Why Do PD-L1-Negative Patients Respond to PD-1 Immunotherapy?
What an immune-cold tumour, dynamic PD-L1 biology and the difference between PD-1 and PD-L1 blockade reveal about a baseline “negative” result.
I came to this question as a patient.
I was trying to understand one of the great breakthroughs in cancer treatment: drugs such as nivolumab (Nivo) and pembrolizumab (Keytruda) that block PD-1 and can restore or strengthen an anti-tumour immune response.
The deeper I went, the more one result stopped me.
Patients whose tumours tested PD-L1 positive responded to PD-1 blockade.
That fitted the biological picture.
But patients whose tumours tested PD-L1 negative responded too.
Not one or two patients. A substantial proportion.
In one major nivolumab trial, 41.3% of patients classified as PD-L1 negative had an objective response.
Roughly 4 out of every 10 patients.
That is not a result to step over.
It is part of the evidence, and it creates a question.
If the test says PD-L1 negative, but blocking the PD-1 pathway still works, what exactly did the negative test establish?
I am not starting with an explanation for why those patients responded.
I am starting with the observation that they did.
The breakthrough in PD-1 immunotherapy answered an enormous question.
But inside that answer was another question that should not be left behind.
The question a study sets out to answer is not necessarily the most important question its results uncover.
How James Allison and Tasuku Honjo Changed Cancer Immunotherapy
To understand the unanswered question, first we need to understand the discovery that made the question possible.
Two scientists working independently uncovered different ways the immune system can restrain T-cell activity.
Professor James Allison studied a protein called CTLA-4.
Professor Tasuku Honjo discovered another protein called PD-1.
Think of them as different control points in the immune system's braking system.
These brakes exist for a reason. An immune response needs to be powerful enough to deal with a threat, but controlled enough to limit damage to normal tissue.
Cancer can benefit from these same protective controls.
Allison's work showed that blocking CTLA-4 could release one of these restraints and strengthen an anti-tumour immune response. That work led to ipilimumab (Ipi).
Honjo's discovery of PD-1 opened another path. Research that followed established PD-1 as an inhibitory immune checkpoint and ultimately led to treatments including nivolumab (Nivo) and pembrolizumab (Keytruda).
Allison and Honjo had not discovered two names for the same brake.
They had uncovered different control points capable of restraining T-cell activity.
In 2018, they shared the Nobel Prize in Physiology or Medicine for the discovery of cancer therapy by inhibition of negative immune regulation.
The breakthrough changed the way cancer could be treated.
Instead of only asking how to attack the cancer directly, another question became possible:
What if treatment could release a restraint on the immune system and create an opportunity for the patient's own T cells to attack the cancer?
That question led to extraordinary treatment advances.
But those advances also produced new observations that the original discoveries could not yet explain.
To understand one of them, we need to look closely at the PD-1 pathway.
What PD-1 and PD-L1 Do in Cancer Immunotherapy
Now we can focus on the checkpoint at the centre of this question: PD-1 and PD-L1.
The simplest way to understand PD-1 is to picture a brake on a T cell.
T cells are specialised immune cells capable of recognising and responding to abnormal cells, including cancer cells.
PD-1 is a protein found on the surface of activated T cells.
PD-L1 is a protein that can be expressed by tumour cells and other cells within the tumour environment.
When PD-L1 binds to PD-1, an inhibitory signal is sent into the T cell.
Think of it simply as:
PD-L1 engages the PD-1 brake.
That signal can restrain the T cell's activity.
This is where Nivo and Keytruda come in.
These drugs bind to PD-1 and interfere with inhibitory signalling through that checkpoint.
They do not directly kill the cancer cell.
They block an inhibitory signal, creating an opportunity for anti-tumour T-cell activity to continue or recover when the right immune biology is present.
The mental picture is simple:
T cell → PD-1 brake → PD-L1 engages the brake → PD-1 inhibitor blocks the inhibitory signal.
Now apply first-principles reasoning to that picture.
If a tumour is classified as PD-L1 negative, the simplified model might lead us to expect less benefit from blocking the PD-1 pathway.
Yet substantial numbers of patients classified as PD-L1 negative responded.
That apparent contradiction is not something to explain away. It is information.
Why the PD-L1-Negative Response to Nivolumab Matters
Now we come to the result that changed the question for me.
In the phase 3 CheckMate 067 trial, patients with previously untreated advanced melanoma received nivolumab, nivolumab plus ipilimumab, or ipilimumab alone.
Tumour tissue was also assessed for PD-L1 expression.
Patients whose tumours were PD-L1 positive showed substantial responses to nivolumab.
But then look at the patients whose tumours were classified as PD-L1 negative.
This was not the only nivolumab trial showing responses among patients without detectable baseline PD-L1.
In the phase 3 CheckMate 066 trial, the objective response rate among patients with negative or indeterminate PD-L1 status receiving nivolumab was 33.1%.
So the observation was not simply that an occasional PD-L1-negative patient responded.
Substantial groups of patients did.
These patients are not noise around an otherwise successful result.
They are part of the result.
And an anomaly is information.
It does not tell us what the explanation is.
But when a substantial group behaves differently from what a simplified biological model might lead us to expect, it tells us that something about the measurement, the model, or both may be incomplete.
The proof points establish that the question exists. They do not establish the answer.
If nearly 4 in 10 patients classified as PD-L1 negative responded to nivolumab, what exactly does the word "negative" establish biologically?
What Did the Patients Actually Experience?
Before looking at the results, three things need to be made clear: response, complete response and durability.
ORR means Objective Response Rate: the percentage of patients who achieved either a partial response or a complete response. Under RECIST 1.1, a partial response generally requires the measured target tumours to shrink by at least 30%.
CR means Complete Response: all detectable target cancer disappeared on the scans used to assess the response. A CR records what was observed at that assessment. It does not mean the patient has been proven cured.
A CR can later be followed by relapse. If that happens, the earlier complete response still stands as a CR. What changes is its duration.
Durable CR means the complete response continues over time. Durability can only be established by continuing to follow the patient. This is why the follow-up point of a trial matters when reading complete-response results.
The simplest way to remember the difference is: CR tells us what happened. Durability tells us how long it lasted.
Nivo = nivolumab: a drug that blocks the PD-1 immune checkpoint.
Keytruda = pembrolizumab: another drug that blocks PD-1.
