 20–30-residue vaccine peptide that must be processed by professional presenting cells, avoiding the tolerizing direct loading possible with minimal epitopes.

Lipid nanoparticle — the mRNA delivery vehicle; simultaneously carrier and much of the adjuvant effect.

Prophylactic vaccine — given to healthy people to prevent disease; extremely high safety bar, intact immune system, no time pressure. Therapeutic vaccine — given to treat existing disease against an immune system that has already failed.

Allergen immunotherapy — repeated controlled allergen exposure intended to induce tolerance; established for several allergic conditions in its whole-allergen form.

Regulatory T cell — a suppressive T cell; an obstacle in cancer vaccines and a goal in allergy immunotherapy.

Spaced Review

1. (Ch 26 + Ch 1) Using Chapter 1 §1.2, explain why changing a single residue at an anchor position can abolish presentation entirely, while changing a residue that points upward out of the groove may leave presentation intact and still change whether any T cell recognizes it.

2. (Ch 26 + Ch 25) Chapter 25's antimicrobial peptides act directly on microbes within minutes and leave no memory. This chapter's peptides do nothing to a pathogen at all and may produce protection lasting years. Explain what each system trades away to get what it has, and why an antimicrobial peptide could never do a vaccine's job.

3. (Ch 26 + Ch 5) A news report says a personalized cancer vaccine "worked" in a trial. Write the five questions you would need answered before assigning any rating — and state which single answer, if unfavorable, would make the others irrelevant.

4. (Ch 26 + Ch 3) Chapter 3 argued that half-life and exposure profile determine a peptide drug's effect. Explain why a vaccine's half-life is nearly irrelevant, what replaces it as the design variable, and why an adjuvant that keeps antigen at the injection site longer is not reliably better.

5. (Ch 26 integrative) Two claims: (a) T cells recognize peptides presented on HLA molecules; (b) vaccinating a patient with peptides from their tumor's mutations extends survival. This chapter rates (a) ✅ and (b) 🔬. Explain to someone with no immunology background why both are correct at once, and why using (a) to argue for (b) is precisely the error the rating system exists to prevent.