Skip to main content
Premium Trial:

Request an Annual Quote

Immunotherapy Developers, Clinical Researchers Continue to Seek New Dx Strategies

Premium

NEW YORK (GenomeWeb) – As recognition has grown that existing immunotherapy biomarkers like PD-L1, microsatellite instability, and tumor mutational burden predict only some of the response variability seen in this new and highly popular class of drugs, pharma firms and independent researchers are broadening their attention, hoping to capture new features of the complex interaction between a cancer and the human immune system that can guide which patients to treat with which therapies.

Amongst new possibilities are updates to traditional cytology and immunohistochemistry that could make existing biomarkers like PD-L1 more predictive by narrowing down the cell types and compartments in which PD-L1 expression truly impacts immunotherapy response.

Investigators also continue to test a variety of gene expression signatures that, similarly, offer a way to glean a combined molecular picture of both the tumor and the host-response.

Both areas were highlighted in presentations at last week's annual meeting of the American Association for Cancer Research, amongst other efforts incorporating more peripheral protein markers, new methods for calculating microsatellite instability, and novel measures of tumor heterogeneity.

In one presentation, investigators from Bristol Myers Squibb shared data from their work exploring two different avenues — alternative methods of PD-L1 analysis and various gene expression signatures —using samples and response data from the CheckMate 032 phase 1/2 study of nivolumab and ipilimumab in patients with metastatic gastroesophageal cancer.

In a group of 40 study participants, the investigators first recorded tumor-only PD-L1 immunohistochemistry via the established Dako PD-L1 IHC 28-8 pharmDx assay. They then calculated a combined score by evaluating PD-L1 expression from the same slides with a new algorithm that assessed both tumor and immune cells.

According to Ming Lei, who presented the results, when she and her coauthors compared the two IHC approaches with patient outcome data from a median of about two years follow-up, the combined PD-L1 score method correlated significantly better with efficacy, and at a higher cutoff, than tumor-only measurement.

Using RNA sequencing in the same group, investigators also measured several gene expression signatures, including two immune cell activation/infiltration signatures, a Bristol-Myers Squibb-developed inflammatory signature, as well as a genomic readout of PD-L1 gene expression.

Although numerous companies have explored gene expression, some early plans to bring specific signatures through regulatory approval as companion diagnostics have since dissipated. Merck, for example, dropped its CDx plans with NanoString in late 2017.

However, pharma firms' interest in immune-focused signatures, including NanoString's, does not seem to have died down, as companies continue to test existing panels and develop new ones as part of retrospective trial analyses.

In the new BMS analysis last week, authors reported that CheckMate 032 responders had higher scores in aggregate across all the gene expression signatures examined. "Despite the small sample size, all signatures evaluated showed a trend of association with response," Lei said during her presentation. But among the four analyzed, the BMS inflammatory signature, which includes four genes, had the strongest link to drug efficacy.

Localizing PD-L1

In another presentation focused on PD-L1, investigators from David Rimm's lab at Yale University shared data from their efforts to break down PD-L1 expression to different cell types, and to evaluate variable predictiveness based on this precise localization. The group has concluded from their early work that it is actually macrophages, not tumor cells, in which PD-L1 expression is predictive of immunotherapy response, something that could potentially significantly affect the accuracy of current IHC methods approved for use as companion or complementary diagnostics.

In a set of tissue samples from treated patients and controls, investigators localized PD-L1 by differentiating natural killer (NK) cells, macrophages, and cytotoxic T cells via quantitative immunofluorescence.

According to Yale PhD student Yuting Liu, who presented the study, the results indicated that PD-L1 is more highly localized in macrophages compared to NK cells and CD8 T cells in both tumor and stroma compartments. In stroma, for example, 80 percent of PD-L1 was expressed by macrophages, the authors reported. And even in the tumor compartment, 40 percent of PD-L1 expressing cells were macrophages rather than cancer cells.

Based on this finding, the group hypothesized that PD-L1 in macrophages is what "carries the predictive power of the biomarker," Liu said.

To test this, the group took another cohort of non-small cell lung cancer patients treated with a variety of immunotherapy drugs. The cohort had 81 patients in total, but the team only analyzed 62 who were on a monotherapy and had pre-treatment samples available.

Liu said that the level of PD-L1 in macrophages correlated strongly with the level of PD-L1 in tumor, as well as with high levels of CD8+ T cells in the study samples. Elevated PD-L1 in macrophages overall was also associated with high PD-L1 level in tumor tissue, but while high PD-L1 expression in macrophages was correlated with better overall survival, high PD-L1 expression in verified tumor cells was not.

The group didn't have a validation cohort, but validated the finding using a second method of quantitative immunofluorescence that uses DNA probes, with largely the same result.

In addition, Liu said that colleagues have since demonstrated that colocalization of PD-L1 binding and the binding of CMTM-6 (a recently described marker involved in the stabilization of PD-L1) also favors macrophages, as well as that macrophage PD-L1 (and not melanocyte PD-L1) is predictive of immunotherapy response in melanoma.

"Our proposed alternative mechanism of action is that in the context of immune checkpoint blockade … macrophages alter the tumor microenvironment, turning 'cold' tumors 'hot,'" Liu concluded.

