NEW YORK – BostonGene wants to further validate a blood-based, machine learning-powered immune profiling platform that has shown potential in classifying cancer patients' immune phenotypes and monitoring their treatment responses in early research.
The firm and its scientific collaborators detailed the development, training, and initial validation of the platform this month in Cancer Cell, noting that it gauges different immune markers in peripheral blood samples using multiparameter flow cytometry and bulk RNA sequencing data. While it is in early stages of development, the platform, if further validated, could help clinicians routinely characterize the composition of patients' immune systems and predict their ability to respond to immunotherapy treatments, said Michael Goldberg, senior author of the study and VP of R&D at BostonGene.
Currently, it is not possible to predict which patients are likely to respond to immune checkpoint inhibitors with great accuracy despite the association of high PD-L1 expression levels and tumor mutational burden with better response. Moreover, Goldberg and colleagues point out in their paper that standard molecular markers like PD-L1 and TMB, as well as attempts to directly characterize immune responses in tumor tissue by, for example, measuring tumor-infiltrating cells, don't provide a complete picture of the ability of patients' overall immune systems to attack the cancer in response to specific treatments.
"What we've done differently here is we've developed an automated analysis platform to identify different immune cell types very rapidly … so that it can be done consistently and can be scaled up to meet a potential clinical need," Goldberg said. "We developed an end-to-end clinical-grade cytometry pipeline to measure these particular [immune] cell types. Then, the platform analytically converts the different cell types that you see in patients into different groups."
In the study, Goldberg and colleagues analyzed thousands of peripheral blood samples using bulk RNA-seq, then, using BostonGene's machine learning-based cell deconvolution platform Kassandra, grouped the immune cell populations, including four populations with immune response biomarkers.
Factoring in distinct cell types and gene expression profiles, researchers then developed a scoring system for classifying patients' blood samples into five immunotypes, dubbed G1 to G5.
The immunotypes were discovered using an internal dataset of peripheral blood samples from 850 patients, which included both healthy donors and cancer patients. Once characteristics of the five immunotype groups were established, the researchers then expanded the analysis to an external dataset of peripheral blood transcriptomes from 17,800 individuals with and without cancer to verify the presence of these immunotypes in a larger independent dataset.
G1 or G1-naïve samples, for example, exhibited high frequencies of naïve CD4-positive T cells, naïve CD8-positive T cells, and naïve B cells, suggesting less activation of the immune system. On the other end of the spectrum, G5 or G5-suppressive samples had markers consistent with immune system dysfunction, such as higher levels of classical monocytes, HLA-DR-low monocytes, and neutrophils, along with smaller frequencies of lymphocytes.
Once researchers established the immunotype scores, they began retrospective analyses to gauge whether the scores correlated with treatment outcomes. Goldberg noted that these analyses were intended to test whether the platform's approach of stratifying patients using many immune markers could correlate with response.
"Immunologists generally are looking at individual [cell] populations or small combinations of individual [cell] populations and focus on a particular immune cell type," he said. "There isn't a very good consensus method for looking at large perturbations in the immune system [based on treatment]."
He added that researchers started testing the platform in peripheral blood samples taken from breast cancer patients treated with neoadjuvant chemotherapy because chemo tends to dramatically change the immune system. The researchers' later analyses explored the immunotypes in patients treated with chemo and immunotherapy and then immunotherapy alone, which doesn't suppress the immune system as much as treatment with chemo.
In that first analysis of blood samples taken from an external cohort of breast cancer patients who had gotten neoadjuvant chemo, researchers found that patients who achieved a pathologic complete response to neoadjuvant chemo were more frequently assigned to the G5-suppresive immunotype than those who had residual disease.
The researchers then moved to samples from a Phase II randomized trial evaluating Bristol Myers Squibb's checkpoint inhibitor Opdivo (nivolumab) plus chemo, with or without Pyxis Oncology's anti-CD40 immunotherapy sotigalimab as a first-line treatment for metastatic pancreatic cancer. They found that the on-treatment pancreatic cancer samples with a G3-progressive immunotype were associated with significantly longer overall survival than samples assigned to the other immunotypes.
