Article Text

Original research
High tumor mutational burden predicts favorable response to anti-PD-(L)1 therapy in patients with solid tumor: a real-world pan-tumor analysis
  1. Jaeyun Jung1,
  2. You Jeong Heo2 and
  3. Sehhoon Park2
  1. 1Innovative Institute for Precision Medicine, Samsung Medical Center, Seoul, Korea (the Republic of)
  2. 2Division of Hematology-Oncology, Department of Medicine, Samsung Medical Center, Sungkyunkwan University School of Medicine, Seoul, Korea, Seoul, Gangnam-gu, Korea (the Republic of)
  1. Correspondence to Dr Sehhoon Park; sehhoon.park{at}


Background Tumor mutation burden (TMB) is an important biomarker to predict response to anti-PD-L1 treatment across cancer types. TruSight Oncology 500 (TSO500) is currently used globally as a routine assay for TMB.

Methods Between 2019 and 2021, 1744 patients with cancer received TSO500 assay as part of a real-world clinical practice at the Samsung Medical Center, and 426 received anti-PD-(L)1 treatment. Correlations between TMB and clinical outcomes of anti-PD-(L)1 were analyzed. Digital spatial profiling (DSP) was used to investigate the tumor immune environment’s influence on the treatment response to anti-PD-(L)1 in high TMB (TMB-H) patients (n=8).

Results The incidence of TMB-H (≥10 mutations (mt)/megabase (Mb)) was 14.7% (n=257). Among TMB-H patients, the most common cancer type was colorectal cancer (n=108, 42.0%), followed by gastric cancer (GC; n=49, 19.1%), bladder cancer (n=21, 8.2%), cholangiocarcinoma (n=21, 8.2%), non-small cell lung cancer (n=17, 6.6%), melanoma (n=8, 3.1%), gallbladder cancer (GBC; n=7, 2.7%), and others (n=26, 10.1%). The response rate to anti-PD-(L)1 therapy was substantially higher in GC (71.4% vs 25.8%), GBC (50.0% vs 12.5%), head and neck cancer (50.0% vs 11.1%), and melanoma (71.4% vs 50.7%) among TMB-H patients when compared with low TMB (TMB-L) (<10 mt/Mb) patients with statistical significance. Additional analysis of patients with TMB ≥16 mt/Mb demonstrated prolonged survival after anti-PD-(L)1 therapy compared with patients with TMB-L (not reached vs 418 days, p=0.03). The benefit of TMB ≥16 mt/Mb was greater when combined with microsatellite status and PD-L1 expression profiles. Among the TMB-H patients, those who responded to anti-PD-L1 therapy had numerous active immune cells that infiltrated the tumor regions during the DSP analysis. Natural killer cells (p=0.04), cytotoxic T cells (p<0.01), memory T cells (p<0.01), naïve memory T cells (p<0.01), and proteins related to T-cell proliferation (p<0.01) were observed in a responder group compared with a non-responder group. In contrast, exhausted T-cell and M2 macrophage counts were increased in the non-responder group.

Conclusions The overall incidence of TMB status was analyzed by the TSO500 assay, and TMB-H was observed in 14.7% of the pan-cancer population. In a real-world setting, TMB-H identified by a target sequencing panel seemed to predict response to anti-PD-(L)1 therapy, especially in patients with a higher proportion of immune cells enriched in the tumor region.

  • tumor microenvironment
  • immunotherapy
  • tumor biomarkers

Data availability statement

Data are available upon reasonable request. The data generated in this study are available on request from the corresponding author.

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.


  • Tumor mutation burden (TMB) has emerged as a predictive biomarker of immune checkpoint inhibitor therapy. The gold standard for the assessment is whole-exome sequencing, which limits its clinical utility. Limited data show the general proportion of TMB assessed by the target sequencing panel and the clinical outcome of PD-(L)1 therapy at the pan-cancer level.


  • This study comprehensively analyzed the TMB profile of patients who underwent the TruSight Oncology 500 assay as part of routine clinical practice. In this study, we identified that TMB predicts anti-PD-(L)1 therapy, and the response was more enriched in TMB with a higher cut-off of 16 mutations/megabase. Among a high TMB population, responders were highlighted by a higher number of immune cells infiltrating the tumor region.


