Article Text

Circulating immune cell dynamics as outcome predictors for immunotherapy in non-small cell lung cancer
  1. Alvaro Marcos Rubio1,2,3,
  2. Celine Everaert1,2,3,
  3. Eufra Van Damme1,2,3,
  4. Katleen De Preter1,2,3 and
  5. Karim Vermaelen2,4
  1. 1VIB UGent Center for Medical Biotechnology, Ghent, Belgium
  2. 2Cancer Research Institute Ghent (CRIG), Ghent University, Ghent, Belgium
  3. 3Department of Biomolecular Medicine, Ghent University, Gent, Belgium
  4. 4Tumor Immunology Laboratory, Department of Pulmonary Medicine, University Hospital Ghent, Ghent, Belgium
  1. Correspondence to Prof. Dr. Karim Vermaelen; karim.vermaelen{at}
  • AMR and CE are joint first authors.


The use of immune checkpoint inhibitors (ICIs) continues to transform the therapeutic landscape of non-small cell lung cancer (NSCLC), with these drugs now being evaluated at every stage of the disease. In contrast to these advances, little progress has been made with respect to reliable predictive biomarkers that can inform clinicians on therapeutic efficacy. All current biomarkers for outcome prediction, including PD-L1, tumor mutational burden or complex immune gene expression signatures, require access to tumor tissue. Besides the invasive nature of the sampling procedure, other disadvantages of tumor tissue biopsies are the inability to capture the complete spatial heterogeneity of the tumor and the difficulty to perform longitudinal follow-up on treatment. A concept emerges in which systemic immune events developing at a distance from the tumor reflect local response or resistance to immunotherapy. The importance of this cancer ‘macroenvironment’, which can be deciphered by comprehensive analysis of peripheral blood immune cell subsets, has been demonstrated in several cutting-edge preclinical reports, and is corroborated by intriguing data emerging from ICI-treated patients. In this review, we will provide the biological rationale underlying the potential of blood immune cell-based biomarkers in guiding treatment decision in immunotherapy-eligible NSCLC patients. Finally, we will describe new techniques that will facilitate the discovery of more immune cell subpopulations with potential to become predictive biomarkers, and reflect on ways and the remaining challenges to bring this type of analysis to the routine clinical care in the near future.

  • Immune Checkpoint Inhibitors
  • Non-Small Cell Lung Cancer
  • Blood Cell Count
  • Biomarkers, Tumor
  • Review

This is an open access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited, appropriate credit is given, any changes made indicated, and the use is non-commercial. See

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

The current predictive biomarker landscape for immune checkpoint inhibition in lung cancer

Immune checkpoint inhibitors (ICIs) have revolutionized the treatment of an increasing number of cancer types, including non-small cell lung cancer (NSCLC). Lung cancer constitutes by far the largest segment of disease indication for these drugs, which include PD-1 and PD-L1-blocking antibodies (in some regimen combined with chemotherapy), and more recently combinations with CTLA-4 blockade. Together with targeted therapies, ICIs have become part of standard of care in all stages of NSCLC, be it metastatic, locally advanced, and more recently also in early-stage disease as part of a neoadjuvant approach. Pembrolizumab, nivolumab, and cemiplimab are all PD-1-blocking monoclonal antibodies approved by the US Food and Drug Administration (FDA) and the European Medicines Agency, while durvalumab and atezolizumab are clinically approved anti-PD-L1 antibodies. Recently, ipilimumab (anti-CTLA-4) in combination with nivolumab and chemotherapy has also been approved by the FDA as a first-line treatment of metastatic NSCLC. Currently, these drugs are administered as monotherapy only with a Tumor Proportion Score of PD-L1 expression >50%; otherwise they are given in combination with chemotherapy.1 However, 60%–80% of NSCLC patients do not benefit from ICI treatment, and serious toxicities still occur in 10%–30% of the cases, which highlights the need for biomarkers that predict which patients are more likely to respond.