Finally, the researchers explored the ability of the immune profiling platform to predict outcomes of head and neck cancer patients treated with frontline immunotherapy using samples from trials testing the activity of AstraZeneca's Imfinzi (durvalumab) and Opdivo. For these analyses, researchers explored the relationship between treatment outcomes and immunotypes assigned using pre-treatment and on-treatment samples.
In 32 patients from the Imfinzi trial, they noted that overall more responders tended to have the G4-chronic immunotype based on their on-treatment samples. In the pre-treatment samples, they also found that the G4-chronic immunotype had more responders.
To evaluate the predictive value of the immunotype assigned based on pre-treatment samples and its association with response, they compared the positive predictive value of the G4-chronic signature to the predictive value of tumor PD-L1 expression and found that the G4-chronic signature from this cohort outperformed tissue PD-L1 expression in stratifying response at baseline.
Using 35 head and neck cancer patients' samples from the Opdivo trial, the researchers found that in both pre-treatment and on-treatment samples more responders had the G2-primed immunotype. Within this cohort, the G2-primed immunotype was both prognostic and predictive of treatment response to Opdivo, Goldberg and colleagues reported.
These analyses in breast, pancreatic, and head and neck cancer provided an initial look at the immune profiling platform's ability to stratify responders and non-responders to different immunotherapies. "The rationale that undergirded all of this is that people are using immunotherapies, which modulate the immune system, and we're not really assessing the patient's immune system in a systematic way," Goldberg said.
His team's analyses showed, however, "that there was a systemic signature that corresponded with a response to the treatment either predictively or prognostically." Moreover, the researchers noted that the immunotypes associated with the best responses differed across treatment types due to the differences in how these treatments work.
The BostonGene team is now looking to further validate its platform for clinical use through several partnerships and sees an opportunity for the platform in longitudinal monitoring, said BostonGene CSO Joe Lennerz. "Instead of evaluating what is the predictive power of a one-time test, we believe the utility [of this platform] would be that we can provide a tool to monitor immune phenotype, or immunotype, changes in the peripheral blood on treatment," Lennerz said.
Goldberg added that using a blood-based assay to monitor patients could address some of the limitations of repeat tumor biopsies, which are often difficult to acquire and get reimbursement for, he said. Meanwhile, other blood-based assays to monitor patient response, such as cell-free DNA tests, often bring some diagnostic uncertainty, especially when they yield a negative result.
"We also have a cell-free DNA test [at BostonGene], but what we noticed is that it is monitoring a dead system," he explained. "[In] a disintegrating cell, some DNA comes out and is floating [in the blood]. But if you do not detect it, you don't know whether that was just an analytical absence or whether there was truly nothing there. … 'Is a negative result a true or false negative?' is a big question in the field of cell-free DNA."
Screening immune markers with the immune profiling platform, however, is monitoring a "living system," Lennerz said, which will always have a result.
One recently inked partnership with the Parker Institute for Cancer Immunotherapy stands to advance the immune profiling platform. Through the partnership, BostonGene will have access to samples taken at different time points in the care trajectory of more than 1,200 pan-cancer patients who received standard-of-care immune checkpoint inhibitor regimens.
Goldberg said the partnership is an "opportunity to expand this immunotype analysis to a larger cohort of patients and use that information to drill down on the specific tumor types and patients that may most potentially benefit from this diagnostic."
Goldberg noted that BostonGene also has other undisclosed partnerships in which it is evaluating the platform for longitudinal monitoring in large breast cancer, non-small cell lung cancer, and melanoma cohorts. Beyond treatment response, the team is also exploring whether the platform can identify patients who will develop toxicity or experience immune-related adverse events on immunotherapy, he added.
BostonGene is seeking more partnerships with research institutions and pharmaceutical companies to explore the potential of this platform across different cancer types and treatment modalities and test its use even outside of cancer, in autoimmune disorders, Lennerz said. In addition, BostonGene has developed a tumor microenvironment platform, called Tumor Portrait, which characterizes pan-cancer tumor microenvironment subtypes for predicting immunotherapy response.
"There's this gigantic gap between immunology research and clinical practice," Lennerz said. "We can contribute to bridging that gap by putting [a new platform] in place that provides a clinically readable assay output. To convey the immune system in a clinician-understandable fashion hasn't been accomplished so far, and that, to me, makes this paper really shine."