  • This study provides additional clinical evidence on TMB, which is simply assessed from a target sequencing panel that predicts PD-(L)1 therapy response. To enhance the predictive value of TMB, further adjustments or considerations based on the tumor microenvironment status need to be taken.


Tumor mutation burden (TMB) is a genomic biomarker that is calculated by the total number of somatic alterations identified using exome sequencing of the coding region.1 Based on the concept of cancer immunity cycle,2 neoantigens, which are likely to be expressed in highly mutated tumors, cause T-cell activation, leading to the hypothesis that TMB may be a useful predictive biomarker of response to immune checkpoint inhibitor (ICI) treatment. Tumors with high TMB (TMB-H) demonstrated positive treatment outcomes with PD-(L)1 or CTLA-4 blockade in several studies.3–6 A pan-tumor analysis of 12 trials with pembrolizumab (n=1772) demonstrated that TMB-H patients (≥175 mutations (mt)/exome) were associated with significant improvement in pembrolizumab efficacy.6 7

Recently, the TruSight Oncology 500 (TSO500) assay, a comparably large target sequencing panel, was evaluated in archival pan-tumor samples from 294 patients enrolled in eight different clinical trials treated with pembrolizumab, a PD-1 inhibitor.8 In this study, TMB calculated from the TSO500 assay was compared with TMB calculated from FoundationOneCDx and whole-exome sequencing to evaluate the correlation between the tests.8 In this analysis, a high correlation between the target sequencing panel (TSO500, FoundationOneCDx assay, and WES) was observed when a cut-off of 10 mt/megabase (Mb) was used for TMB-H.8

Because the target sequencing panel has become a standard diagnostic procedure for oncology practice, we conducted a comprehensive analysis using the TSO500 assay results from a real-world clinical practice to evaluate the overall incidence of TMB-H and the clinical outcome of ICI treatment among a pan-cancer population. In addition, the tumor microenvironment of the TMB-H population based on the ICI response was investigated using digital spatial profiling (DSP).


Patient enrollment

At the Samsung Medical Center (SMC), the genomic profiles of 1753 patients with solid cancer who underwent the TSO500 assay between October 2019 and April 2021 were collected. Nine patients were excluded from the final analysis owing to the poor quality of the genomic profiling. Clinicopathological data, including the anti-PD-(L)−1 response, were extracted from electronic medical records using SMC clinical data version 1.2. The best response was described as a complete response (CR), partial response (PR), stable disease (SD), or progressive disease (PD) using the Response Evaluation Criteria in Solid Tumors 1.1. Patients who showed either CR or PR were considered responders.

DNA extraction and TruSight Oncology 500 assays

Forty nanograms of DNA was quantified using the Qubit dsDNA HS assay (Thermo Fisher Scientific, Waltham, MA, USA) on a Qubit 2.0 fluorometer (Thermo Fisher Scientific) and then sheared using a Covaris E220 focused-ultrasonicator (Woburn, MA, USA) and eight microTUBE–50 Strip AFA Fiber V2, according to the manufacturer’s instructions. Treatment time was optimized using fresh, frozen, and paraffin-embedded tissues. Treatment settings were as follows: peak incident power, 75 W; duty factor, 15%; cycles per burst, 500; treatment time, 360 s; temperature, 7°C. For DNA library preparation and enrichment, a TSO500 Kit (Illumina, San Diego, CA, USA) was used according to the manufacturer’s instructions. Post-enriched libraries were quantified, pooled, and sequenced using the NextSeq 500 (Illumina). The quality of NextSeq 500 sequencing runs was assessed using the Illumina sequencing analysis viewer. Sequencing data were analyzed using TSO500 Local App V. (Illumina). The TSO500 is a comprehensive tumor profiling assay designed to identify known and emerging tumor biomarkers, including small variants, splice variants, fusions, TMB, microsatellite instability (MSI), and the degree of mutation in homologous recombination (HR) genes (homologous recombination deficiency, HRD).

Tumor mutation burden was reported as mutations per megabase sequenced, and TMB-H was defined as ≥10 mt/Mb.9 Variants with an observed allele count ≥10 in any of the GnomAD exome, genome, or 1000 genome databases were excluded. Both non-synonymous and synonymous single nucleotide variants (SNVs) and insertions and deletions (indels) in the coding region with a variant frequency ≥5% were included.10 The effective size for TMB calculation was considered to be the total coding region with coverage >50×. Microsatellite instabilities were calculated from the microsatellite sites for evidence of instability relative to a set of normal baseline samples based on information entropy metrics. The percentage of unstable MSI sites among the total assessed MSI sites was reported as the sample-level microsatellite score.