In a brief question and answer period, audience members raised the question of whether macrophage PD-L1 is only predictive when the macrophages are in the tumor compartment, or if the overall population contributes. Some other studies of non-tumor PD-L1 have had poor results.

Liu said that the group hasn't yet calculated whether the tumor-bound macrophages are driving prediction, or whether the whole population, including macrophages in the stroma, is required. But she said that what is clear from the results so far is that it is not PD-L1 in actual tumor cells that is producing a predictive signal.

Improving Gene Expression Tests

In addition to efforts to improve PD-L1 analyses, investigators are also working on ways to increase the accuracy of gene expression profiles like those being tested by BMS and others.

In data from the ongoing TRACERx trial in lung cancer, investigators led by Cancer Research UK Chief Clinician Charles Swanton shared how they have been exploring the ways genomic heterogeneity in tumors may confound existing gene expression biomarkers that are applied to tissue samples.

Most of the team's work has focused on gene expression signatures for prognosis in early-stage cancer, but University College London PhD student Dhruva Biswas, who presented the results, said that the group is now pushing this into the realm of immunotherapy prediction.

In the context of TRACERx, the team has developed a tool called the "Outcome Risk Associated Clonal Lung Expression" (ORACLE) assay, which measures a set of genes that are expressed homogeneously within individual tumors (enriched in regions associated with things like mitosis and nucleosome assembly) but heterogeneously between patients.

Other presenters highlighted how existing tests might meet thresholds for clinical utility if applied in specific ways.

A team from the I-SPY2 trial in triple-negative breast cancer, for example, reported on an effort to study the overlap of immune-related gene expression with what they call the Parpi7 signature, identified as part of the trial's biomarker program as a measure of DNA repair deficiency that is predictive of response to PARP-inhibitor therapy.

An issue that arose as a result of I-SPY2 is that two different effective treatments have emerged: a combination of the PARP inhibitor veliparib, being developed by Abbvie, with chemotherapy, and pembrolizumab (Merck's Keytruda).

Hoping to hit on a way to potentially prioritize one or the other treatment in the clinic, the team set out to compare Parpi7 alongside an immune signature developed by NanoString in 153 patients from the control, pembrolizumab, and veliparib arms of the trial.

According to the authors, 40 percent of the cohort were positive for both biomarkers, 40 percent for only one or the other, and 20 percent for neither.

Using statistical methods, the group calculated that for the double positives, there was no significant difference in their response rates to veliparib versus pembrolizumab and both were much better than standard chemotherapy — patients in the control arm had an 18 percent response rate compared to either 83 percent or 84 percent with either of the two new therapies.

Denise Wolf, who presented that data, said that this suggests that double-positive patients can be confidently treated with either of the two. However, the presence of single positive does look to have important relevance to therapy choice. Immune-positive DRD-negative women responded much more frequently to immunotherapy, while Immune-negative DRD-positive had higher chances of responding to veliparib than to pembrolizumab.

Among double-negatives, only 27 percent responded to PARP inhibition, while 43 percent responded to immunotherapy, so research will have to continue to identify effective treatment for this subgroup.

MSI

With the first tissue-agnostic cancer drug approval in 2017, in which the US Food and Drug Administration endorsed pembrolizumab for solid tumors characterized by high microsatellite instability or mismatch repair deficiency, MSI has become an established predictive biomarker, even though FDA did not designate a specific companion diagnostic to accompany the drug.

Legacy methods include a variety of assays established initially for screening individuals for Lynch syndrome, which use IHC or PCR. But as broad sequencing assays have started to become more commonly used to guide cancer therapy, groups have also been keen to develop methods to read out MSI from NGS data.

During the same session at AACR last week, a team from the Broad Institute and Massachusetts General Hospital described their development of an NGS-based MSI assay, called MSIClass, that can be applied to limited quantities of cell-free DNA in the context of liquid biopsy.

"We were interested to see if we can detect this from blood because we thought this could be very helpful for patients who are too ill, too old, or maybe could be used for early detection," Broad postdoc Yosef Maruvka said at the meeting.

Other liquid biopsy developers are incorporating MSI into their panels in anticipation of broader use of blood-based NGS tests, especially in advanced cancer patients who cannot easily be biopsied, many of whom would be eligible for immunotherapy drugs.

But according to Maruvka, other sequencing methods require ultra-deep sequencing of pre-specified MSI loci, which may not capture differences between different tumor types and is also relatively expensive to perform.

He and his colleagues' development of MSIClass began with discovery of unique genome-wide mutational features that distinguish MSI vs. microsatellite-stable tumors, which the group then validated using whole-genome sequencing data from The Cancer Genome Atlas.

"We are not interested in mutations in specific MSI loci. … Instead [we] aggregate information from whole-genome even if it is at a very low confidence for any specific locus," Maruvka said.

According to the authors, the test requires minimal genomic information — working with sequencing coverage as low as 0.05x even in samples with as low as 0.1 percent tumor DNA fraction.

At AACR, the group shared data from 29 colon cancer cases, of which 22 were microsatellite stable and 7 were MSI-high. Maruvka said that the assay was able to correctly identify six of the seven.

In response to audience questions, he added that the team has done further work showing that the method also holds up when applied to smaller sequencing panels down to about 50 genes.