news

Immunotherapy has brought revolutionary changes to the treatment of malignant tumors, but there are still some patients who cannot benefit. Therefore, appropriate biomarkers are urgently needed in clinical applications to predict the effectiveness of immunotherapy, in order to maximize efficacy and avoid unnecessary toxicity.

FDA approved biomarkers

PD-L1 expression. The evaluation of PD-L1 expression levels by immunohistochemistry (IHC) yields the tumor proportion score (TPS), which is the percentage of partially or completely membrane stained tumor cells of any intensity in surviving tumor cells. In clinical trials, this test serves as an auxiliary diagnostic test for the treatment of advanced non-small cell lung cancer (NSCLC) with pembrolizumab. If the TPS of the sample is ≥ 1%, PD-L1 expression is considered; TPS ≥ 50% indicates high expression of PD-L1. In the initial Phase 1 trial (KEYNOTE-001), the response rate of patients in the PD-L1 TPS>50% subgroup using pembrolizumab was 45.2%, while regardless of TPS, the response rate of all patients receiving this immune checkpoint inhibitor (ICI) treatment was 19.4%. The subsequent phase 2/3 trial (KEYNOTE-024) randomly assigned patients with PD-L1 TPS>50% to receive pembrolizumab and standard chemotherapy, and the results showed a significant improvement in overall survival (OS) in patients receiving pembrolizumab treatment.

However, the application of PD-L1 in predicting ICI responses is limited by various factors. Firstly, the optimal threshold for different types of cancer varies. For example, Pabolizumab can be used when the tumor PD-L1 expression of patients with gastric cancer, esophageal cancer, bladder cancer cancer and lung cancer is 1%, 10% and 50% respectively. Secondly, evaluating the cell population of PD-L1 expression varies depending on the type of cancer. For example, the treatment of recurrent or metastatic squamous cell carcinoma of the head and neck may choose to use another FDA approved testing method, the Comprehensive Positive Score (CPS). Thirdly, there is almost no correlation between PD-L1 expression in various cancers and ICI response, indicating that tumor background may be a key factor in predicting ICI biomarkers. For example, according to the results of the CheckMate-067 test, the negative predictive value of PD-L1 expression in melanoma is only 45%. Finally, multiple studies have found that PD-L1 expression is inconsistent across different tumor lesions in a single patient, even within the same tumor. In summary, although initial clinical trials of NSCLC prompted research on PD-L1 expression as a possible predictive biomarker, its clinical utility in different types of cancer remains unclear.

Tumor mutation burden. Tumor Mutation Burden (TMB) has been used as an alternative indicator of tumor immunogenicity. According to the clinical trial results of KEYNOTE-158, among the 10 types of advanced solid tumors treated with pembrolizumab, patients with at least 10 mutations per megabase (high TMB) had a higher response rate than those with low TMB. It is worth noting that in this study, TMB was a predictor of PFS, but it was unable to predict OS.

The immune therapy response is mainly driven by T cell recognition of new antigens. The immunogenicity associated with higher TMB also depends on various factors, including the tumor neoantigen presented by the tumor; The immune system recognizes tumor neoantigens; The ability of the host to initiate antigen-specific responses. For example, data suggests that tumors with the highest infiltration of some immune cells may actually have inhibitory regulatory T cell (Treg) clone amplification. In addition, the range of TMB may differ from the potential of TMB neoantigens, as the exact site of the mutation also plays a significant role; Mutations that mediate different pathways of antigen presentation can affect the presentation (or non presentation) of new antigens to the immune system, indicating that tumor intrinsic and immunological characteristics must be consistent in order to produce optimal ICI responses.

At present, TMB is measured through next-generation sequencing (NGS), which may vary among different institutions (internally) or commercial platforms used. NGS includes whole exome sequencing (WES), whole genome sequencing, and targeted sequencing, which can be obtained from tumor tissue and circulating tumor DNA (ctDNA). It is worth noting that different types of tumors have a wide range of TMB, with immunogenic tumors such as melanoma, NSCLC, and squamous cell carcinoma having the highest TMB levels. Similarly, detection methods designed for different tumor types have different definitions of TMB threshold values. In the study of NSCLC, melanoma, urothelial carcinoma, and small cell lung cancer, these detection methods use different analytical methods (such as WES or PCR detection for specific numbers of related genes) and thresholds (TMB high or TMB low).

Microsatellites are highly unstable. Microsatellite highly unstable (MSI-H), as a pan cancer biomarker for ICI response, has excellent performance in predicting ICI efficacy in various cancers. MSI-H is a result of mismatch repair defects (dMMR), leading to a high mutation rate, especially in microsatellite regions, resulting in the production of a large number of new antigens and ultimately triggering a clonal immune response. Due to the high mutation burden caused by dMMR, MSI-H tumors can be considered as a type of high mutation burden (TMB) tumor. Based on the clinical trial results of KEYNOTE-164 and KEYNOTE-158, the FDA has approved pembrolizumab for the treatment of MSI-H or dMMR tumors. This is one of the first pan cancer drugs approved by the FDA driven by tumor biology rather than histology.