From the TSO500 panel, the following genes classified as HR genes were identified: ARID1A, ATM, ATRX, BAP1, BARD1, BLM, BRCA1, BRCA2, BRIP1, CHEK1, CHEK2, FANCA, FANCC, FANCD2, FANCE, FANCF, FANCG, FANCL, MRE11A, NBN, PALB2, PTEN, RAD50, RAD51, and RAD51B. Sequence data were also analyzed for clinically relevant classes of genomic alterations, including SNVs and small indels, copy number variations (CNVs), and rearrangements/fusions. SNVs and small indels with a variant allele frequency of <2% were excluded. Average CNVs of ≥2 were considered copy number gains (amplification), and those <1 were considered copy number (CN) losses. In addition, RNA translocation-supporting reads of >4–12 were considered translocations, which were dependent on sample quality. Data outputs exported from the TSO500 pipeline (Illumina) were annotated using the Ensembl Variant Effect Predictor annotation engine, with information from the dbSNP, gnomAD genome, 1000 genomes, ClinVar, COSMIC, RefSeq, and Ensembl databases. Processed genomic changes were categorized according to the four-tier system proposed by the American Society of Clinical Oncology/College of American Pathologists and annotated with proper references.

PD-L1 immunohistochemistry

Tissue sections were freshly cut into 4 μm thick sections, mounted on Fisherbrand Superfrost Plus Microscope Slides (Thermo Fisher Scientific), and then dried at 60°C for 1 hour. Immunohistochemistry was performed using a Dako Autostainer Link 48 system (Agilent Technologies, Santa Clara, CA, USA) using a Dako PD-L1 immunohistochemistry 22C3 pharmDx kit (Agilent Technologies) with an EnVision FLEX visualization system and then counterstained with hematoxylin, according to the manufacturer’s instructions. Protein expression of PD-L1 was quantified using both combined positive scores (CPS), which were calculated as the number of PD-L1-stained cells divided by the total number of viable tumor cells multiplied by 100.

Digital spatial profiling with 43 immune cell markers

Patients with a TMB-H (≥10 mt/Mb) and microsatellite stability (MSS) were selected for DSP (n=8). First, 5 µm thick tissue sections were trimmed from formalin-fixed paraffin-embedded tissue blocks and mounted on glass slides. Samples were prepared according to the manufacturer’s instructions. A total of 46 antibodies, including morphological markers (panCK and CD45), were used. The tissue sections were baked at 60°C for at least 2 hours. After baking, the tissue sections were deparaffinized using Neo-Clear (a xylene substitute; Sigma Aldrich, St. Louis, MO, USA) and rehydrated. Antigen retrieval (1× citrate buffer, pH 6.0; Sigma Aldrich) was performed under high pressure and temperature conditions. Tissue sections were washed using 1× Tris-buffered saline with Tween-20 (Cell Signaling Technology, Danvers, MA, USA) and blocked using buffer W (NanoString Technologies, Seattle, WA, USA) for 1 hour to prevent non-specific binding. The tissues were stained with fluorescence-labeled antibodies, including pan-cytokeratin (AE1+AE3) and CD45 (2B11+PD7/26; GeoMX Solid Tumor TME Morphology Kit), to segment the morphology and antibody cocktails that included 44 antibodies (immune cell profiling core, IO drug target module, immune activation status module, and immune cell typing module). Nuclear staining was performed for nuclear segmentation.

Stained tissues were analyzed using the GeoMx DSP system. The region of interest was selected based on the morphological markers. Digital counts of each protein were collected and normalized to the internal spike-in controls and endogenous genes. All statistical analyses of the DSP results were performed using R software V.4.2.0.

Nearest neighbor (NN) analysis

Cell segmentation was performed using HighPlex FL algorithms in HALO 3.4v (Indica Labs, USA) based on the DNA staining signal. The cells were divided into two groups: tumor and immune cells. Pankeratin-positive cells were annotated as tumor cells, and CD45-positive cells were annotated as immune cells. After the annotation, distances between the tumor and immune cells were calculated using the NN algorithms of the spatial analysis modules in the HALO software. The Wilcoxon rank-sum test was performed to compare the distance between tumor and immune cells in the responder and non-responder groups.

