Of the 102 studies, we identified 27 studies for NSCLC, 40 studies for melanoma, 10 studies for urothelial cancer, and 5 studies for renal cell cancer indications. Fewer studies were identified for other cancer types, such as squamous-cell carcinoma of the head and neck (
n = 2), colorectal cancer (
n = 3), gastric cancer (
n = 2), breast cancer (
n = 1), Hodgkin’s disease (
n = 1), Merkel cell cancer (
n = 1), small cell lung cancer (n = 1), and pancreatic cancer (
n = 1). Some studies presented data for multiple indications. Table
1 lists the indications we identified for each drug of interest. The studies were highly heterogenous, investigating a range of biomarker cutoffs with a variety of biomarker assays. The study characteristics and reported outcomes of all the studies of interest are presented in Additional file
1: Table S1.
Table 1
Indications Included in Identified Studies by Intervention
CTLA-4 inhibitors |
Ipilimumab | Melanoma (n = 25), NSCLC (n = 3), mCRC (n = 1), SCLC (n = 1), RCC (n = 1), pancreatic (n = 1) |
Tremelimumab | Melanoma (n = 3) |
PD-1 inhibitors |
Nivolumab | NSCLC (n = 14), melanoma (n = 14), RCC (n = 3), urothelial (n = 2), SCCHN (n = 1), GC (n = 1), mCRC (n = 1), Hodgkin’s disease (n = 1), SCLC (n = 1) |
Pembrolizumab | Melanoma (n = 9), NSCLC (n = 6), urothelial (n = 2), mCRC (n = 2), GC (n = 1), breast cancer (n = 1), SCCHN (n = 1) |
PD-L1 inhibitors |
Atezolizumab | NSCLC (n = 6), melanoma (n = 2), urothelial (n = 3), RCC (n = 2) |
Avelumab | NSCLC (n = 1), Merkel cell cancer (n = 1), urothelial (n = 1) |
Durvalumab | NSCLC (n = 2), urothelial (n = 1) |
NSCLC
We identified 27 studies (69 references, including 3 pooled analyses) that presented outcome data of interest for NSCLC. Eleven studies presented data for nivolumab as treatment, 5 for atezolizumab, and 3 for pembrolizumab; the remaining studies reported data on other treatments or mixed treatments.
Six studies reported OS or PFS data for populations using TMB as a biomarker, as shown in Table
2. The cutoff points used included < 10, ≥ 10, < 12, ≥ 12, ≥ 13, < 14, ≥ 14, ≥ 16, < 16, < 20, and ≥ 20 mutations per megabase; some studies also reported TMB as low, medium, or high. Due to the varying definitions of TMB, it is difficult to draw direct comparisons between studies.
Table 2
Tumor Mutation Burden as Predictor of Non-small Cell Lung Cancer Outcome: OS and PFS Data
CheckMate 026 Carbone et al. (2017) [ 17] Socinski et al. (2016) [ 18] | High TMB | NIVO 3 mg/kg Q2W | 47 | 18.3 (11.4-NE) | 1.1 (0.64–1.88) | 9.7 (5.1-NE) | 0.62 (0.38–1.0) |
Platinum-based chemotherapy Q3W | 60 | 18.8 (11.3-NE) | 5.8 (4.2–8.5) |
Low or medium TML | NIVO 3 mg/kg Q2W | 111 | 12.7 (9.9–16.1) | 0.99 (0.71–1.4) | 4.1 (2.8–5.4) | 1.82 (1.3–2.55) |
Platinum-based chemotherapy Q3W | 94 | 13.2 (9.5–15.2) | 6.9 (5.5–8.6) |
CheckMate 227 Hellmann et al. (2018) [ 4] | TMB ≥ 10 mutations per mb | NIVO + IPI | 139 | NR | NR | 7.2 (5.5–13.2) | 0.58 (97.5% CI, 0.41–0.81) |
Chemotherapy | 160 | NR | NR | 5.5 (4.4–5.8) |
TMB < 10 mutations per mb | NIVO + IPI | 191 | NR | NR | 3.2 (2.7–4.3) | 1.07 (0.84–1.35) |
Chemotherapy | 189 | NR | NR | 5.5 (4.3–5.6) |
OAKa Rittmeyer et al. (2017) [ 6] Gadgeel et al. (2017) [ 19] Barlesi et al. (2016) [ 20] Gandara et al. (2017) [ 22] | TMB ≥ 10 | ATEZO vs. DTX | 251 | NR | 0.69 (NR) | NR | 0.73 (NR) |
TMB ≥ 16 | | 158 | NR | 0.64 (NR) | NR | 0.65 (NR) |
TMB ≥ 20 | 105 | NR | 0.65 (NR) | NR | 0.61 (NR) |
POPLARa Fehrenbacher et al. (2016) [ 23] Mazieres et al. (2016) [ 25] Vansteenkiste et al. (2015) [ 26] Gandara et al. (2017) [ 22] | TMB ≥ 10 | ATEZO vs. DTX | 96 | NR | 0.59 (NR) | NR | 0.68 (NR) |
TMB ≥ 16 | 63 | NR | 0.56 (NR) | NR | 0.57 (NR) |
TMB ≥ 20 | 42 | NR | 0.51 (NR) | NR | 0.58 (NR) |
Yaghmour (2016) | TML: top quintile | ≥ First line, NIVO or IPI | 50 (overall patients) | NR | 3.29 (0.75–25.53) | NR | NR |