PD-L1 expression in the tumor microenvironment (TME) assessed by immunohistochemistry (IHC) is still used to predict the likelihood of response to PD-1/PD-L1 blockade. A proportion of ≥1% or ≥50% of PD-L1+ cancer cells on biopsy predicts high response rates to several of these ICIs as monotherapy in the first line setting.2 Nonetheless, PD-L1 as a predictive biomarker presents limitations: a proportion of patients with high tumor PD-L1 expression do not respond, and some patients who benefit from ICI in monotherapy have tumors with negative/low PD-L1 expression.3 Other issues are spatial and temporal intratumoral heterogeneity,4 and interassay variation of performance as several anti-PDL1 clones are used in IHC diagnostic kits.5 Alternatives to PD-L1 have been proposed, such as other immune response-related gene expression signatures in the TME. For instance, a peripheral T cell6 or an effector T cell signature,7 are associated with a durable clinical benefit from anti-PD-1 and anti-PD-L1 therapy. However, they have not been proven to have a better predictive potential than PD-L1 expression testing.8 Additionally, the proportion or abundance of tumor infiltrating lymphocytes (TILs) has also been investigated as a predictive biomarker,9 considering both the total number of lymphocytes and specific subpopulations, such as exhausted PD-1+ CD8+ T cells.10 A recently developed method, the Immunoscore IC assay, calculates a risk score by measuring the density of CD8+T cells and PD-L1+cells as well as the spatial distribution of these cells on a single tissue segment, and has shown predictive potential in metastatic colorectal cancer.11 Moreover, in two NSCLC patient cohorts, a low-risk Immunoscore IC correlated with prolonged overall survival (OS) and progression-free survival (PFS).12

Tumor mutational burden (TMB), another potential response predictor, represents the number of somatic mutations in a tumor. Overall, tumor types with a high TMB are more likely to respond to ICIs,13 and thus NSCLC patients with high TMB tumors experience improved objective response, durable clinical benefit, and PFS.14 Moreover, TMB>10 mutations/megabase has been recently approved as a tumor-agnostic biomarker for selecting patients with cancer more likely to respond to pembrolizumab.15 However, this fixed threshold still misses a considerable fraction of ICI-responding microsatellite-stable solid tumors, and this fraction is very variable across tumor types. In addition, TMB measurement by whole-exome sequencing (WES) is subject to sequencing depth bias, which reduces its robustness across different studies.16

These and other tissue-based biomarkers for ICI-response prediction in NSCLC patients have been discussed extensively before.17 18 However, none of these biomarkers are accurate enough to be implemented in a clinical setting, and all of them require the availability of tumor tissue from the patient. As an invasive procedure, tumor tissue collection through biopsy or surgery entails certain risks, and it is not always possible to perform, depending on the accessibility of the tumor and the condition of the patient. Also, longitudinal follow-up of treatment response by serial biopsy is practically not feasible or at least very burdensome for most patients.

Looking beyond the tumor bed: systemic factors associated with therapeutic outcome under ICI

The destruction of cancer cells by T cells in the tumor bed is the last event of an essential anti-tumoral process called the ‘cancer-immunity cycle’.19 The initiating event is the capture and processing of tumor neoantigens by antigen-presenting cells such as dendritic cells (DCs), which home to the lymph nodes to present epitopes on MHC class I and II molecules to T cells along with powerful costimulatory and polarizing signals. Proper priming of T cells leads to the acquisition of an effector phenotype with the capacity to leave the lymph nodes, enter the blood circulation to finally infiltrate tumorous foci to kill cancer cells expressing the relevant antigen(s) (figure 1). This means that an important segment of the antitumor immune response occurs outside the tumor, offering an opportunity to probe for successful antitumoral immune response by tracking changes in peripheral blood immune cells. Indeed, neoantigen-specific T cells have been found in peripheral blood of melanoma patients,20 and patients with cancer who best respond to anti-PD-L1 therapy experience a clonotypic expansion of circulating T cell clones that are also present in the tumor, suggesting a replenishment of newly primed T cells from outside the tumor.21 Besides T cells, other peripheral immune cell types in blood and secondary lymphoid organs (comprizing what has been called the ‘immune macroenvironment’22), such as natural killer (NK) cells, DCs, and B cells, rise in numbers after ICI treatment, indicating that systemic immune events accompany the ICI-induced antitumor response.23 Moreover, tumors can alter hematopoietic processes in the bone marrow, increasing the frequency of immunosuppressive immature myeloid cells (see below) and decreasing the proportion and activity of different antitumoral immune cell subtypes.22 This suggests that the presence of a tumor-induced dysfunctional hematopoiesis can contribute to a reduced effectiveness of ICIs. Apart from the tumor, other factors shape the peripheral immunological compartment during a patient’s lifespan, such as aging and age-related diseases,24 past infections,25 the nutritional status26 and the gut microbiome,27 and therefore, might influence the patient outcome after treatment with ICIs.