Ipi = ipilimumab: a drug that blocks a different immune checkpoint, CTLA-4.
Did Patients Classified as PD-L1 Negative Respond?
This is the result at the centre of the investigation.
Look first only at patients whose tumours were classified as PD-L1 negative.
| Trial | Treatment | Patients being measured | ORR | CR |
|---|---|---|---|---|
| CheckMate 067 | Nivo | PD-L1 negative | 41.3% | Not separately identified |
| CheckMate 066 | Nivo | PD-L1 negative or indeterminate | 33.1% | Not separately identified |
| KEYNOTE-006 | Keytruda | PD-L1 negative | 25% | Not separately identified here |
| KEYNOTE-006 | Ipi | PD-L1 negative | 13% | Not separately identified here |
| CheckMate 067 | Nivo + Ipi | PD-L1 negative | 54.8% | Not separately identified |
| CheckMate 067 | Ipi | PD-L1 negative | 17.8% | Not separately identified |
| CheckMate 069 | Nivo + Ipi | PD-L1 negative | 55% | Not separately identified |
| CheckMate 069 | Ipi | PD-L1 negative | 4% | Not separately identified |
| KEYNOTE-029 | Keytruda + Ipi | PD-L1 negative | 50.0% | 33.3% |
These trials are not one head-to-head comparison. They involved different patient groups, study designs, treatments and PD-L1 testing methods. The percentages should not be used to declare one treatment better than another.
But the patient should be able to see the observation plainly.
Patients classified as PD-L1 negative responded to PD-1 blockade.
Nivo showed it.
Keytruda showed it.
Combination checkpoint blockade showed it.
This was not an observation confined to one drug or one trial.
Did PD-L1-Negative Patients Achieve Complete Responses?
This question matters because tumour shrinkage and complete response are materially different outcomes.
In the PD-L1-negative subgroup reported from KEYNOTE-029, 50.0% responded to Keytruda + Ipi and 33.3% achieved a complete response.
One in three patients in this PD-L1-negative subgroup achieved a complete response.
That is not a peripheral observation. It is a consequential result that demands explanation and further pursuit.
The KEYNOTE-029 analysis was reported at a median follow-up of 36.8 months. Across the whole treatment group, the median duration of response had not been reached and the estimated proportion of responses still ongoing at 36 months was 84.2%.
Those durability figures describe the treatment group as a whole. They do not establish the durability specifically of the PD-L1-negative complete responses.
And the complete-response picture remains incomplete across the other PD-L1-negative groups.
For the PD-L1-negative patients receiving Nivo in CheckMate 067, 41.3% responded.
What I cannot establish from the published data I have found is how many of those PD-L1-negative patients achieved a complete response.
That missing number matters. A complete response is not the same outcome as tumour shrinkage, and a complete response itself does not tell us how long that response lasted.
If the subgroup CR result is not available, it should remain visibly missing. I will not replace it with the CR rate from the whole treatment arm and allow that number to appear to answer a different question.
What Happened Across the Full Trial Populations?
Now look separately at the broader treatment results.
These are whole-arm results, including patients classified as PD-L1 positive and PD-L1 negative where both were enrolled in that treatment arm. Some trial populations also included patients whose PD-L1 status was indeterminate, unevaluable or otherwise not classified.
These results provide important context, but they must not be mistaken for results specifically from PD-L1-negative patients.
| Trial | Treatment | Patients being measured | ORR | CR |
|---|---|---|---|---|
| CheckMate 067 | Nivo | All patients | 43.7% | 8.9% |
| KEYNOTE-006 | Keytruda | All patients, every 2 weeks | 33.7% | Not separately shown here |
| KEYNOTE-006 | Keytruda | All patients, every 3 weeks | 32.9% | Not separately shown here |
| KEYNOTE-006 | Ipi | All patients | 11.9% | Not separately shown here |
| CheckMate 067 | Nivo + Ipi | All patients | 57.6% | 11.5% |
| KEYNOTE-029 | Keytruda + Ipi | All patients | 62.1% | 27.5% |
| CheckMate 067 | Ipi | All patients | 19.0% | 2.2% |
| CheckMate 069 | Nivo + Ipi | BRAF wild-type cohort | 61% | 22% |
| CheckMate 069 | Ipi | BRAF wild-type cohort | 11% | 0% |
These trials should not be ranked against one another as though they were a single comparison. Different trials involved different populations, designs, dosing, assessment points and lengths of follow-up. The purpose of showing the results together is to make the evidence visible, not to declare a winning treatment.
The timing matters too.
For example, the original KEYNOTE-006 analysis had a median follow-up of 7.9 months, while the KEYNOTE-029 results shown here were reported at a median follow-up of 36.8 months.
A CR percentage tells us how many patients had reached complete response in that analysis. Follow-up tells us how much time had been available to observe what happened next.
What Does This Pattern Tell Us?
Now stand back from the individual trials and look at the pattern.
Patients classified as PD-L1 negative responded to Nivo.
Patients classified as PD-L1 negative responded to Keytruda.
PD-L1-negative patients also responded when PD-1 blockade was combined with Ipi.
And in the PD-L1-negative subgroup reported from KEYNOTE-029, one-third achieved a complete response.
So PD-L1 negative plainly did not mean that the PD-1 pathway was biologically irrelevant or that blocking it could not work.
That takes us back to the biology.
PD-L1 expression is dynamic. It can be induced as immune conditions change, including in response to interferon-gamma released during an active T-cell response.
A tumour classified as PD-L1 negative by a baseline biopsy is therefore not necessarily biologically incapable of expressing PD-L1. The biopsy tells us what was detected in the tissue sampled, using that assay and threshold, at that point in time.
A static test is being used to measure dynamic biology.
There is another limitation.
PD-1 has two established ligands: PD-L1 and PD-L2. A PD-L1 test measures PD-L1. It does not tell us whether PD-L2 is contributing to PD-1 signalling.
This matters when considering Nivo and Keytruda because both drugs block PD-1 itself. By blocking the receptor, they interrupt inhibitory signalling through PD-1 from both PD-L1 and PD-L2.
So a patient can be classified PD-L1 negative while the biology relevant to PD-1 blockade remains more complex than that classification suggests.
Biology missed by the baseline PD-L1 test, particularly dynamic PD-L1 induction, is more likely than not to explain a substantial proportion of the responses seen in patients classified as PD-L1 negative.