Differentially expressed protein (DEP) analysis

Differentially expressed proteins were identified using the DESeq package in R software V.4.2.0. The Wilcoxon rank-sum test was performed to compare the responder and non-responder groups. Enrichment analysis was performed using the database for annotation, visualization, and integrated discovery.

Statistical analyses

Descriptive statistics were reported as proportions and medians. Data are presented as numbers (%) for categorical variables. Correlations between TMB status and clinicopathological features were analyzed using the t-test, Fisher’s exact test, or one-way analysis of variance, as appropriate. The Mann-Whitney U test was used to compare the differences between the TMB-H and low TMB (TMB-L) groups. Overall survival (OS) was defined as the time from the first treatment to the date of death. Progression-free survival (PFS) was defined as the time from treatment initiation to the date of disease progression or all-cause mortality. Kaplan-Meier estimates were used in the analysis of all time-to-event variables, and the 95% CI for the median time-to-event was computed. From the differentially expressed gene analysis, a log2-fold change of <0.75 or >0.75 satisfying a-Log10 p of 1.25 was considered significant. Throughout the manuscript, a p value of <0.05 was considered significant. All statistical analyses were performed using Prism 8 (GraphPad Software, San Diego, CA, USA; or R software for Windows (V.4.1.2; The RStudio desktop 1.4 was used to draw all graphics (Rstudio Team, 250 Northern Ave, Boston, MA 02210, USA;


Characteristics of TMB-H patients

Among 1744 patients, the incidence of TMB-H (≥10 mt/Mb) in the TSO500 assay was 14.7% (n=257) (figure 1A). The entire cohort (n=1744) comprised 33 types of cancers: colorectal cancer (CRC, n=547, 31.4%), gastric cancer (GC, n=381, 21.8%), sarcoma (n=155, 8.9%), pancreatic cancer (n=125, 7.2%), melanoma (n=92, 5.3%), non-small cell lung cancer (NSCLC, n=77, 4.4%), bladder cancer (n=65, 3.7%), hepatocellular carcinoma (HCC, n=36, 2.1%), and others (online supplemental figure 1A). Among the patients (n=257), the most common cancer type with TMB-H was CRC (n=108, 42.0%), followed by GC (n=49, 19.1%), bladder cancer (BLCA, n=21, 8.2%), cholangiocarcinoma (CCC, n=21, 8.2%), NSCLC (n=17, 6.6%), melanoma (n=8, 3.1%), GBC (n=7, 2.7%), and others (n=26, 10.1%). We further divided the TMB categories into TMB-L (<10 mt/Mb, n=1487) versus TMB intermediate high (TMB-IH, ≥10 and <16 mt/Mb, n=186), and TMB very high (TMB-VH, ≥16 mt/Mb, n=71) (figure 1B). In the TMB-IH cohort (n=186), there were 84 (45.1%) CRC, 29 (15.6%) GC, 16 (8.6%) CCC, 13 (7.0%) NSCLC, 12 (6.5%) BLCA, and other cancer types (n=26, 14.0%), and 77.4% (n=144) of the patients had TMB scores between 10 and 13. All the TMB-IH patients had MSS. In the TMB-VH cohort (n=71), there were 24 (33.8%) CRC, 20 (28.2%) GC, 9 (12.7%) BLCA, 5 (7.0%) CCC, 4 (5.6%) NSCLC, and other cancer types (n=9, 12.7%). Of note, 34 of the 71 TMB-VH patients (47.9%) had high MSI (MSI-H) using TSO500 (CRC (n=18, 47.4%); GC (n=13, 34.2%); CCC (n=4, 10.5%); BLCA (n=1, 2.6%); metastases of unknown origin (MUO) (n=1, 2.6%); uterine cancer (n=1, 2.6%)). Finally, we compared PD-L1 (CPS score) between the TMB-H, TMB-IH, and TMB-L groups, which showed a positive correlation between high PD-L1 and TMB-H scores (p=0.04) (figure 1C).