Figure 1

The cancer-immunity cycle. (1) Cancer cell death releases tumor antigens such as mutanome-derived ‘neoantigens’, cancer-testis antigens, overexpressed shared antigens or oncoviral proteins, which are captured and processed by dendritic cells (DCs) or other antigen-presenting cells (APCs). (2) APCs present tumor antigen-derived epitopes on major histocompatibility (MHC) I or MHCII molecules to T cells in the lymph nodes to activate them. (3) Neoantigen-specific cytotoxic T cells (CTLs) migrate through the blood stream. (4) CTLs extravasate through the endothelium and infiltrate the tumor. (5) CTLs recognize and target the cancer cells. (6) CTLs release cytotoxic enzymes and kill the cancer cells.

Given the aforementioned limitations of tumor tissue-based biomarkers, the interrogation of the systemic immune compartment through sampling of the peripheral blood of patients witt cancer has recently gained interest as a means to develop non-invasive predictive biomarkers. Blood-based molecular biomarkers for ICI response prediction in lung cancer have been reviewed elsewhere,28 29 including circulating PD-L1 (soluble or expressed in exosomes, as a surrogate of PD-L1 expression in the TME), circulating tumor DNA (ctDNA) levels and the TMB in ctDNA (bTMB). Next to these a-cellular biomarkers, circulating tumor cell (CTC) count and PD-L1 expression on CTCs have also been studied in relation to ICI response prediction. However, in this review, we will specifically focus on circulating immune cells, examining how their diversity, frequency, and activation/differentiation status relates to response and toxicity during ICI therapy in lung cancer. We will also provide an overview of several new groundbreaking techniques and how they will expand the repertoire of biomarkers for ICI response prediction in the coming years.

Peripheral lymphoid cells as predictors of response to ICIs in NSCLC patients

Absolute lymphocyte counts (ALCs) in whole blood are easy to obtain as part of routine blood analysis in patients with cancer and therefore an attractive candidate biomarker for therapeutic prediction and monitoring. Several studies involving anti-PD-1 or anti-PD-L1 treated NSCLC patients showed that a higher ALC at baseline was associated with an increased OS and PFS.30–33 Although a lower ALC appears correlated with more advanced disease, confounding factors complicate the interpretation of these results. Reductions of ALC can be induced by previous treatments as is the case with radiotherapy-associated lymphopenia.34

Among lymphocytes, T cell-related parameters are obvious candidate predictors of response to ICIs as they are direct targets of CTLA-4 or PD-1 blockade (figure 2).

Figure 2

Peripheral blood lymphoid cell frequencies associated with a better response to ICIs in NSCLC patients. ICIs, immune checkpoint inhibitors; NSCLC, non-small cell lung cancer.

Within T-cells, cytotoxic CD8+ T lymphocytes (CTLs) constitute the main effectors during the antitumor immune response. Besides naïve and effector CTLs, central and effector memory CTL subsets can also play an important role, since these cells are activated much faster when they re-encounter a specific tumor neoantigen. Although it has always been thought that therapy with PD-1 inhibitors restores pre-existing, but suppressed CTL antitumor response, several recent studies examining T cell clonality suggest that it can also trigger new responses by priming circulating peripheral T cells.35 36 These findings support the hypothesis that the frequency of tumor-specific peripheral CD8+ T cells can be a good predictive biomarker for ICI response, which has been investigated in several studies involving NSCLC patients. Indeed, peripheral blood PD-1+ CD8+ T cells are significantly more frequent in responders compared with non-responders before treatment with nivolumab.37 Furthermore, the number of circulating PD-1+ CD8+ T cells that express Ki-67 (a marker of cell proliferation) increased in the blood of responders within 138 or 439 weeks after the start of anti-PD-1 therapy. In addition, these T-cells exhibited an activated effector-like phenotype (CD38+ HLA-DR+ CD45RA- CCR7- PD1+ CD8+ T cells), and expressed co-stimulatory molecules (CD27, CD28, ICOS) and inhibitory receptors like CTLA-4. These changes seemed restricted to tumor-specific CD8+ T cells as Epstein-Barr Virus-specific CD8+ T cells did not show similar signs of activation and proliferation. Another costimulatory receptor, CD137, was expressed in higher levels in CD8+ T cells from anti-PD1-treated patients with better clinical response.40 By contrast, in a different study, CD8+ T cell subsets expressing terminal differentiation signatures (CD95+ CD69; CD45RA+ CCR7), or lacking co-stimulatory markers (CD28, ICOS), were significantly more present in responders than non-responders at baseline and during nivolumab treatment.41 Peripheral memory CD8+ T cells have also been associated with the response to ICIs in NSCLC patients. One study involving a small cohort (n=22) of patients treated with nivolumab found that a high central memory to effector CD8+ T cell ratio at baseline was associated with a longer PFS.42