The size of that contribution has not been sufficiently followed through and quantified.
That is not evidence that the contribution is small. It identifies the unanswered question.
We need to determine how much of the observed response is explained by dynamically induced PD-L1, how much PD-L2 contributes, how much is hidden by spatial and sampling differences, and what other biology remains to be accounted for.
This leads to a larger question.
Have we mistaken the limitations of PD-L1 testing for limitations of PD-L1 biology?
If PD-L1 is dismissed because a static baseline test is an imperfect predictor, we risk dismissing the biology with the measurement.
The problem may not be PD-L1. The problem may be what we have asked a single PD-L1 test to tell us.
The question is no longer simply whether PD-L1 is a good biomarker.
The question is whether an incomplete measurement of dynamic checkpoint biology could leave patients who are capable of responding to PD-1 blockade unidentified.
Why Does This Matter So Much in Cholangiocarcinoma?
Surgery remains the established curative pathway for cholangiocarcinoma when the cancer can be completely removed.
Once that opportunity is no longer available, complete responses are rare.
But there is something we must not lose sight of: durable complete responses to checkpoint blockade have occurred in patients with advanced cholangiocarcinoma.
We do not need to turn those few patients into the many to make that observation important.
Nor should we describe the outcome as reproducible when cholangiocarcinoma does not yet provide the numbers to establish that.
We should preserve what the patients have already demonstrated.
Advanced cholangiocarcinoma can, in some patients, respond completely to checkpoint blockade, and complete responses can remain durable for years.
That is a demonstrated biological possibility.
The next question is not whether it has happened. It has.
The question is: what biology enabled it, and who else carries that biology?
That makes understanding PD-1, PD-L1, PD-L2 and the dynamic immune conditions surrounding them more than an academic biomarker question.
It is a question about identifying survival opportunity.
If an inadequate measurement causes us to underestimate the biology of checkpoint response, the consequence for a patient could be a missed objective response, a missed complete response, or in an exceptional responder, a missed durable complete response measured in years.
We do not inflate the few into the many. But neither do we allow the few to disappear because they are few.
These exceptional responders are not statistical inconvenience. They are evidence of what the biology has already shown itself capable of doing.
And that leaves the Patient-Led question that must remain in pursuit:
If static PD-L1 testing is failing to capture dynamic checkpoint biology, who are we failing to identify who could respond, achieve a complete response, or achieve a durable complete response?
What is known must remain known. What is missing must remain visible. And what the evidence demands must remain in pursuit.
What Does a PD-L1-Negative Biopsy Actually Prove?
This brings us back to the word at the centre of the problem:
Negative.
What does "PD-L1 negative" actually mean?
This is where we need to separate the measurement from the underlying biology.
A PD-L1 test usually begins with tumour tissue obtained during biopsy or surgery.
The laboratory can use immunohistochemistry (IHC) to look for PD-L1 protein in that tissue.
The result is interpreted using a particular assay, scoring method and threshold.
The sample may then be classified:
PD-L1 positive.
Or:
PD-L1 negative.
Before turning that classification into a conclusion about the tumour's biology, ask four simple questions.
What was measured?
Where was it measured?
When was it measured?
What could the test not see?
The test examined PD-L1 expression in the tissue available for testing, according to the assay, scoring method and threshold being used.
That is a real measurement.
But it is a bounded measurement.
It does not continuously observe the tumour.
It does not necessarily represent every part of the cancer.
And it cannot, by itself, tell us what that tumour environment might express later if the biology changes.
So a PD-L1-negative result establishes something real, but limited:
PD-L1 was not detected at the level required for a positive classification in the tissue examined, using that test, scoring method and threshold, at that point in time.
It does not establish:
This tumour is biologically incapable of expressing PD-L1.
Those are different claims.
PD-L1 negative is a test classification. It is not a biological identity.
A test is a measurement of biology. It is not the biology itself.
Or more simply:
A biopsy is a photograph. Biology is a movie.
How Dynamic PD-L1 Expression Changes During an Immune Response
Now the photograph problem becomes more important.
PD-L1 expression can change.
PD-L1 is not simply something cancer invented.
It is part of normal immune regulation.
During an immune response, the body needs ways to restrain immune activity and help limit damage to normal tissue.
Cancer can benefit from that same protective biology.
Picture a tumour-reactive T cell recognising and engaging a cancer cell.
An activated T cell can release a signalling molecule called interferon-gamma (IFN-γ).
Think of interferon-gamma as a message travelling through the tumour environment:
The immune system is here and active.
Cells receiving that signal can respond by increasing PD-L1 expression.
PD-L1 can then engage PD-1 on T cells and contribute to restraining their activity.
This is one recognised form of adaptive PD-L1 expression, sometimes described as adaptive immune resistance.
The sequence can be pictured simply:
T-cell activity → interferon-gamma signal → PD-L1 increases → PD-L1 engages PD-1 → T-cell activity is restrained.
The important point is that PD-L1 expression can therefore reflect what is happening during an immune interaction.
The cancer did not invent the brake. It can benefit from a normal biological brake that already exists within us.
This changes how we need to think about a PD-L1 measurement.
A tumour environment showing little or no detectable PD-L1 at one moment is not necessarily being observed under the same biological conditions at another moment.
PD-L1 can be a dynamic expression of biology, while the biopsy remains a static measurement of one moment within it.
Can a Static PD-L1 Test Represent Dynamic Tumour Biology?
Now put two established facts beside each other.
The biopsy is a static measurement.
PD-L1 expression can be dynamic.
Neither fact is controversial on its own.
But together they create an important question.
Imagine a biopsy shows little or no detectable PD-L1.
The laboratory may have measured the tissue accurately.
The result may correctly classify that sample as PD-L1 negative according to the assay, scoring method and threshold used.
The photograph may be perfectly accurate and still incomplete.
Because the test tells us what was detected in the tissue examined at that moment.
It cannot, by itself, tell us everything that the tumour environment is capable of expressing under different biological conditions.
So we need to keep two propositions separate:
PD-L1 was not detected at the level required for a positive classification.
And:
PD-L1 cannot be expressed.
The first can be established by the test.
The second does not logically follow from it.
Now return to the patients who were classified as PD-L1 negative and still responded to PD-1 blockade.
Their response does not prove that dynamic PD-L1 induction explains what happened.
