To PD-L1 or Not to PD-L1: That Is the Question

These days, it seems that I write mostly about immune checkpoint blockade drugs, or some other new immunotherapy treatment for cancer. This post is no different—it covers PD-L1, a protein that is at the center of clinical decisions for selecting patients who are likely to benefit from treatment with an anti-PD-1 or anti-PD-L1 drug.

The U.S. Food and Drug Administration (FDA) has approved anti-PD-1 drugs for melanoma (nivolumab and pembrolizumab) and squamous non-small cell lung cancer (NSCLC—nivolumab). Along with the PD-L1 antibody atezolizumab, these drugs are expected to be FDA approved for some other cancers fairly soon. These treatments are superior to chemotherapy, but still, they do not work for all patients.

Expression of the protein PD-L1 on tumors has been thought to be a good predictor of a patient’s response. However, none of the three FDA approvals mention expression of PD-L1 as a prerequisite for prescribing these drugs. Attempts to figure out the prognostic significance of PD-L1 are at their peak, though it is not yet clear what lies beyond. Nonetheless, it is of obvious importance: there is no compelling reason to give these very expensive drugs to patients who are not likely to benefit from them.

Here is a very brief history of this complicated connection: in 2012, the very first sensational publication reported on the activity of nivolumab in three different cancers. The study looked at PD-L1 expression in about 15% of patients who participated in the clinical trial. Of 17 patients with PD-L1-negative (no PD-L1 detected) tumors, none had a measurable response, whereas 9 of 25 patients (36%) with PD-L1-positive tumors responded. This report certainly put PD-L1 at the center stage of anti-PD-1 development, but it was only the beginning.

Numerous following trials have reported that the efficacies of pembrolizumab (Keytruda) and nivolumab (Opdivo) indeed correlate with expression of PD-L1, but only to a degree. For example, in the KEYNOTE-001 trial, patients whose lung tumors had high levels of the PD-L1 protein had a response rate of 45%, but 10% patients whose tumors tested negative for PD-L1 also responded.

Even the anti-PD-L1 drugs in development, atezolizumab (previously known as MPDL3280A) and durvalumab (MEDI4736), do not show a strict dependence on the presence of their target PD-L1 in tumors, as would have been expected. Across several tumor types, responses to these drugs have been observed more often in PD-L1-positive tumors, but also in a lower proportion of PD-L1-negative ones.

Three years later, there is still no consensus regarding the value of PD-L1 expression in NSCLC as a predictive marker of response. Across the board, positivity for PD-L1, while predictive of response, does not always guarantee it, and lack of PD-L1 does not exclude response. Obviously, one must conclude that PD-L1 is just one of several (many?) factors that determine the likelihood of response. To compound the PD-L1 problem, there is a lack of defined criteria about what ‘PD-L1-positive tumor’ really means:

  1. The criteria used to define the PD-L1 status of a tumor biopsy differ vastly. In trials of nivolumab, the cutoff used is 1% to 5% of positive cells in a biopsy. For pembrolizumab, the cutoff is 1% to 50%! And for atezolizumab it is 1% to 10%.
  2. Different companies in the checkpoint drug business use PD-L1 reagents from different commercial sources to measure its levels and, without doubt, these different reagents have different sensitivities.
  3. Possibly the most confounding factor is that PD-L1 expression in tumors is not stable: not only can it be confined to different parts of tumors, but it can appear and disappear, depending on treatments and other poorly understood influences.

Leading clinicians agree that they would give anti-PD-1 drugs to patients with metastatic lung cancer or melanoma regardless of PD-L1 status, according to FDA guidelines. At the same time, many ongoing clinical trials testing these drugs group patients based on PD-L1 expression, or simply exclude patients without PD-L1 expression in their tumors.

Right now, PD-L1 is just not a good enough biomarker to exclude patients from receiving an anti-PD-1 drug, because even PD-L1-negative patients derive more benefit from PD-1 blockade than from docetaxel (a standard chemotherapy drug). If patients with PD-L1-negative NSCLC have a 10% to 12% chance of a meaningful response, they may be strongly inclined to receive a PD-1 or a PD-L1 blocker rather than going for a harsh conventional chemotherapy that may provide only a short-lived response.

One more problem is that most trials with checkpoint blockers only accept PD-L1-positive patients. This not only deprives patients from potentially life-prolonging treatment, but precludes a meaningful analysis of what determines the possibility of response in patients with PD-L1-negative tumors.

A recent study conducted a meta-analysis of 20 trials (1,475 patients) with nivolumab, pembrolizumab, and atezolizumab in NSCLC, melanoma, and genitourinary cancers. It looked at correlations between the presence of PD-L1 on tumors and patients’ responses. Significant differences were seen in NSCLC and melanoma, but not genitourinary cancers. Ignoring the different cutoffs for PD-L1 expression in different trials, responses were observed in 34.1% of PD-L1-positive and 19.9% of PD-L1-negative patients who were treated with nivolumab and pembrolizumab. This difference is obviously significant, but not tremendous.

Excluding PD-L1-negative patients from trials of immune checkpoint inhibitors appears to be a fallacy. Ignoring many other possibilities, it is entirely probable that because of the shortcomings of reagents, and the elusive nature of PD-L1, many PD-L1-negative tumors may be ‘false-negative‘; they really do express enough PD-L1 to respond to immune checkpoint drugs.