Markers more specifically reflecting a cytotoxic functional status have the potential to be more predictive of therapeutic response. Accordingly, proliferation of circulating CD8+ T cells expressing CX3CR1 was increased in the blood of responders, but not in non-responders after treatment with pembrolizumab.43 This marker showed a better accuracy (88.9% vs 44.4%), specificity (87% vs 17.4%) and a similar sensitivity (92.3%) compared with PD-L1 expression in the tumor. CX3CR1 surface expression delineates differentiated CD8+ T cells expressing, among others, the cytolytic proteins granzyme B and perforin.44 However, preclinical experiments indicate that injected CX3CR1 CD8+ T cells from melanoma tumors are fully rescued by anti-PDL1 treatment resulting in enhanced antitumor efficacy, whereas depletion of CX3CR1hi T cells does not impact tumor control.45

Besides investigating the complete pool of a given CD8+ T cell subset, researchers have also focused on neoantigen-specific CD8+ T cells. Although technically feasible, tracking the presence of these rare cells in the blood requires a complex, lengthy and fully patient-individualized assay pipeline. It has, however, delivered interesting insights on the response to adoptive T cell therapy in melanoma46 and anti-PDL1 therapy in metastatic NSCLC.47 In the latter study, neoantigen-specific T cell clones exhibited a differentiated effector phenotype in responders, as opposed to a more memory-like phenotype in non-responders.

As opposed to CTLs, CD4+ T helper cells (Th cells) were thought to play a secondary role in the antitumor immune response process. However, it has been recently demonstrated that they are essential in initiating and sustaining the anti-PD1-induced systemic response.48 This crucial involvement of Th cells during the anti-tumor immune response supports the idea that they could be good predictor and/or indicators of response to ICIs in patients with cancer. Similarly to CD8+ T cells, an early rise in PD-1+ CD4+ T cells after 1 week of treatment has been correlated with a longer PFS in NSCLC patients treated with ICIs.49 In addition, several studies point to pretreatment levels of CD4+ T cells with a highly differentiated or memory phenotype as being associated with better response or a longer PFS.42 50 51

Despite being mainly enriched in the TME compared with peripheral blood, circulating regulatory T cells (Tregs) have also been proposed as a potential biomarker for predicting or monitoring the response to ICIs in NSCLC patients, due to their role as antitumor immunity suppressors.52 In a previously mentioned study, the frequency of Tregs was higher in non-responders prior to treatment with anti-PD-1, and Treg counts, especially the PD1+ Treg subset, decreased 1 week after treatment with anti-PD-1 or anti-PD-L1 only in the blood of responders.51 53 Interestingly, Tregs increased in patients experiencing hyperprogression, a phenomenon of accelerated disease evolution that is considered by some to be triggered by the immunotherapy itself.53 By contrast, another study reported a higher percentage of FoxP3+ Tregs in the blood of responders 1 week after treatment with anti-PD-1, thus contradicting the previous publications.54

NK cells express several immune checkpoint molecules, including LAG-3,55 CTLA-456 and PD-157 (although the latter has been disputed58), suggesting that NK cells might be a potential target for ICIs. In several studies, the percentage of circulating NK cells was higher in responders to anti-PD-1 treatment, both before37 59 and after therapy administration.60 In addition, one of these studies investigated multiple NK subsets, including CD56+, CD56dim, perforin+, granzyme B+ and CD3ζ+ NK cells, which were two times more frequent in responders than in non-responders at baseline. Moreover, responders showed an increase in the NK cell frequency 4 weeks after treatment, while it decreased in non-responders.37

B lymphocytes, despite being one of the infiltrating immune populations in the lung TME61 and promoting immunotherapy responses in tertiary lymphoid structures by differentiating into plasma cells and producing tumor-specific antibodies that recognize and react against tumor-associated antigens,62 have not yet been investigated as non-invasive ICI response predictors in NSCLC patients alone. Only two studies including 36% and 20% of NSCLC patients among pan-cancer cohorts, found a negative association between peripheral blood B cells and the response to ICI therapy at baseline63 and at the first response evaluation,64 respectively.