But neither does their baseline negative test establish that dynamic PD-L1 expression could not have occurred.
That leaves us with a more consequential question than the baseline classification alone can answer:
What is this tumour environment capable of expressing when the biological conditions within it change?
That is the difference between measuring a moment and understanding a biological system.
THE COLD-TUMOUR QUESTION
What Does PD-L1 Negative Mean in a Cold Tumour?
If cholangiocarcinoma commonly begins immune-cold, what checkpoint biology is actually present before treatment, and what changes once treatment begins?
Cholangiocarcinoma is commonly characterised as an immune-cold cancer.
A cold tumour typically contains fewer tumour-infiltrating CD8 T cells and less active anti-tumour immune engagement. By contrast, immune-hot cholangiocarcinomas show greater CD8 T-cell infiltration, increased interferon-gamma activity and increased expression of checkpoint molecules including PD-1 and PD-L1.
That relationship matters.
Research in human intrahepatic cholangiocarcinoma has shown that activated CD8 T cells and interferon-gamma can rapidly increase PD-L1 expression in cholangiocarcinoma cells.
So PD-L1 expression can be partly a consequence of an immune interaction already taking place.
THE OBSERVATION
If the tumour is cold before treatment, with little active T-cell engagement and little adaptive PD-L1 expression, a baseline biopsy may be measuring the tumour before the checkpoint biology we are looking for has been fully induced.
This creates a particularly important question in cholangiocarcinoma because first-line treatment commonly includes Durva.
Durva is a checkpoint inhibitor. Nivo and Keytruda are also checkpoint inhibitors.
But that does not make them the same treatment.
AN IMPORTANT DISTINCTION
They share a treatment class. They do not block the same checkpoint biology.
The words “checkpoint inhibitor” describe a family of treatments. Within that family, the molecular target matters.
NIVO / KEYTRUDA
Block PD-1, the receptor on the T cell.
PD-L1 → PD-1: BLOCKED
PD-L2 → PD-1: BLOCKED
DURVA
Blocks PD-L1, one of the ligands.
PD-L1 → PD-1: BLOCKED
PD-L2 → PD-1: REMAINS AVAILABLE
IPI
Blocks CTLA-4, a different checkpoint.
Combined CTLA-4 and PD-1 blockade can enhance T-cell function beyond either pathway alone.
This distinction is established biology. Nivo and Keytruda block PD-1 itself and therefore prevent inhibitory signalling through both PD-L1 and PD-L2. Durva blocks PD-L1, leaving PD-L2 able to signal through PD-1.
Now return to the cold tumour.
If little PD-L1 is present at the beginning of treatment, then the amount of PD-L1 available for Durva to block is correspondingly limited at that moment.
Nivo and Keytruda enter the biology differently.
They bind PD-1 itself. Their target is therefore not dependent on PD-L1 first being expressed as the ligand carrying the inhibitory signal. They also prevent PD-L2 from signalling through PD-1.
If PD-1 blockade successfully restores activity in responsive tumour-reactive T cells, T-cell function and cytokine production can increase. Increased interferon-gamma can then induce PD-L1 expression.
This raises an apparent paradox worth investigating:
Could checkpoint treatment help create the immune conditions in which PD-L1 becomes expressed, even though the tumour was classified PD-L1 negative before treatment?
THE PATIENT-LED PROBLEM
When “they are both checkpoint inhibitors” becomes incorrect treatment advice.
I regularly speak with cholangiocarcinoma patients who are receiving, or have progressed on, Durva and want to understand whether Nivo or Keytruda should be investigated.
Some tell me they have been advised that there would be little point because Durva, Nivo and Keytruda are essentially the same treatment.
At the level of mechanism, that assertion is incorrect.
They belong to the same broad family of checkpoint inhibitors, but they do not block the same target.
Durva blocks PD-L1. Nivo and Keytruda block PD-1 itself, preventing inhibitory signalling through both PD-L1 and PD-L2.
That difference does not prove that a patient who progresses on Durva will respond to Nivo or Keytruda.
But it does mean that the possibility should not be dismissed simply because the treatments have all been placed beneath the label “checkpoint inhibitor”.
A treatment class is not a mechanism.
WHAT THE PATIENT NEEDS TO UNDERSTAND
Progression on PD-L1 blockade is not, by itself, evidence that blocking PD-1 would be biologically identical.
The clinical question remains unanswered: does that mechanistic difference create a meaningful treatment opportunity for an individual cholangiocarcinoma patient after Durva?
That requires evidence, not assumption.
But understanding the difference changes the conversation. Instead of asking simply whether another “checkpoint inhibitor” might work, the patient can ask a much more precise question:
THE RIGHT QUESTION
My tumour has progressed on PD-L1 blockade. Given that PD-1 blockade targets the receptor itself and also prevents PD-L2 signalling through PD-1, what evidence in my tumour supports, or argues against, investigating PD-1 blockade?
THE QUESTIONS THIS EXPOSES
If a cholangiocarcinoma is truly immune-cold and expressing little PD-L1, how much PD-L1-mediated inhibition is available for Durva to block at the beginning of treatment?
Does blocking PD-1 itself create a different biological opportunity because it prevents inhibitory signalling through both PD-L1 and PD-L2?
If PD-1 blockade restores T-cell activity, does the resulting increase in interferon-gamma induce PD-L1 that was absent, low or undetected before treatment?
Does the tumour therefore move from one checkpoint environment before treatment to another after treatment has begun?
What happens to PD-L2 during that change?
Are PD-1 blockade and PD-L1 blockade clinically equivalent in that changing environment, or have we assumed equivalence because both treatments sit beneath the same “checkpoint inhibitor” label?
And when a patient progresses on Durva, what evidence should determine whether PD-1 blockade is investigated next?
WHAT THE EVIDENCE DOES NOT YET ESTABLISH
The mechanistic difference does not establish that Nivo or Keytruda is superior to Durva in cholangiocarcinoma, or that switching from PD-L1 blockade to PD-1 blockade will benefit an individual patient. Those are clinical questions. The established point is narrower, but important: these treatments do not block identical checkpoint biology.
THE CONSEQUENTIAL QUESTION
Are we testing an immune-cold tumour before treatment, calling it PD-L1 negative, and then failing to investigate what checkpoint biology emerges once treatment changes the immune environment?