Despite significant success, there are also issues to be aware of when using MSI status. For example, up to 50% of dMMR colorectal cancer patients have no response to ICI treatment, highlighting the importance of other features in predicting response. Other intrinsic features of tumors that cannot be evaluated by current detection platforms may be contributing factors. For example, there have been reports that patients with mutations in genes encoding important catalytic subunits of polymerase delta (POLD) or polymerase ε (POLE) in the DNA region lack replication fidelity and exhibit a “super mutation” phenotype in their tumors. Some of these tumors have significantly increased microsatellite instability (thus belonging to MSI-H), but mismatch repair proteins are not lacking (therefore not dMMR).

In addition, similar to TMB, MSI-H is also affected by the new antigen types generated by microsatellite instability, host recognition of new antigen types, and host immune system responsiveness. Even in MSI-H type tumors, a large number of single nucleotide mutations have been identified as passenger mutations (non driver mutations). Therefore, relying solely on the number of microsatellites identified in the tumor is not enough; The actual type of mutation (identified through specific mutation profiles) can improve the predictive performance of this biomarker. In addition, only a small proportion of cancer patients belong to MSI-H tumors, indicating the current need for more widely applicable biomarkers. Therefore, identifying other effective biomarkers to predict efficacy and guide patient management remains an important research area.

Organizational based biomarker research

Given that the mechanism of action of ICI is to reverse immune cell suppression rather than directly targeting the intrinsic pathways of tumor cells, further research should focus on systematically analyzing the tumor growth environment and the interaction between tumor cells and immune cells, which may help elucidate the factors affecting ICI response. Many research groups have studied tumor or immune features of specific tissue types, such as tumor and immune gene mutation features, tumor antigen presentation deficits, or multicellular immune centers or aggregates (such as tertiary lymphoid structures), which can predict responses to immunotherapy.

Researchers used NGS to sequence the tumor and immune exome and transcriptome of patient tissues before and after ICI treatment, and conducted spatial imaging analysis. By using multiple integrated models, combined with techniques such as single-cell sequencing and spatial imaging, or multi omics models, the predictive ability of ICI treatment outcomes has been improved. In addition, a comprehensive method for evaluating tumor immune signals and intrinsic tumor characteristics has also shown stronger predictive ability. For example, a comprehensive batch sequencing method that simultaneously measures tumor and immune characteristics is superior to a single analytical variable. These results highlight the necessity of simulating ICI efficacy in a more comprehensive manner, including incorporating evaluation results of host immune capacity, intrinsic tumor characteristics, and tumor immune components into individual patients to better predict which patients will respond to immunotherapy.

Given the complexity of incorporating tumor and host factors in biomarker research, as well as the potential need for longitudinal integration of immune microenvironment features, people have begun to explore biomarkers using computer modeling and machine learning. At present, some groundbreaking research achievements have emerged in this field, indicating the future of personalized oncology assisted by machine learning.

The challenges faced by tissue based biomarkers

Limitations of analytical methods. Some meaningful biomarkers perform well in certain tumor types, but not necessarily in other tumor types. Although tumor specific gene features have stronger predictive ability than TMB and others, they cannot be used for the diagnosis of all tumors. In a study targeting NSCLC patients, gene mutation features were found to be more predictive of ICI efficacy than high TMB (≥ 10), but more than half of the patients were unable to detect gene mutation features.

Tumor heterogeneity. The tissue based biomarker method only samples at a single tumor site, which means that the evaluation of specific tumor parts may not accurately reflect the overall expression of all tumors in the patient. For example, studies have found heterogeneity in PD-L1 expression between and within tumors, and similar issues exist with other tissue markers.

Due to the complexity of biological systems, many previously used tissue biomarkers may have been oversimplified. In addition, cells in the tumor microenvironment (TME) are usually mobile, so the interactions displayed in spatial analysis may not represent the true interactions between tumor cells and immune cells. Even if biomarkers can ideally represent the entire tumor environment at a specific time point, these targets can still be induced and dynamically change over time, indicating that a single snapshot at a time point may not represent dynamic changes well.

Patient heterogeneity. Even if known genetic changes related to ICI resistance are detected, some patients carrying known resistance biomarkers may still benefit, possibly due to molecular and/or immune heterogeneity within the tumor and at different tumor sites. For example, β 2-microglobulin (B2M) deficiency may indicate new or acquired drug resistance, but due to the heterogeneity of B2M deficiency between individuals and within tumors, as well as the interaction of immune recognition replacement mechanisms in these patients, B2M deficiency may not strongly predict individual drug resistance. Therefore, despite the presence of B2M deficiency, patients may still benefit from ICI therapy.

Organizational based longitudinal biomarkers
The expression of biomarkers may change over time and with the impact of treatment. Static and single assessments of tumors and immunobiology may overlook these changes, and changes in tumor TME and host immune response levels may also be overlooked. Multiple studies have shown that obtaining samples before and during treatment can more accurately identify changes related to ICI treatment. This highlights the importance of dynamic biomarker assessment.