TML: other quintiles | NR | NR | NR |
B-F1RST | Blood-based TMB ≥ 12 | ATEZO | 22 | NR | NR | 3 | 0.95 (90% CI, 0.55–1.63) |
Blood-based TMB < 12 | 36 | NR | NR | 3.2 |
Blood-based TMB ≥ 14 | 14 | NR | NR | 3.4 | 0.73 (90% CI, 0.39–1.39) |
Blood-based TMB < 14 | 44 | NR | NR | 3.2 |
Blood-based TMB ≥ 16 | 11 | NR | NR | 9.5 | 0.49 (90% CI, 0.23–1.04) |
Blood-based TMB < 16 | 47 | NR | NR | 2.8 |
Blood-based TMB ≥ 20 | 8 | NR | NR | 9.5 | 0.23 (90% CI, 0.08–0.62) |
Blood-based TMB < 20 | 50 | NR | NR | 2.7 |
The most commonly applied TMB cutoff points were ≥ 10, ≥ 16, and ≥ 20 mutations per megabase. However, the studies that used these cutoff points used different definitions of TMB (blood or tissue based). B-F1RST [
29] reported the greatest increase of median PFS (9.5 months) at the cutoff point ≥16 when using cutoff points ranging from ≥12 to ≥20.
The CheckMate 227 study [
4] reported a median PFS of 3.2 and 7.2 months for TMB < 10 and TMB ≥ 10, respectively, for patients treated with first-line nivolumab 3 mg/kg plus ipilimumab 1 mg/kg. Nivolumab 3 mg/kg also was the first-line treatment used in CheckMate 026 [
17]; the median PFS was 4.1 months for low or medium TMB and 9.7 for high TMB. A higher OS (18.3 vs. 12.7 months) was reported for the high-TMB group than for the low- or medium-TMB group. Interestingly, despite this study finding no association between PD-L1 expression and TMB, patients with both a high TMB and a PD-L1 expression of ≥50 had a higher response rate (75%) than patients with one (32–34%) or neither (16%) of these factors, suggesting that they are independent biomarkers predictive of response. It should be noted that the CheckMate 227 and CheckMate 026 studies used different methods to assess TMB (FoundationOne CDx assay and whole exome sequencing, respectively).
Two studies looked at TMB in second-line therapy and beyond when comparing atezolizumab and docetaxel therapy. Both OAK [
6] and POPLAR [
23] studies used the cutoff points ≥ 10, ≥ 16, and ≥ 20, and both reported an inverse relationship between TMB and OS HR. The OAK and POPLAR studies both used blood-based approaches to assess TMB. The OS HRs for the individual TMB cutoffs differed between studies: in OAK, they were 0.69, 0.64, and 0.65, respectively, for the three cutoff points, while in POPLAR, they were 0.59, 0.56, and 0.51, respectively [
6,
23]. This difference could be attributed to the difference in population sizes or because patients with known EGFR or anaplastic lymphoma kinase mutations were excluded in Rittmeyer et al. [
6]. In addition, increasing TMB may be prognostic but not predictive; i.e., tumors with higher levels of TMB would be less responsive to chemotherapy. As no confidence intervals (CIs) were reported for either study, it is not possible to determine the degree of significance of the difference in OS HR results between the studies.
Finally, Yaghmour et al. [
28] investigated patients who had solid tumors, were treated with any checkpoint inhibitor, and had undergone next-generation sequencing. This study reported that OS was significantly higher in patients who were in the top quintile for TMB (hazard ratio [HR] = 5.78; 95% CI, 1.40–15.12). However, no significant difference was found in the population of patients who had NSCLC (
P = 0.205; HR = undefined [95% CI, 0.53–25.70]).
Sixteen studies reported OS or PFS data in patients with NSCLC and with PD-L1 expression as a biomarker, as shown in Additional file
1: Table S2. The cutoff values for PD-L1 expression in tumor and/or immune cells used included < 1%, < 5%, < 10%, < 50, 1 to 49%, ≥ 1%, ≥ 5%, ≥ 10%, and ≥ 50%. Unfortunately, not all studies reported the PD-L1 expression cutoffs used. Additionally, study durations differed and, in some studies, the median OS or the upper limit of the CI was not reached.