Why "Not Proven" Is Not Proof of the Opposite
We have now reached the point where the unanswered question must remain attached to the evidence.
Before going further, we need to separate what is established, what I infer from that evidence, and what remains unknown.
Substantial numbers of patients classified as PD-L1 negative have responded to PD-1 blockade.
PD-L1 expression can be dynamic.
Immune activity, including interferon-gamma signalling, can induce PD-L1 expression.
A biopsy measures tissue from a particular place at a particular point in time.
PD-1 has another established ligand, PD-L2, which a PD-L1 test does not measure.
Biology missed by the baseline PD-L1 test, particularly dynamic PD-L1 induction, is more likely than not to explain a substantial proportion of the responses seen in patients classified as PD-L1 negative.
How much of the observed response is explained by dynamically induced PD-L1, how much PD-L2 contributes, how much is hidden by spatial and sampling differences, and what other biology remains to be accounted for.
The response of PD-L1-negative patients does not, by itself, prove which mechanism caused their response.
But a baseline PD-L1-negative result does not prove that the tumour environment was biologically incapable of expressing PD-L1 later.
Nor does a PD-L1-negative result establish that PD-1 signalling through PD-L2 was absent.
That boundary matters.
Not known does not mean not true.
Not proven is not proof of the opposite.
And not investigated proves neither.
My informed assertion is that biology missed by the baseline PD-L1 test, particularly dynamic PD-L1 induction, is more likely than not to explain a substantial proportion of these responses.
What has not been established is how large that contribution is.
We still need to determine how much is explained by dynamically induced PD-L1, how much PD-L2 contributes, how much is hidden by spatial and sampling differences, and what other biology remains to be accounted for.
There are competing explanations.
Different parts of the same tumour can differ. A biopsy can miss an area where PD-L1 is present. Assays, scoring methods and thresholds matter. PD-L2 provides another route into PD-1 signalling that a PD-L1 test does not measure. Other biology within the tumour immune environment may also contribute.
Those possibilities do not make the dynamic question disappear.
They tell us what needs to be measured.
The uncertainty is in how much each mechanism contributes. It is not in whether this biology exists.
Those are not the same uncertainty, and patients should not have them presented as though they are.
The proof points establish that the question exists. They do not establish how much each mechanism explains.
An unanswered question must remain attached to the evidence when its answer could change the meaning of what we think we already know.
Could T-Cell Exhaustion Help Create a PD-L1-Negative Baseline?
There is another part of the biology worth examining.
T cells fighting cancer can become exhausted or dysfunctional, and they can also be excluded from parts of a tumour.
This matters because adaptive PD-L1 expression depends, in part, on what is happening within the immune environment.
When tumour-reactive T cells are actively engaged, they can release interferon-gamma.
That signalling can increase PD-L1 expression within the tumour environment.
Now consider the opposite biological state.
What if, in the tissue being sampled at baseline, tumour-reactive T cells are exhausted, dysfunctional, excluded, absent or simply not effectively engaging the tumour?
There may then be less active interferon-gamma signalling and therefore less adaptive PD-L1 expression for the biopsy to detect.
The pathology result may correctly classify that tissue as:
PD-L1 negative.
But the biological question is different.
Could exhaustion, dysfunction or exclusion help create the baseline conditions in which adaptive PD-L1 expression is low or absent?
If so, the baseline biopsy may be accurately describing the immune conditions present at that moment without describing what the tumour environment is capable of expressing when those conditions change.
This does not establish that T-cell exhaustion, dysfunction or exclusion caused a particular tumour to test PD-L1 negative.
It identifies another testable part of the biological chain.
T-cell state → interferon-gamma signalling → adaptive PD-L1 expression → what the baseline biopsy can detect.
The question is therefore not whether the baseline result was wrong.
The question is what biological conditions produced that result, and whether those conditions remain the same once treatment begins.
How Could PD-L1-Negative Responses to PD-1 Blockade Be Tested?
The patient group already exists.
Patients classified as PD-L1 negative who nevertheless respond to PD-1 blockade.
So follow the biology.
Start with the baseline PD-L1-negative tumour sample.
Then, in an appropriate research setting, examine the tumour again early during PD-1 blockade and compare what has changed.
The investigation can be pictured simply:
baseline biopsy
blockade
tumour reassessment
the biology
Then ask:
Has T-cell presence within the tumour changed?
Has T-cell activity changed?
Has interferon-gamma signalling changed?
Has PD-L1 expression changed?
Where within the tumour have those changes occurred?
And most importantly:
Do those changes help explain why a patient classified as PD-L1 negative at baseline responded to PD-1 blockade?
The investigation should not be designed simply to prove one proposed explanation.
It should distinguish between competing explanations.
Was PD-L1 already present elsewhere in the tumour but missed by the original sample?
Did PD-L1 expression change during treatment?
Did the location or activity of tumour-reactive T cells change?
Did interferon-gamma signalling change?
Or does another part of the tumour immune environment better explain the response?
The purpose is not to make the evidence fit the assertion.
It is to make the competing explanations answer to the evidence.
If dynamic PD-L1 expression is part of the explanation, the investigation should be capable of detecting evidence for it.
If it is not, the investigation should be capable of showing that too.
That is how an unexplained clinical observation becomes a researchable biological question.
"We don't know" can be an accurate answer. But where the answer could change our understanding of a treatment opportunity, it should also be the beginning of the next investigation.
How Unanswered Cancer Questions Can Become Half-Truths
This is where the problem becomes larger than PD-L1.
A scientific study can successfully answer the question it was designed to answer.
But its results can also expose another question that the study was not designed to resolve.
What happens to that second question matters.
Scientific findings travel.
They move through papers, conferences, guidelines, clinical practice and eventually into conversations with patients.
During that journey, complicated findings have to be summarised.
The answered part becomes easier to carry forward.
The unanswered part can be left behind.
That does not require bad science or bad medicine.
It can happen because complex knowledge is compressed as it moves through a system.
But that compression can have a consequence.
A bounded scientific finding can gradually begin to sound like a complete biological fact.
What began as:
"We don't yet know."
can eventually be heard as:
"It isn't so."
Those statements are not equivalent.
I have sat in a cancer conference and heard experienced oncologists, surgeons and radiologists confronted with the question of why patients classified as PD-L1 negative can respond to PD-1 blockade.