Peripheral myeloid cells as predictors of response to ICIS in NSCLC patients

Cells from the myeloid compartment are known to play very diverse roles in antitumor immunity. The TME conditions myeloid cells such as macrophages and granulocytes to a tumor-supporting/immune suppressing status. Chronic inflammation is one of the hallmarks of cancer,65 and the associated immunosuppressive climate is largely mediated by myeloid cell infiltration in the TME. Under proper stimuli, however, these cells can elicit powerful antitumoral effects. Myeloid cells express a whole array of ligands for checkpoint receptors such as PD-L1/PD-L2, HVEM, Galectin-9 and MHC class II, which can engage respectively PD-1, BTLA, TIM-3 and LAG-3 on the T-cell. In addition, so-called myeloid-derived suppressor cells (MDSCs), which are enriched both intratumorally and systemically in patients with cancer, can directly suppress antitumoral T-cell function through specific metabolic mechanisms (discussed below). As such, several myeloid cell subsets, for example, neutrophils, monocytes, DCs and MDSCs have been investigated as predictors of ICI therapeutic efficacy (figure 3).

Figure 3

Peripheral blood myeloid cell frequencies associated with a better response to ICIs in NSCLC patients. ICIs, immune checkpoint inhibitors; NSCLC, non-small cell lung cancer.

Neutrophils are the most frequent immune cell subpopulation in the NSCLC TME,66 and high levels of circulating neutrophils have been associated with poor disease outcomes in patients with advanced-stage cancer.67 The latter observation could be due to the fact these cells are actually granulocytic myeloid suppressor cells, as discussed below. In peripheral blood, a lower absolute neutrophil count (ANC) was associated with a higher OS and better response to anti-PD-1 antibodies.32 68 69 In line with the positive prognostic impact of circulating lymphocyte counts, the neutrophil to lymphocyte ratio (NLR), which is the ratio of ANC over ALC, and the derived NLR (dNLR), calculated as ANC/(white cell count—ANC), emerge as significant predictors of ICI response in many cancers including NSCLC. Several studies show that a lower NLR or dNLR before treatment with nivolumab or pembrolizumab is associated with a superior disease control rate and longer OS and/or PFS.68–73 In addition, some studies have found significant differences in the NLR between responders and non-responders a few weeks after the start of the treatment with anti-PD-1 antibodies,74 75 which suggests that NLR might also be a potential biomarker for monitoring the response to ICIs.

Monocyte generation, also called monopoiesis, is influenced by tumor-released cytokines in the blood stream, such as granulocyte colony-stimulating factor (G-CSF), granulocyte-macrophage colony-stimulating factor (GM-CSF), macrophage colony-stimulating factor (M-CSF), IL-6, TGF-β. These cytokines can change the frequency of circulating monocytes and reprogram them towards an immunosuppressive phenotype.76 Few studies have investigated the potential predicting ability of circulating monocytes for ICIs response in NSCLC. In one of these studies, a higher absolute monocyte count at baseline was associated with an increased risk of death and disease progression (including a shorter PFS) in patients treated with anti-PD-1 antibodies.69 Similarly to neutrophils, this might be based on the notion that these cells are in fact indistinguishable from monocytic MDSCs (mMDSCs) (see below). Monocytes are also known to express surface PD-L1, and in a different study a lower proportion of circulating PD-L1+ monocytes at baseline was associated with better response and a longer PFS and OS under anti-PD1 therapy.77

MDSCs constitute a heterogeneous group of bone marrow-derived heterogeneous cells that share the ability to suppress immune responses in several pathological conditions, including sepsis and cancer. In addition, a multitude of studies has found high frequencies of circulating MDSCs being associated with a worse prognosis in patients with advanced cancer.78 79 Two major MDSC subtypes have been proposed: polymorphonuclear or granulocytic MDSCs (gMDSCs) which are virtually indistinguishable from immature neutrophils, and mMDSCs. The distinctive features of gMDSCs versus mMDSCs have been studied in great detail in preclinical (ie, murine) cancer models, but are harder to define in the human setting. One of the reasons is that routine blood processing and cryopreservation procedures applied to clinical sample collections artefactually eliminate most of the granulocytic compartment. Two studies showed that the frequency of circulating mMDSCs decreased in responders during anti-PD-1 therapy.72 80 The inverse effect was consistently observed in non-responders. A low frequency of both mMDSCs and gMDSCs has also been found prior to anti-PD-1 treatment in the blood of responders.81 82 However, Passaro et al showed opposite results, as they found that high baseline levels of gMDSCs and a low baseline ratio of CD8 T cells/gMDSC were associated with a better clinical response, and longer PFS and OS.83 MDSCs have also been incorporated in other ratio-based predictive measures such as the Treg/gMDSC and the NK cell/gMDSC ratio (NMR).60 84 The NMR was compared with the predictive performance of standard PD-L1 expression in tumor tissue: in a subset of 48 patients a receiver operating characteristic (ROC) curve analysis showed that the AUC value for the NMR was significantly higher than for PD-L1 expression (online supplemental table 1).