Supplemental material

Figure 1

Sample classification and TMB. (A) Clinical features (cancer type, MSI, and HRD) and PDL-1 expression profiles in TMB-H and TMB-L groups. (B) Distribution of TMB scores in the three groups was categorized by TMB score. (C) Correlation between PD-L1 combined proportion and TMB scores. HRD, homologous recombination deficiency; MSI, microsatellite instability; TMB, tumor mutation burden; TMB-H, high TMB; TMB-L, low TMB.

Tumor mutation burden status and response to anti-PD-(L)1 therapy

The distribution and median TMB scores for each cancer type are presented in figure 2A and online supplemental table 1. The median TMB score was lowest for thyroid cancer (1.6) and highest for bladder cancer (7.8). Across tumor types, the response rates to anti-PD-L1 therapy were substantially higher in GC (71.4% vs 25.8%, p=0.007; TMB-H vs TMB-L, respectively), GBC (50.0% vs 12.5%, p=0.003; TMB-H vs TMB-L, respectively), head and neck cancer (HNC, 50.0% vs 11.1%, p<0.001; TMB-H vs TMB-L, respectively), and melanoma (71.4% vs 50.7, p=0.027; TMB-H vs TMB-L, respectively) with statistical significance (figure 2B). Similarly, the median TMB score in responders to anti-PD-L1 therapy was higher in patients with GBC (p=0.036) and HNC (p=0.00.38) (figure 2C). Across all tumor types, PD-L1 CPS-positive patients responded better to anti-PD-L1 therapy than PD-L1 CPS-negative patients (figure 2D). In the TMB-VH cohort (≥16 mt/Mb), a higher response to anti-PD-L1 therapy (50.0%) was observed in PD-L1-negative patients compared with the PD-L1-positive patients in the TMB-IH/L cohort (34.5%) (figure 2E). In addition, within the TMB-H cohort (≥10 mt/Mb), MSI-H patients responded to anti-PD-L1 therapy better than non-MSI-H patients (p=0.042) (figure 2F). The detailed characteristics of MSI-H patients who were treated with anti-PD-(L)1 therapy are described in online supplemental table 2. Finally, in the survival analysis, all the patients with TMB-VH demonstrated longer OS and PFS following anti-PD-L1 therapy compared with patients with TMB-L (TMB-VH vs TMB-L; not reached vs 418 days, p=0.03; OS: not reached vs 147 days, p<0.01 for PFS) (figure 2G,H). However, there were no statistically significant differences in the OS or PFS following anti-PD-L1 treatment between the TMB-IH and TMB-L cohorts or the TMB-H and TMB-L cohorts.

Supplemental material

Figure 2

Variables affecting response rate to ICI treatment. (A) Distributions of TMB across 1744 samples from various cancer types, with a median of the total sample (black dot) and TMB-H sample (red triangle) (TMB score was log10-transformed). (B) Response rates to ICI treatment in TMB-H and TMB-L groups. (C) Comparison of the distribution of TMB score between TMB-H and low TMB (TMB-L) groups in various cancer types. (D) Response rates depend on PD-L1 positivity and TMB score (cut-off: 10 mt/Mb). (E) Response rates depend on PD-L1 positivity and TMB score (cut-off: 16 mt/Mb). (F) Response rates depend on MSI and TMB scores (cut-off: 10 mt/Mb). Analysis of OS (G) and PFS (H) of the ICI treatment based on TMB. AOV, ampulla of Vater; BLCA, bladder cancer; CA, cancer; CCC, cholangiocarcinoma; CRC, colorectal cancer; GBC, gallbladder cancer; GC, gastric cancer; GIST, gastrointestinal stromal tumor; HCC, hepatocellular cancer; HNC, head and neck cancer; ICI, immune checkpoint inhibitor; Mb, megabase; M.S.F.T, malignant solitary fibrous tumor; mt, mutation; MUO, metastases of unknown origin; NET, neuroendocrine tumor; NSCLC, non-small cell lung cancer; OC, ovarian cancer; OS, overall survival; PACA, pancreas adenocarcinoma; PFS, progression-free survival; TMB, tumor mutation burden; TMB-H, high TMB; TMB-L, low TMB.