The CheckMate 227 study [
4] reported OS and PFS data in patients with NSCLC and both PD-L1 expression and TMB status as biomarkers (Additional file
1: Table S3).
The median OS for first-line treatment with nivolumab was highest in the subgroup with PD-L1 expression ≥50% [
17]. In CheckMate 026 [
17], the median OS for PD-L1 expression ≥1% was 13.7 months with nivolumab 3 mg/kg, compared with 20.2 months with nivolumab 10 mg/kg as treatment in CheckMate 012 [
30]. The median OS for second-line treatment with nivolumab at a dose of 3 mg/kg ranged from 9.3 months to 17.7 months for patients with a PD-L1 expression ≥1%, 10.0 to 19.4 months for PD-L1 expression ≥5%, 11 to 19.9 months for PD-L1 expression ≥10%, and 8.7 to 10.5 months for PD-L1 expression < 1% in CheckMate 057 and 017 [
31,
32].
For second-line treatment with atezolizumab (1200 mg), the median OS ranged from 15.5 to 15.7 months for PD-L1 expression ≥1%, 15.5 to 16.3 months for PD-L1 expression ≥5%, 15.1 to 20.5 months for ≥50%, and 9.7 to 12.6 months for PD-L1 expression < 1% in OAK and POPLAR [
6,
23]. Rittmeyer et al. [
6] differentiated the patient population into squamous and nonsquamous NSCLC. Comparing the median OS for PD-L1 expression ≥50% showed that survival appeared to be better in patients with nonsquamous NSCLC (22.5 months) than in patients with squamous NSCLC (17.5 months). Similar differences were shown for the other PD-L1 expression cutoffs.
Melanoma
We identified 40 studies (53 references) that presented outcome data of interest for melanoma; however, limited OS and PFS data were available. Only 3 studies reported OS or PFS data using TMB as a biomarker (Table
3), while 5 studies reported OS or PFS data using PD-L1 expression (Additional file
1: Table S4).
Table 3
Tumor Mutation Burden as Predictor of Melanoma Outcome: OS and PFS Data
Johnson et al. (2016) [ 33] | High (> 23.1 mutations per mb) | NIVO, PEM, and ATEZO | 65 | NE | NR | NE | NR |
Intermediate (3.3–23.1 mutations per mb) | 65 | 9.9 (NR) | NR | 2.9 (NR) | NR |
Low (< 3.3 mutations per mb) | 65 | 12.3 (NR) | NR | 2.8 (NR) | NR |
Roszik et al. (2016) [ 34] | Predicted TML ≤ 100 | IPI | 19 | 19.14 (NR) | 0.35 (0.16–0.77) | NR | NR |
Predicted TML > 100 | 57 | Undefined (NR) | NR | NR |
Yaghmour et al. (2016) [ 28] | TML: top quintile | NIVO, PEM, and IPI | 50 (overall patients) | NR | 3.29(0.75–25.53) | NR | NR |
TML: other quintiles combined | NR | NR | NR |
Yaghmour et al. [
28] reported that OS was higher in patients with a TMB in the top quintile (median genomic alterations = 16.5) than OS in patients with a TMB in the lower quintiles (median genomic alterations = 2) (HR = 3.29; 95% CI, 0.75–25.53). Patients were treated with nivolumab, pembrolizumab, or ipilimumab. Roszik et al. [
34] also found that OS was higher in patients treated with ipilimumab who had a high predicted TMB (> 100) compared to those with a low predicted TMB (≤100) (median = undefined vs 582 days,
P < 0.006). Additionally, Johnson et al. [
33] and Yaghmour et al. [
28] found that patients with a high TMB (> 23.1 mutations/megabase) had higher OS and PFS than those with intermediate TMB (3.3–23.1 mutations/megabase) or low TMB (< 3.3 mutations/megabase) (OS: median = not reached vs. 300 days vs. 375 days,
P < 0.001; PFS: median = not reached vs. 89 days vs. 86 days,
P < 0.001). Patients were treated with nivolumab, pembrolizumab, or atezolizumab.
Two studies (KEYNOTE-001 and KEYNOTE-006) used a PD-L1 expression cutoff of 1%, both investigating pembrolizumab [
35‐
38]. The median OS was significantly higher in patients with PD-L1 expression ≥1% than in patients with PD-L1 expression < 1%, with an HR between 0.55 and 0.83 [
35,
36]. Three other studies (CheckMate 066, CheckMate 067, CheckMate 069) used a PD-L1 expression cutoff of 5% [
36,
39,
40], but the results are inconclusive.