The answer was essentially:
We don't know.
I do not see that as a failure of those clinicians.
I see it as a reason to ask what happened to the unanswered question on its journey from biological discovery, through clinical trials, and into clinical knowledge.
The trial result travelled.
Did the unresolved question travel with it?
A scientist may reasonably ask:
Did the study answer the question it was designed to answer?
A clinician may reasonably ask:
What does the available evidence support me doing for this patient?
Patient-Led reasoning must also ask:
What did the evidence uncover that remains unresolved and could change my opportunity for survival?
These are not competing questions.
They are different responsibilities looking at the same evidence.
But only one person carries the whole consequence when an important unanswered question is lost.
The patient.
When an unanswered observation could change a patient's opportunity to cure or extend survival, the unanswered question must travel with the answer until it is resolved, disproven, or shown not to matter.
An unanswered question is not an inconvenience attached to the evidence.
When its answer could change the consequence for the patient, it remains part of what needs to be understood.
Could the Unanswered PD-L1 Question Be Bigger Than the Original Answer?
Checkpoint blockade answered an enormous question.
The immune system could be released from inhibitory checkpoints and produce meaningful, sometimes extraordinary, anti-cancer responses.
That discovery changed cancer treatment.
But a breakthrough answer does not make the unanswered observations inside it less important.
Patients classified as PD-L1 negative responded to PD-1 blockade.
Substantial numbers of them responded.
We know PD-L1 expression can be dynamic.
We know immune signalling can induce PD-L1 expression.
We know the immune state within a tumour can change.
And we know a biopsy is a bounded measurement of that biology at a particular place and time.
None of this establishes the mechanism responsible for the PD-L1-negative responder group.
But the proof points establish that the question exists. They do not establish the answer.
The unresolved biological question is:
In patients classified as PD-L1 negative who nevertheless respond to PD-1 blockade, what biological mechanism enables that response?
And within that question is a specific explanation that should be tested:
Is dynamic PD-L1 expression part of the explanation, and if so, what is a static baseline biopsy failing to show us?
Following this question through leads somewhere larger than PD-L1.
It is the relationship between a measurement and the biological reality we allow that measurement to represent.
If a static measurement is allowed to become a biological identity, we need to ask where else in cancer medicine the same thing may be happening.
What was measured?
What was observed?
What remains unexplained?
What could explain it?
Has that explanation actually been tested?
And when the answer could change a patient's opportunity to cure or extend survival:
Has the unanswered question travelled with the answer?
A test is a measurement of biology. It is not the biology itself.
A biopsy is a photograph. Biology is a movie.
The question a study sets out to answer is not necessarily the most important question its results uncover.
And when a consequential question emerges from the evidence, it should remain visible until it is answered, disproven, or shown not to matter.
That is not a challenge to science.
It is a Patient-Led requirement that the investigation follows the consequence all the way through.
It is the patient who bears the consequence. Therefore, it is the patient's benefit that must govern every decision.
Cure. Extend. Prevent.
Steve Holmes
ABOUT THE AUTHOR
Steve Holmes
Steve Holmes is Founder and CEO of Cholangiocarcinoma Foundation Australia. His own journey through cholangiocarcinoma included 25 hours of major surgery, a near-fatal ruptured aneurysm, Stage IV recurrence, and ultimately a complete response after matching to a new checkpoint immunotherapy trial.
Today, Steve puts that lived experience to work, turning it into expertise that connects scientific evidence with patient understanding and action, focused on three outcomes: Cure. Extend Survival. Prevent.
ABOUT CHOLANGIO TODAY
Understanding changes what happens next.
A cholangiocarcinoma diagnosis is a hole. Late or no Understanding deepens it. Earlier Understanding is the ladder out.
That ladder is built from lived expertise, experience put to work, examined against evidence, science and what has been learned from those who have travelled before. Each rung becomes a Right Action the patient can take.
Through a Patient-Led lens, Steve Holmes follows Understanding wherever it leads, across cognition, biology, physiology, medicine, science, technology and lived experience. There are no disciplinary boundaries when the answer could change what happens next for the patient.
Earlier Understanding creates more curative opportunity, extends life and prevents tomorrow's diagnoses. Earlier Right Action delivers it.
CURE MORE. EXTEND LIFE. PREVENT MORE.
References & Source Material
The clinical results, biological mechanisms and Patient-Led questions examined in this investigation are grounded in the original clinical trials, foundational checkpoint research, cholangiocarcinoma biology and current drug-mechanism evidence below.
What Happened to the Patients?
These sources support the response, complete-response and durability evidence that first exposes the PD-L1-negative responder question.
- Larkin J, Chiarion-Sileni V, Gonzalez R, et al. Combined Nivolumab and Ipilimumab or Monotherapy in Untreated Melanoma. New England Journal of Medicine. 2015;373:23-34. CheckMate 067. Includes the 41.3% objective response rate among patients classified as PD-L1 negative who received nivolumab, together with the broader nivolumab, nivolumab plus ipilimumab and ipilimumab treatment-arm results. DOI: 10.1056/NEJMoa1504030
- Robert C, Long GV, Brady B, et al. Nivolumab in Previously Untreated Melanoma without BRAF Mutation. New England Journal of Medicine. 2015;372:320-330. CheckMate 066. Reports an objective response rate of 33.1% among nivolumab-treated patients with negative or indeterminate PD-L1 status. DOI: 10.1056/NEJMoa1412082
- Robert C, Schachter J, Long GV, et al. Pembrolizumab versus Ipilimumab in Advanced Melanoma. New England Journal of Medicine. 2015;372:2521-2532. KEYNOTE-006. Establishes the original phase 3 pembrolizumab versus ipilimumab treatment results used as context in this investigation. DOI: 10.1056/NEJMoa1503093
- Schachter J, Ribas A, Long GV, et al. Pembrolizumab versus Ipilimumab for Advanced Melanoma: Final Overall Survival Results of a Multicentre, Randomised, Open-Label Phase 3 Study (KEYNOTE-006). The Lancet. 2017;390(10105):1853-1862. Longer follow-up from KEYNOTE-006, supporting the continuing assessment of response and survival after pembrolizumab. DOI: 10.1016/S0140-6736(17)31601-X
- Robert C, Ribas A, Schachter J, et al. Pembrolizumab versus Ipilimumab in Advanced Melanoma (KEYNOTE-006): Post-hoc 5-Year Results. Lancet Oncology. 2019;20(9):1239-1251. Five-year follow-up, demonstrating why duration of follow-up must be distinguished from response status recorded at an earlier assessment. DOI: 10.1016/S1470-2045(19)30388-2
- Postow MA, Chesney J, Pavlick AC, et al. Nivolumab and Ipilimumab versus Ipilimumab in Untreated Melanoma. New England Journal of Medicine. 2015;372:2006-2017. CheckMate 069. Supports the nivolumab plus ipilimumab response evidence, including responses among patients classified as PD-L1 negative. DOI: 10.1056/NEJMoa1414428
- Long GV, Atkinson V, Cebon JS, et al. Standard-Dose Pembrolizumab Plus Alternate-Dose Ipilimumab in Advanced Melanoma: KEYNOTE-029 Cohort 1C, Long-Term Results. Clinical Cancer Research. Among the reported PD-L1-negative subgroup, 50.0% achieved an objective response and 33.3% achieved a complete response. The broader cohort also provides longer-term response-duration context. DOI: 10.1158/1078-0432.CCR-20-0177
- Eisenhauer EA, Therasse P, Bogaerts J, et al. New Response Evaluation Criteria in Solid Tumours: Revised RECIST Guideline (Version 1.1). European Journal of Cancer. 2009;45(2):228-247. RECIST 1.1. Provides the standard response definitions underlying partial response, complete response and progressive disease. DOI: 10.1016/j.ejca.2008.10.026
What Biology Could Explain the Observation?