Among the study population, 426 patients were treated with anti-PD-(L)1 treatment and are available for clinical outcomes (figure 3A). Thirty-two of the 71 patients with TMB-VH received anti-PD--(L)1 treatment (figure 3B,C). Excluding the MSI-H patients (n=34), there were 33 TMB-VH/MSS (n=33, 47.9%) and 4 TMB-VH/low MSI (MSI-L, n=4, 5.6%) patients, including 18 (47.4%) CRC, 13 (34.2%) GC, 4 (10.5%) CCC, and 3 other cancer types. Next, we analyzed the confounding factors using multivariate analysis (figure 3D). While TMB-VH, TMB-H, and MSI-H were significant variables in predicting the response to anti-PD-L1 therapy, multivariate analysis demonstrated that only TMB-VH and MSI-H were significant independent predictors of the response to anti-PD-(L)1 therapy, indicating that TMB-VH and MSI-H were two important factors in predicting the response to anti-PD-(L)1 therapy, regardless of the PD-L1 status in our data set. The benefit of TMB-VH in predicting the response to anti-PD-(L)1 therapy was greater when combined with PD-L1 expression (relative risk (RR) = 1.87, p=0.002) and MSI-H (RR=3.97, p=0.002) statuses. Additional details are provided in online supplemental figure 2.

Supplemental material

Figure 3

Statistics of patients who received ICI treatment in the TMB-VH group. (A) Information about the best response and PD-L1 in patients undergoing ICI therapy. (B) The cancer types of patients who received ICI therapy and had TMB-VH status. (C) Histogram by TMB scores (left panel), the proportion of TMB-VH subgroup (middle panel), and the number of patients in each cancer type (right panel). (D) Relative risk about responder/non-responder by single or multiple factors. ICI, immune checkpoint inhibitor; TMB-VH, very high tumor mutation burden.

Correlation between TMB status and other genomic aberrations

The number of mutations in the HR genes substantially increased as TMB scores increased across the tumor types (figure 4A). The probability of mutations in the HR genes was higher in the TMB-VH group for the following genes: ATRX (p<0.0001), ARID1A (p<0.0001), PTEN (p<0.0001), ATM (p<0.0001), BRCA1 (p<0.0001), BRIP1 (p<0.0001), FANCF (p=0.031), NBN (p<0.0001), BLM (p<0.0001), FANCD2 (p<0.0001), BARD1 (p<0.0001), FANCA (p<0.0001), and RAD50 (p<0.024) (figure 4B). Other HR genes unrelated to TMB are shown in online supplemental figure 3. We compared the number of mutations in other genes in the TMB-VH group with the TMB-IH or TMB-L group, presenting more mutations occurring in the TMB-VH group, regardless of cancer type (figure 4C). In addition, we compared the mean CN of the top 20 genes in diverse cancer types according to the TMB status (online supplemental figure 4A). The average CNs of CCNE1, KRAS, and JAK2 were relatively higher in the TMB-VH group, but no significant trend was observed when the analysis was conducted using the overall CN (online supplemental figure 4B,C). Additional details are shown in online supplemental figure 4D–G.

Supplemental material

Supplemental material

Figure 4

Characteristics of HRD and mutations in other genes in the TMB-VH group. (A) Distribution of the TMB score depends on the degree of mutation in homologous recombination (HR) genes. (B) Estimated ORs with 95% CI in each HR gene. (C) Several SNVs per sample in TMB-VH, TMB-IH, and TMB-L groups. HR, homologous recombination; HRD, homologous recombination disorder; L-TMB, low TMB; SNV, single nucleotide variant; TMB, tumor mutation burden; TMB-IH, intermediate high MB; TMB-VH, very high TMB.

Immune cell type profile according to response to ICIs

We focused on the tumor immune environment of tumors in the responder (n=2) and non-responder (n=6) groups within the TMB-VH/MSS cohort (n=8) who had tumor tissue available for further analysis. The DSP analysis was performed to profile immune cell infiltration in the tumor. Patients who responded to anti-PD-L1 therapy had more enriched peritumoral immune cell infiltration (figure 5A–C). The responder and non-responder groups differed not only in the expressed immune cell types but also in the locations of immune cells. Immune cells expressed in the responder group infiltrated the tumor regions or were located near the tumor regions (figure 5A and online supplemental figure 5A,B). The non-responder group had low expression of immune cells infiltrating the tumor region or had immune cells located far from the tumor regions (figure 5A and online supplemental figure 5C–H). A dramatic responder to pembrolizumab was a patient in his 50s with small bowel adenocarcinoma who had MSS, which was also confirmed by multiplex PCR analysis, TMB-H tumor (426.8 mt/Mb), and high PD-L1 CPS 20. The patient failed to respond to first-line irinotecan/5-fluorouracil/leucovorin (FOLFIRI) and received eight cycles of pembrolizumab as part of the practice. The patient showed complete metabolic remission to the eight cycles of pembrolizumab and was still in CR for 3 years after the treatment at the time of this study (figure 5D).