Blood based biomarkers
The advantage of blood analysis lies in its ability to biologically evaluate all individual tumor lesions, reflecting average readings rather than specific site readings, making it particularly suitable for evaluating dynamic changes related to treatment. Numerous research results have shown that using circulating tumor DNA (ctDNA) or circulating tumor cells (CTC) to evaluate minimal residual disease (MRD) can guide treatment decisions, but these tests have limited information on predicting whether patients can benefit from immunotherapies such as ICI. Therefore, ctDNA testing needs to be combined with other methods to measure immune activation or host immune capacity. In this regard, progress has been made in the immunophenotyping of peripheral blood mononuclear cells (PBMCs) and proteomic analysis of extracellular vesicles and plasma. For example, peripheral immune cell subtypes (such as CD8+T cells), high expression of immune checkpoint molecules (such as PD1 on peripheral CD8+T cells), and elevated levels of various proteins in plasma (such as CXCL8, CXCL10, IL-6, IL-10, PRAP1, and VEGFA) may all serve as effective supplements to ctDNA dynamic co biomarkers. The advantage of these new methods is that they can evaluate changes within the tumor (similar to changes detected by ctDNA) and may also reveal changes in the patient’s immune system.

Radiomics
The predictive factors of image data can effectively overcome the limitations of tissue biomarker sampling and biopsy, and can observe the entire tumor and possible other metastatic sites at any time point. Therefore, they may become an important part of non-invasive dynamic biomarkers in the future. Delta radiomics can quantitatively calculate the changes in multiple tumor features (such as tumor size) at different time points, such as before and after ICI treatment, during treatment, and subsequent follow-up. Delta radiomics can not only predict initial or no response to early treatment, but also identify acquired resistance to ICI in real-time and monitor any recurrence after complete remission. The imaging model developed through machine learning technology is even better than the traditional RECIST standard in predicting treatment response and possible adverse events. Current research indicates that these radiomics models have an area under the curve (AUC) of up to 0.8 to 0.92 in predicting immune therapy response.

Another advantage of radiomics is its ability to accurately identify pseudo progression. The radiomics model constructed through machine learning can effectively distinguish between true and false progression by re measuring CT or PET data for each tumor, including factors such as shape, intensity, and texture, with an AUC of 0.79. These radiomics models may be used in the future to avoid premature termination of treatment due to misjudgment of disease progression.

Intestinal microbiota
The biomarkers of gut microbiota are expected to predict the therapeutic response of ICI. Numerous studies have shown that a specific gut microbiota is closely related to the response of various types of cancer to ICI treatment. For example, in patients with melanoma and liver cancer, the abundance of Ruminococcaceae bacteria is associated with PD-1 immunotherapy response. Akkermansia muciniphila enrichment is common in patients with liver cancer, lung cancer, or renal cell carcinoma, who respond well to ICI treatment.

In addition, the new machine learning model can be independent of tumor types and associate specific gut bacterial genera with the therapeutic response of immunotherapy. Other studies have also revealed the specific role that individual bacterial groups play in regulating the host immune system, further exploring how to prevent or promote immune escape of cancer cells.

Neoadjuvant therapy
Dynamic evaluation of tumor biology can guide subsequent clinical treatment strategies. The neoadjuvant therapy trial can evaluate the therapeutic effect through pathological remission in surgical specimens. In the treatment of melanoma, the primary pathological response (MPR) is associated with recurrence free survival rate. In the PRADO trial, researchers determine the next clinical intervention measures, such as surgery and/or adjuvant therapy, based on patient specific pathological remission data.

Among various types of cancer, several new adjuvant therapy options still lack head to head comparison. Therefore, the choice between immunotherapy monotherapy or combination therapy is often decided jointly by the attending physician and the patient. Currently, researchers have developed an interferon gamma (IFN gamma) feature containing 10 genes as a biomarker for predicting pathological remission in melanoma after neoadjuvant therapy. They further integrated these features into an algorithm to select patients with strong or weak responses to neoadjuvant therapy. In a follow-up study called DONIMI, researchers used this score, combined with more complex analysis, not only to predict treatment response, but also to determine which stage III melanoma patients require the addition of histone deacetylase inhibitors (HDACi) to enhance response to neoadjuvant ICI treatment.

Tumor model derived from patients
In vitro tumor models have the potential to predict patient specific responses. Unlike the in vitro platform used for drug response spectrum analysis of hematologic malignancies, solid tumors face greater challenges due to their unique tumor microstructure and tumor immune interactions. Simple tumor cell culture cannot easily replicate these complex features. In this case, tumor like organs or organ chips originating from patients can compensate for these structural limitations, as they can preserve the original tumor cell structure and simulate interactions with lymphoid and myeloid immune cells to evaluate ICI responses in a patient specific manner, thereby more accurately reproducing biological features in a more realistic three-dimensional environment.

Several breakthrough studies in China and the United States have adopted this new high fidelity three-dimensional in vitro tumor model. The results show that these models can effectively predict the response of lung cancer, colon cancer, breast cancer, melanoma and other tumors to ICI. This lays the foundation for further verifying and standardizing the predictive performance of these models.