Effects of different sequencing platforms, sequencing methods and races on TMB values
To study the effect of different sequencing platforms and methods on TMB calculation, data of 4126 tumor samples sequenced with different platforms and methods were obtained. As shown in Figure 1A, no significant difference in TMB calculation from different sequencing platforms, including Illumina and MGI, was found. TMB from panel sequencing was higher than whole-exome sequencing (WES), however, there was no significant difference. To further verify the effect of different sequencing methods on TMB calculation, public sequencing data from 3680 samples performed with different methods were analyzed (Figure 1B). TMB values calculated from whole-genome sequencing (WGS) were significantly lower than those calculated from WES and panel sequencing, which suggested that the TMB values may be affected by sequencing depth. Furthermore, TMB in different races was analyzed. As shown in Figure 1C, TMB values of the East Asian populations were significantly lower than that of European and American populations in both WES sequencing and panel sequencing.
TMB calculation was affected by the sequencing depth and tumor purity
To investigate the effect of sequencing depth on TMB calculation, TMB calculated from lung cancer at different panel sequencing depths were analyzed. As shown in Figure 2A, the TMB calculated from sequencing depths ≥ 500 was significantly higher than that calculated from sequencing depths < 500. To determine the effect of tumor purities on TMB calculation, the distribution of TMB values with different tumor purities was analyzed. As shown in Figures 2B and 2C, the TMB values were more stable with tumor purity estimated by HE staining between 0.2-0.8 or tumor purity estimated by NGS ≥ 0.4.
High correlation of TMB calculated with the optimized SGZ algorithm and paired normal-tumor sequencing in East Asian populations
TMB calculation requires paired normal samples to remove germline mutations, which increases the cost of sequencing. In addition, in terms of clinical accessibility, paired normal samples are sometimes unavailable, which limits the clinical application of TMB. The SGZ algorithm for TMB calculation with tumor-only sequencing samples was designed based on European and American populations. For East Asian populations, the TMB calculated by the SGZ algorithm had a low correlation with the TMB calculated by the method of paired normal-tumor sequencing (Figure 3A). In order to calculate TMB from tumor-only sequencing samples in Asian populations, the SGZ algorithm was optimized. As shown in Figure 3B, we added the mutation filtering step, and constructed a background library with normal samples from East Asian patients to filter germline mutations, which ultimately reduced the false positives of TMB. To verify the accuracy of TMB calculation with an optimized algorithm, TMB calculated from the methods of the optimized algorithm and paired normal-tumor sequencing in sixty-two samples were compared. As shown in Figures 3C and 3D, their correlation coefficient is 0.95 and 82.7% of the mutations identified from the method of paired normal-tumor sequencing could be identified with the optimized SZG algorithm. These results demonstrate the accuracy of TMB calculation from the tumor-only sequencing with the optimized SGZ algorithm in East Asian populations.
Identification of the TMB cut-off for predicting immunotherapy efficacy in the training cohort
The cut-off of TMB for predicting the efficacy of immunotherapy in European and American populations is 10 mut/Mb [20]. However, since the TMB of East Asian populations is lower than that of European and American populations, a TMB cut-off of 10 mut/Mb may not be suitable for East Asian populations. To determine the cut-off of TMB for predicting immunotherapy efficacy in the East Asian population,
tumor samples from sixty-six lung patients treated with ICIs were retrospectively collected as a training cohort. TMB was calculated with the optimized SGZ algorithm. The performance of TMB for predicting patient durable clinical benefit was analyzed with a ROC curve and Log-rank test. As shown in Figure 4A, the optimal cut-off of TMB was 7.19 mut/Mb with AUC = 0.742, and validated with Log-rank analysis (Figure 4B). Furthermore, the response rate and survival period were higher in patients with TMB ≥ 7.19 mut/Mb than in those with TMB < 7.19 mut/Mb, and the TMB cut-off of 7.19 mut/Mb is better than the TMB cut-off of 10 mut/Mb in East Asian populations (Figure 4C-4F). To further investigate the role of TMB in predicting the efficacy of immunotherapy, the effects of TMB, medication type, tumor type and age group on patient survival were analyzed through multi-factor cox-regression. It was found that TMB was a favorable factor for patient survival, while other factors had no significant effect (Figure 4G).
Validation of TMB cut-off in the test cohort
To further validate the TMB cut-off of 7.19 mut/Mb, genomic and clinical data of sixty-nine East Asian NSCLC patients treated with ICIs were collected [30]. Consistent with the above results, the survival period of patients with TMB ≥ 7.19 mut/Mb was longer than those with TMB < 7.19 mut/Mb (Figure 5A). Furthermore, the predicting efficacy with the TMB cut-off of 7.19 mut/Mb is better than the TMB cut-off of 10 mut/Mb (Figure 5B).