These sources establish adaptive and dynamic PD-L1 biology, on-treatment immune change, PD-L2 biology and the molecular distinction between blocking PD-1 and blocking PD-L1.
- Taube JM, Anders RA, Young GD, et al. Colocalization of Inflammatory Response with B7-H1 Expression in Human Melanocytic Lesions Supports an Adaptive Resistance Mechanism of Immune Escape. Science Translational Medicine. 2012;4(127):127ra37. Important evidence linking tumour immune activity, interferon-gamma signalling and adaptive PD-L1 expression. DOI: 10.1126/scitranslmed.3003689
- Spranger S, Spaapen RM, Zha Y, et al. Up-Regulation of PD-L1, IDO, and Tregs in the Melanoma Tumor Microenvironment Is Driven by CD8+ T Cells. Science Translational Medicine. 2013;5(200):200ra116. Supports the relationship between active tumour-reactive CD8+ T cells and adaptive changes in the tumour immune environment, including PD-L1. DOI: 10.1126/scitranslmed.3006504
- Vilain RE, Menzies AM, Wilmott JS, et al. Dynamic Changes in PD-L1 Expression and Immune Infiltrates Early During Treatment Predict Response to PD-1 Blockade in Melanoma. Clinical Cancer Research. 2017;23(17):5024-5033. Paired biopsies before and early during pembrolizumab or nivolumab treatment demonstrated dynamic changes in PD-L1 and immune-cell infiltration. Early on-treatment biology provided information not captured by the pretreatment PD-L1 measurement alone. DOI: 10.1158/1078-0432.CCR-16-0698
- Yearley JH, Gibson C, Yu N, et al. PD-L2 Expression in Human Tumors: Relevance to Anti-PD-1 Therapy in Cancer. Clinical Cancer Research. 2017;23(12):3158-3167. Establishes PD-L2 as another ligand of PD-1. PD-L2 was detected without PD-L1 in subsets of tumours and was associated with pembrolizumab response independently of PD-L1 in the studied head-and-neck cancer population. DOI: 10.1158/1078-0432.CCR-16-1761
- Current nivolumab prescribing information. Nivolumab binds PD-1 and blocks its interaction with both PD-L1 and PD-L2. This establishes the molecular distinction used in this investigation between PD-1 blockade and PD-L1 blockade. Nivolumab prescribing information
- Current pembrolizumab prescribing information. Pembrolizumab binds PD-1 and blocks its interaction with PD-L1 and PD-L2. Pembrolizumab prescribing information
- Current durvalumab prescribing information. Durvalumab binds PD-L1 and blocks PD-L1 interactions with PD-1 and CD80. The prescribing information also recognises that PD-L1 expression can be induced by inflammatory signals including interferon-gamma. Because durvalumab targets PD-L1 rather than PD-1, it does not block PD-L2 from binding PD-1. Durvalumab prescribing information
- Leach DR, Krummel MF, Allison JP. Enhancement of Antitumor Immunity by CTLA-4 Blockade. Science. 1996;271(5256):1734-1736. Foundational experimental evidence showing that blockade of CTLA-4 could enhance anti-tumour immune responses. DOI: 10.1126/science.271.5256.1734
- Ishida Y, Agata Y, Shibahara K, Honjo T. Induced Expression of PD-1, a Novel Member of the Immunoglobulin Gene Superfamily, upon Programmed Cell Death. EMBO Journal. 1992;11(11):3887-3895. The original report from Tasuku Honjo's laboratory describing PD-1. DOI: 10.1002/j.1460-2075.1992.tb05481.x
- The Nobel Assembly at Karolinska Institutet. The 2018 Nobel Prize in Physiology or Medicine: Discovery of Cancer Therapy by Inhibition of Negative Immune Regulation. Scientific background to the Nobel Prize awarded jointly to James P. Allison and Tasuku Honjo for their independent checkpoint discoveries. Nobel Prize scientific background
What Does the Evidence Show in Cholangiocarcinoma?
These sources bring the investigation into cholangiocarcinoma: the immune-cold tumour environment, inducible PD-L1 biology, limitations of baseline PD-L1 measurement, treatment-induced immune change, the distinction between PD-1 and PD-L1 treatment strategies, and the clinical evidence showing what checkpoint treatment has already achieved in patients.