Supplemental material

Figure 5

Characteristics of the tumor microenvironment in responders with TMB-H (A) Multiplex immunofluorescence images of patients with a high TMB score (left panel: responder; right panel: non-responder). (B) Immune cell abundance of responder and non-responder group. (C) Expression pattern of immune cells was significantly differentially expressed in the responder and non-responder groups. (D) CT images of the patients are in the left lower box. (A) Left and right CT images are before and after ICI therapy, respectively. CPS, combined positive score; ICI, immune checkpoint inhibitor; MSS, microsatellite stability; TMB, tumor mutation burden, TMB-H, high TMB.

In the DEP analysis, the levels of immune checkpoint proteins (LAG3, PD-L2, and B7-H3) were higher in the non-responder group, and IDO1 expression was increased in the responder groups (figure 6A), which is positively correlated with CD8+ T-cell expression.11 Most of the DEPs were related to immune cell regulation (figure 6B), indicating that immune cell regulation pathways were differentially activated between the responder and non-responder groups. Immune cell enrichment was also analyzed based on protein expression. There were more natural killer cells (p=0.04), proteins related to T-cell proliferation (p<0.01), cytotoxic T cells (p<0.01), memory T cells (p<0.01), and naïve memory T cells (p<0.01) in the responder TMB-H cohort than in the non-responder TMB-H cohort (figure 6C and online supplemental figure 6). In contrast, exhausted T cells (p<0.01) and M2 macrophages (p<0.01) were substantially increased in the baseline tumor specimens from the non-responder group (figure 6C).

Supplemental material

Figure 6

Immunological difference between responder and non-responder. (A) Differently expressed proteins (DEPs) between responder and non-responder groups. Red spots represent significantly expressed proteins. (B) Functional enrichment analysis of genes that encode DEPs. (C) Box plot showing the different proportions of immune cells between the responder and non-responder groups. TMB, tumor mutation burden; TMB-VH, very high TMB.

The distance between immune and tumor cells was significantly smaller in the responder group than in the non-responder group (online supplemental figure 7A–H). The average distance from T cells to tumor cells was 48.59 µm in the responder group and 104.51 µm in the non-responder group (p<0.01; online supplemental figure 7I).

Supplemental material


TMB is a candidate biomarker for predicting the response to anti-PD-L1 treatment because a higher number of mutations increases the levels of antigenic peptides (online supplemental figure 8).12 13 To the best of our knowledge, this is one of the largest studies to report the incidence of TMB-H using the TSO500 assay in real-world clinical practice. Among 1744 patients, we identified that the incidence of TMB-H (≥10 mt/Mb) using the TSO500 assay was 14.7% (n=257). The most common cancer type with TMB-H was CRC (n=108, 42.0%), followed by GC (n=49, 19.1%), BLCA (n=21, 8.2%), CCC (n=21, 8.2%), NSCLC (n=17, 6.6%), melanoma (n=8, 3.1%), GBC (n=7, 2.7%), and others (n=26, 10.1%), which is consistent with a previous report. We found that TMB-H predicts the response to anti-PD-(L)1 therapy across some tumor types, such as GC, GBC, HNC, and melanoma. However, by defining the TMB-VH group and increasing the cut-off to 16 mt/Mb, the predictive value of TMB-VH for anti-PD-L1 therapy was greater in terms of OS and PFS.

Supplemental material

The cohorts were divided into three groups: TMB-VH (≥16 mt/Mb), TMB-IH (10≥and <16 mt/Mb), and TMB-L (<10 mt/Mb). Based on the recent approval of pembrolizumab and atezolizumab in FoundationOneCDx TMB-H (≥10 mt/Mb),6 14 it is important to show the concordance of TMB between different target sequencing panels. In this study, an additional subgroup, TMB-VH, used a cut-off of 16 mt/Mb and observed a clear benefit from anti-PD-(L)1 therapy. One plausible explanation for the TMB-H cut-off of 10 mt/Mb not being a significant predictive factor for all the cancer types in our tumors analyzed with the TSO500 assay would be the under-reflected germline single nucleotide polymorphism (SNP) of Asian ethnicity in the TSO500 assay database. A recent large study on the genetic ancestry of >2000 patients with solid tumor types treated with immunotherapy demonstrated that there might be inflation in TMB, especially in African and Asian patients, for the same reason.15 Hence, future studies should extend the germline SNPs from Korean/Asian countries in the assay and correct TMB status in tumor specimens using conventional assays.