Supplemental material

DCs are considered the most efficient antigen presenting cells (APCs), and hence play an essential role in the cancer immunity cycle. However, DCs consist of a heterogeneous group of cells with different phenotypes and functions, not all of which being immunostimulatory. DC subpopulations consist of two conventional myeloid DC subsets (cDC1 and cDC2), a newly described DC3, and the plasmacytoid DCs (pDCs).85 In the TME, pDCs exert tolerogenic or even immunosuppressive effects.86 On the contrary, cDC1 and cDC2 cells are crucial for CD8+ and CD4+ T cell activation (via the antigen presentation process), respectively. Similar to monocytes, DC maturation is also affected by tumor-secreted cytokines, such as vascular-endothelial growth factor (VEGF,87 which represents another example of how tumor cells can shape systemic immunity. Despite their important role in the response to ICIs as APCs, few studies have found DCs as potential predictive biomarkers. This could be due to the fact that DCs are only found in trace quantities in the peripheral circulation (cDCs and pDCs represent 0.3%–0.6% and 0.2%–0.5% of all blood leukocytes, respectively). Möller et al found that a higher percentage of total circulating DCs at the start of the treatment was associated with a longer OS and PFS in NSCLC patients treated with anti-PD-1. In addition, the percentage of DCs in blood experienced a 2-time fold increase after the third and fifth cycle of treatment in responding patients, whereas patients with a stable or progressive disease did not show such an increase. Furthermore, the percentage of pDCs at baseline was significantly lower in patients who had a PFS shorter than 1 month.72 In another study, the baseline PD-L1 expression was higher in pDCs and cDCs of patients without a clinical response to anti-PD-1, and this also correlated with a shorter PFS.77

New technologies enabling systemic biomarker discovery

The discovery of immunobiological parameters for prediction of therapeutic response under ICI is progressing at an accelerated pace thanks to recent technological advances. However, biomarker validation studies and eventually implementation in routine clinical practice will require a technology and assay that is simple, robust and can be deployed at a large scale in the most cost-effective way. In the exploratory context, flow cytometry has been the mainstay for immunophenotyping, and now allows detection of up to 20–30 markers simultaneously. Recently, several technologies have been developed to go beyond current limits of standard flow cytometry. All these technologies deal with reducing spectral overlap, which is an issue with classical fluorchrome-labeled antibodies in a standard flow cytometry experiment and thus limits the panel size.

A first technique that enables large antibody panels is spectral flow cytometry. With spectral flow cytometry fluorophore spillover is reduced by utilizing several detectors to measure the complete spectrum emission of every fluorophore across multiple lasers, resulting in a more comprehensive signature for each fluorophore. A spectral signature is then created for each fluorophore. While standard flow cytometry uses compensation to correct for fluorescence spillover, spectral flow cytometry employs unmixing, a mathematical method that differentiates the numerous fluorophore signatures within a multicolor sample, based on the unique spectral signature of each fluorophore. Fluorophores with almost identical peak emissions may be recognized and used together in a panel using this technique, allowing immunophenotyping panels to be expanded beyond 40 fluorescent parameters. Moreover, cellular autofluorescence can be extracted from the fluorescence signal to increase signal resolution and sensitivity. However, given the continuous nature of the spectral signatures, spectral flow cytometry data cannot be stored in standard FCS file format and therefore is not compatible with normal flow cytometry data analysis software. To solve this problem, several custom analysis tools that support spectral flow cytometry data have been developed.88 Another technique that enables detection of large marker panels by extensive antibody panels (more than 40) is mass cytometry (CyTOF), which combines time of flight mass spectrometry with immunophenotyping by labeling surface protein-specific antibodies with heavy metal ions. In a CyTOF experiment spectral overlap between the markers is virtually eliminated. Drawbacks of CyTOF compared with flow cytometry include a much slower flow rate (cells/second), which makes the data acquisition process longer, a higher cost, and risk of contamination with heavy metals present in laboratory reagents.89 CyTOF has already been applied to identify peripheral blood immune cell types associated with a better response to ICIs, such as NK cells in lung cancer59 and classical monocytes in melanoma.90 Cellular indexing of transcriptomes and epitopes by sequencing (CITE-seq) combines single-cell RNA sequencing (scRNA-seq) with surface protein expression data from the same single cells. CITE-seq allows for an even more comprehensive antibody panel than CyTOF, often comprizing more than 200 different antibodies.