- Loeuillard E, Conboy CB, Gores GJ, Ilyas SI. Immunobiology of Cholangiocarcinoma. JHEP Reports. 2019;1(4):297-311. Describes immune-hot and immune-cold cholangiocarcinoma. T-cell-infiltrated or immune-hot CCA is characterised by greater CD8+ T-cell infiltration, interferon-gamma activity and increased checkpoint molecules including PD-1 and PD-L1. Immune-cold CCA lacks this T-cell-rich environment and contains prominent immunosuppressive components. This provides the biological foundation for asking what a baseline PD-L1-negative result means in an immune-cold tumour. DOI: 10.1016/j.jhepr.2019.06.003
- Gani F, Nagarajan N, Kim Y, et al. Programmed Death Ligand 1 Expression in Human Intrahepatic Cholangiocarcinoma and Its Association with Prognosis and CD8+ T-Cell Immune Responses. Human intrahepatic cholangiocarcinoma tissue and experimental work examining the relationship between PD-L1 and CD8+ T-cell responses. Human iCCA cell lines exposed to interferon-gamma showed increased PD-L1 expression, directly supporting the article's question about inducible PD-L1 biology in CCA. PubMed: 30323667
- Mody K, Starr J, Saul M, et al. Patterns and Genomic Correlates of PD-L1 Expression in Patients with Biliary Tract Cancers. Journal of Gastrointestinal Oncology. 2019;10(6):1099-1109. Examined 652 biliary tract cancers using a defined PD-L1 assay and threshold. PD-L1 positivity was 8.6% overall, including 7.3% in intrahepatic cholangiocarcinoma and 5.2% in extrahepatic cholangiocarcinoma. These percentages belong to that assay and threshold and should not be treated as a universal biological prevalence. DOI: 10.21037/jgo.2019.08.08
- Phase II pembrolizumab plus CAPOX study in advanced biliary tract carcinoma. Included paired tumour biopsies, PD-L1 and immune-infiltrate assessment, together with transcriptomic and genomic analyses. The study demonstrated the feasibility and value of examining how BTC immune biology changes during treatment rather than relying only on a pretreatment sample. PubMed: 35274717
- Phase Ib paired-biopsy study of anti-TIM-3 plus anti-PD-1 treatment in advanced biliary tract cancer. Exploratory paired tumour biopsies showed increased intratumoral CD8 T-cell density and upregulation of gene signatures related to interferon-gamma signalling, antigen presentation and T-cell activation during treatment. These changes were not shown to correlate clearly with efficacy, but they demonstrate that the immune environment in BTC can change after checkpoint- based treatment begins. PubMed: 41825935
- Kim RD, Chung V, Alese OB, et al. A Phase 2 Multi-institutional Study of Nivolumab for Patients With Advanced Refractory Biliary Tract Cancer. JAMA Oncology. 2020;6(6):888-894. Prospective phase 2 evidence of PD-1 blockade with nivolumab in advanced refractory biliary tract cancer. Important clinical evidence that PD-1 blockade has measurable activity in BTC outside the melanoma evidence that generated the original PD-L1-negative responder question. DOI: 10.1001/jamaoncol.2020.0930
- Klein O, Kee D, Nagrial A, et al. Evaluation of Combination Nivolumab and Ipilimumab Immunotherapy in Patients With Advanced Biliary Tract Cancers: Subgroup Analysis of the CA209-538 Trial. JAMA Oncology. 2020. Phase 2 evidence examining combined PD-1 and CTLA-4 blockade in advanced biliary tract cancers, including intrahepatic cholangiocarcinoma and gallbladder cancer. DOI: 10.1001/jamaoncol.2020.2814
- Oh D-Y, He AR, Qin S, et al. Durvalumab plus Gemcitabine and Cisplatin in Advanced Biliary Tract Cancer. NEJM Evidence. 2022;1(8). TOPAZ-1. Phase 3 evidence establishing the clinical benefit of adding the PD-L1 inhibitor durvalumab to gemcitabine and cisplatin in previously untreated advanced biliary tract cancer. This is the principal clinical context for the article's examination of PD-L1 blockade in first-line BTC. DOI: 10.1056/EVIDoa2200015
- Kelley RK, Ueno M, Yoo C, et al. Pembrolizumab in Combination with Gemcitabine and Cisplatin Compared with Gemcitabine and Cisplatin Alone for Patients with Advanced Biliary Tract Cancer (KEYNOTE-966). The Lancet. 2023. Phase 3 evidence establishing clinical benefit from adding the PD-1 inhibitor pembrolizumab to gemcitabine and cisplatin in previously untreated unresectable or metastatic biliary tract cancer. TOPAZ-1 and KEYNOTE-966 establish activity for two different checkpoint strategies in BTC. They are not a head-to-head comparison and do not establish superiority of PD-1 over PD-L1 blockade or the reverse. PubMed: 37075781
- Tanaka S, Umemoto K, Kubo S, et al. Nivolumab for Treating Patients with Occupational Cholangiocarcinoma. Journal of Hepato-Biliary-Pancreatic Sciences. 2022;29:1153-1155. Documents complete responses to nivolumab in recurrent occupational cholangiocarcinoma. The report describes an earlier patient whose complete response persisted for 26 months after nivolumab was discontinued, followed by two additional nivolumab-treated patients who also achieved complete response. These observations establish biological possibility. They do not establish that the outcome is reproducible across cholangiocarcinoma. DOI: 10.1002/jhbp.1215
- Nagrial A, et al. Nivolumab and Ipilimumab Combination Treatment in Patients with Advanced Intrahepatic Cholangiocarcinoma and Gallbladder Cancer: Results from the Phase II MoST-CIRCUIT Trial. Clinical Cancer Research. 2026;32(15):3203-3212. Contemporary phase 2 evidence of combined PD-1 and CTLA-4 blockade in 60 patients with advanced iCCA or gallbladder cancer. The overall objective response rate was 12%, including a 2% complete-response rate. Thirteen patients had previously received durvalumab. The trial demonstrates continued clinical investigation of checkpoint strategies in BTC, but does not establish that switching from PD-L1 blockade to PD-1 blockade benefits an individual patient. DOI: 10.1158/1078-0432.CCR-25-4009
What is known must remain known. What is missing must remain visible.
The evidence establishes that cholangiocarcinoma can be immune-cold, that interferon-gamma can induce PD-L1 in cholangiocarcinoma cells, that checkpoint-based treatment can alter the tumour immune environment, and that PD-1 and PD-L1 inhibitors do not block identical checkpoint biology.
What the evidence does not yet establish is how much dynamic PD-L1 induction explains PD-L1-negative response in cholangiocarcinoma, how much PD-L2 contributes, or whether switching from PD-L1 blockade to PD-1 blockade creates a clinically meaningful opportunity for an individual patient.
We do not inflate the few into the many. But neither do we allow the few to disappear because they are few.