In addition to TMB, the expression of PD-L1 on tumor cells and immune cells has become a widely used predictive biomarker for responsiveness to anti-PD-(L)1 therapy in several cancer types.16–20 However, based on the observation of an unsatisfactory response to anti-PD-(L)1 therapy in some patients despite TMB-VH or PD-L1 positivity, we focused on the differences in the tumor microenvironment. From our DSP analysis results, we found that patients with a high proportion of infiltrated T cells (cytotoxic and memory T cells) close to the tumor benefited from the anti-PD-(L)1 therapy. The tumor microenvironment, which comprises different cell groups of the immune system and their interactions in the tumor microenvironment niche, has been shown to play a key role in carcinogenesis, cancer progression, and responses to immunotherapy.21 T Cells undergo dynamic changes in status from activated to exhausted cells in the tumor microenvironment of cancer patients receiving anti-PD(L)−1 therapy.22–24 Although limited by the small number of patients, our DSP analysis showed that each patient’s tumor within the TMB-H cohort and a high PD-L1 score had a different tumor microenvironment at baseline, which may have contributed to the optimized response to anti-PD-(L)1 therapy. Finally, our findings support the hypothesis that to increase the response rate to anti-PD-(L)1 treatment, we may need to consider three factors: the number of immune cells, the type of immune cells expressed, and the location of immune cells in the baseline tissue specimen before treatment, which needs further validation.

Our study has some limitations. First, this study was conducted retrospectively, which potentially harbors inherent bias, such as the proportion of cancer types or lines of anti-PD-(L)1 therapy. Second, in some cancer types, the predictive value of TMB for PD-(L)1 therapy is still under debate, with inconsistent data. This might also be explained by our results showing that in addition to TMB status, TME also plays a pivotal role in the response to anti-PD-(L)1 therapy at a higher level. Lastly, we observed a difference in the ranking of cancer types with TMB compared with previous reports.25 One of the major explanations for this could be the difference in SNP based on ethnicity. Further efforts in the calculation of TMB should be seriously considered for the clinical utility of TMB using a non-normal sample-matched target sequencing panel. In addition, the number of patients enriched in certain cancer types due to NGS (Next generation sequencing) conducted in a certain pattern in real-world clinical practice, such as NGS performed after failure of standard therapy, might affect the overall distribution of TMB based on the cancer type.

Overall, in a comparably large real-world cohort, we demonstrated that TMB-VH defined with a higher cut-off (such as 16 mt/Mb) using the TSO500 assay could be an important biomarker for predicting the response to anti-PD-L1 therapy, especially in Asian patients. In addition, we observed that the clinical outcome of anti-PD-(L)1 therapy was also affected by the quantity and spatial disposition of immune cells, even in the TMB-VH population.

Data availability statement

Data are available upon reasonable request. The data generated in this study are available on request from the corresponding author.

Ethics statements

Patient consent for publication

Ethics approval

This study was performed in accordance with the principles of the Declaration of Helsinki and Korean Good Clinical Practice guidelines. The collection of specimens and associated clinical data used in this study was approved by the Institutional Review Board of the SMC (IRB File No. 2021-08-046). The requirement for informed consent was waived because of the retrospective nature of the study.


Supplementary materials


  • JJ and YJH contributed equally.

  • Contributors SP: study conception; design; protocol development; patient enrollment; generated, interpreted, and verified the data; conducted the statistical analysis; wrote the paper; and act as gurantor. JJ, YJH, SP: generated the data, conducted the statistical analysis, interpreted and verified the data, and wrote the paper. All authors had full access to the complete study data. The corresponding author assumes the final responsibility for the decision to submit the manuscript.

  • Funding This research was supported by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health and Welfare, Republic of Korea (grant number: HR20C0025).

  • Competing interests None declared.

  • Provenance and peer review Not 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.