The ability of T cells to express a repertoire of T cell receptors (TCRs) capable of recognizing tumor neoantigens presented by MHC molecules on the surface of tumor cells and APCs is a key element of a successful antitumor immune response. The huge diversity of the different complementarity-determining regions present in TCRs is generated by somatic recombination of the variable (V), diversity (D), and joining (J) genes. High-throughput sequencing of the TCRs (TCR-seq) has been developed to assess the diversity of the TCR pool of patients with cancer, which could be associated with the response to ICIs. Indeed, a higher peripheral blood TCR diversity at baseline is predictive of response to anti-PD-1/PD-L1 therapy in PD-1+ CD8+ T cells from NSCLC patients.91 In addition, the TCR repertoire was shown to broaden over the course of anti-PD-L1 treatment in a small cohort of NSCLC patients.92 Bentham et al developed a method for T cell fraction quantification using TCR-seq data obtained by WES.93 The T cell fraction in the TME quantified by this method was able to predict the response to ICIs in several cancer entities, including NSCLC, which suggests that it could also give similar results if applied to peripheral blood samples. Similarly to TCR-seq, B cell receptor sequencing (BCR-seq) also aims to capture the diversity of the different BCR clonotypes generated by B cells in multiple conditions. Interestingly, a study combined TCR-seq and BCR-seq on NSCLC patient to find that TCR and BCR diversity was lower in a subset of responders without an EGFR/ALK mutation after anti-PD-1 treatment. Also, patients within this subset with a reduced TCR diversity had a significantly longer PFS.94 These contradictory results indicate that more studies on TCR and BCR diversity need to be performed to confirm their role on the response of NSCLC patients to ICI therapy.

In recent years, the possibility of deciphering the immune content of a biological sample in a less biased manner using single-cell technologies has helped to discover new immune cell subsets or signatures present in the TME associated with the clinical benefit of ICIs. These technologies include the previously mentioned scRNA-seq and CITE-seq, and single-cell assay for transposase-accessible chromatin using sequencing (scATAC-seq), which allows to identify cell type-specific chromatin accessibility profiles and regulatory elements.95 Zhang et al combined scRNA-seq (including TCR-seq) and scATAC-seq to identify a subset of CXCL13+ T cells that was enriched in the TME of triple negative breast cancer patients responding to anti-PD-L1 plus chemotherapy.96 Another study performed CITE-seq in 35 NSCLC lesions from patients treated with ICIs revealed what the authors called the ‘lung cancer activation module’, which consisted in PD-1+ CXCL13+ activated T cells, IgG+ plasma cells, and SPP1+ macrophages and was enriched in tumor lesions from responsive patients.97 These promising results prompt us to consider applying single-cell technologies to interrogate peripheral blood of patients instead of tumor tissue, potentially enabling the discovery of new minimally invasive biomarkers for ICI response prediction. However, given the current high cost and long turn-around-time of these technologies, it is not possible yet to routinely implement them in a clinical setting, at least at the present day. Algorithms for cell-type deconvolution may represent a less expensive solution for this problem. These computational methods infer the cell-type composition of a heterogeneous sample using bulk RNA-seq data as input. Although not specific for any type of cell in particular, some deconvolution methods are more immune-oriented, and even specifically designed for immuno-oncology.98 The combination of comprehensive single-cell-based and cell type-specific gene expression references from peripheral blood mononuclear cells (PBMC) or whole blood samples, with accurate deconvolution methods could provide a cost-effective alternative for peripheral blood immune cell type proportions quantification as an approach for selecting patients with cancer who are likely to benefit from the treatment with ICIs.

All these techniques, despite allowing a deeper characterization of the immune cell phenotype, still need further development to be as suitable as flow cytometry for a clinical context. However, they can help to reveal previously unknown cell subpopulations and to better characterize the ones already known, which will definitely shed more light in the field of non-invasive biomarkers for ICI response prediction. Moreover, if the of cost of next generation sequencing (NGS) keeps dropping at the current pace in the next years, NGS-based biomarkers might become more and more common in the clinical setting.

Concluding remarks

It is clear from an increasing number of studies involving patients with lung cancer that intratumoral responses to ICIs are reflected by specific and complex changes in circulating immune cells. The challenge remains to identify the systemic immune events that are most strongly and consistently linked to therapeutic outcome, and package this knowledge into a biomarker that is fully validated and technologically easy to implement at scale.

The standardization of flow cytometry protocols and the development in the last years of high-throughput techniques, such as CyTOF and single-cell approaches, have led to several cutting-edge publications that reported on the enrichment/reduction of specific (combinations of) immune cell subtypes in tumor tissue samples of patients with cancer responding to ICIs.21 90 96 97 These results encourage the search for similar biomarkers within the circulating immune compartment, hence avoiding invasive tissue sampling procedures.

A recurring parameter that is associated with therapeutic success under ICI (or in the contrary a dominant immune-suppressed climate) in several cancers is the NLR. Although this parameter seems reductionist from a modern-day cancer immunological point of view, it has the advantage of reflecting the tug of war between MDSCs and effector T cells in an easily implementable way. Within the lymphocytic compartment itself, studies seem to converge on the predictive value of T cells with a CD8+ effector memory phenotype, that are in active cell cycle and express surface PD-1.37–41 43 If these and other findings are validated in larger cohorts, they may constitute a compound biomarker that can be captured and standardized into a relatively easy flow-cytometrical panel, within reach of most clinical hematology labs.

To make the most sense to the clinician, a biomarker should be able to predict outcome before initiating therapy. To answer this need, there are more studies reporting differences in blood immune cell type proportions between responders and non-responders at baseline than after the treatment. Still, immune cell changes occurring very early on after treatment initiation can be useful as well, and may give both caregivers and patients alike confidence that they are on the right path. The lack of consistent biomarkers for treatment monitoring can be explained by the different blood drawing timepoints selected in the different studies. The time range varies from just 1 week to 12 weeks after the first ICI cycle. This means that some studies might have missed significant differences of certain immune cell proportions between responders and non-responders by not measuring at the appropriate time points.

A considerable number of studies featured in this review include small patient cohorts, and in some cases a validation cohort is lacking. Also, most of the cohorts comprise only patients treated with ICIs as monotherapy, either as first-line or second-line treatment. However, patients treated with the combination of ICIs plus chemotherapy, today the dominant first-line therapeutic modality in advanced NSCLC, were not included in these reports. Regarding statistical analysis, results should be interpreted cautiously when the univariate Cox regression model was significant but the multivariate was not, since this could mean that other variables are driving the predictive effect of the biomarker. In addition, in several publications in which multiple cell types and/or clinical characteristics were analyzed simultaneously, the authors do not mention whether a multiple testing correction was performed to avoid false discoveries. Furthermore, ROC curve analysis was performed in only a few studies. These data are also interesting because they allow to compare the performance of different biomarkers within the same study,72 including PD-L1 expression data.60 So far, this is the only biomarker for ICI response prediction that has gone through the whole process that leads to clinical implementation. For all the potential biomarkers presented in this review, this trajectory should involve ROC curve analysis to determine an optimal and standardized cut-off point, and validation on large randomized phase III clinical trials, none of which having been performed yet.

In summary, evidence supports that the proportion of certain blood immune cell types (mostly T cells and neutrophils) is associated with the response to ICIs in NSCLC patients. However, there are no cell subpopulations yet that can solely predict the outcome of ICI-treated NSCLC patients and replace PD-L1 expression testing as the standard clinically available option. A combinatorial approach that takes into account the frequencies of several immune cell types would provide a more comprehensive view of the immunological processes involved in the response to ICIs and could represent a better strategy. Nevertheless, further research and methodological improvements are needed to determine whether the interrogation of immune cell proportions in the blood of the patients can be translated into an accurate predictive biomarker that can outcompete the currently available ones.

Ethics statements

Patient consent for publication


Figures 1-3 were created using


Supplementary materials

  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.


  • Twitter @AMarcosRubio

  • Contributors AMR, CE, KDP, and KV conceptualized the review. AMR performed the literature search, created the figures and wrote the first draft. CE, EVD, KDP and KV critically reviewed and commented on the manuscript. CE, KDP and KV supervised the work and obtained funding.

  • Funding This work received no dedicated funding. AMR is supported by Kom op tegen Kanker (Stand up againstCancer), while CE and EVD are supported by Fund for Scientific Research Flanders (FWO) postdoctoraland doctoral grants (grant IDs 11L7122N and 1226821N).

  • Competing interests AMR, CE, EVD and KDP declare no conflict of interest. KV has served as consultant for Bristol-Myers Squibb (BMS), Merck (MSD) and Roche Pharma.

  • Provenance and peer review Commissioned; externally peer reviewed.

  • Supplemental material This content has been supplied by the author(s). It has not been vetted by BMJ Publishing Group Limited (BMJ) and may not have been peer-reviewed. Any opinions or recommendations discussed are solely those of the author(s) and are not endorsed by BMJ. BMJ disclaims all liability and responsibility arising from any reliance placed on the content. Where the content includes any translated material, BMJ does not warrant the accuracy and reliability of the translations (including but not limited to local regulations, clinical guidelines, terminology, drug names and drug dosages), and is not responsible for any error and/or omissions arising from translation and adaptation or otherwise.