Combined PD-1, BRAF and MEK inhibition in advanced BRAF-mutant melanoma: safety run-in and biomarker cohorts of COMBI-i

Reinhard Dummer 1 , Celeste Lebbé2, Victoria Atkinson3, Mario Mandalà4, Paul D. Nathan5, Ana Arance6, Erika Richtig7, Naoya Yamazaki8, Caroline Robert 9, Dirk Schadendorf 10, Hussein A. Tawbi 11, Paolo A. Ascierto 12, Antoni Ribas 13, Keith T. Flaherty 14, Neha Pakhle15, Catarina D. Campbell16, Daniel Gusenleitner16, Aisha Masood17, Jan C. Brase18, Eduard Gasal17 and Georgina V. Long 19

Immune and targeted therapies achieve long-term survival in metastatic melanoma; however, new treatment strate- gies are needed to improve patients’ outcomes1,2. We report on the efficacy, safety and biomarker analysis from the single-arm safety run-in (part 1; n = 9) and biomarker (part 2; n = 27) cohorts of the randomized, placebo-controlled, phase 3 COMBI-i trial (NCT02967692) of the anti-PD-1 anti- body spartalizumab, in combination with the BRAF inhibitor dabrafenib and MEK inhibitor trametinib. Patients (n = 36) had previously untreated BRAF V600-mutant unresect- able or metastatic melanoma. In part 1, the recommended phase 3 regimen was identified based on the incidence of dose-limiting toxicities (DLTs; primary endpoint): 400 mg of spartalizumab every 4 weeks plus 150 mg of dabrafenib twice daily plus 2 mg of trametinib once daily. Part 2 characterized changes in PD-L1 levels and CD8+ cells following treatment (primary endpoint), and analyzed additional biomarkers. Assessments of efficacy and safety were key secondary end- points (median follow-up, 24.3 months). Spartalizumab plus dabrafenib and trametinib led to an objective response rate (ORR) of 78%, including 44% complete responses (CRs). Grade ≥3 treatment-related adverse events (TRAEs) were experienced by 72% of patients. All patients had temporary dose modifications, and 17% permanently discontinued all three study drugs due to TRAEs. Early progression-free sur- vival (PFS) events were associated with low tumor mutational burden/T cell–inflamed gene expression signature (GES) or high immunosuppressive tumor microenvironment (TME) GES levels at baseline; an immunosuppressive TME may also preclude CR. Overall, the efficacy, safety and on-treatment biomarker modulations associated with spartalizumab plus
dabrafenib and trametinib are promising, and biomarkers that may predict long-term benefit were identified.
First-line treatment with anti-programmed death receptor 1 (anti-PD-1) with or without anti-cytotoxic T-lymphocyte-associated protein 4 immune checkpoint inhibitor therapy was associated with 5-year overall survival (OS) rates of 39–44% with monotherapy and 52% with nivolumab plus ipilimumab in phase 3 trials in patients with metastatic melanoma1,3. In patients with BRAF-mutant metastatic melanoma treated with BRAF plus MEK inhibitor targeted thera- pies, 5-year OS rates in phase 3 trials ranged from 31 to 34% (refs. 2,4). While this represents substantial improvement over historical survival rates5,6, many patients continue to experience disease progression, highlighting the need for new strategies to further improve outcomes. Preclinical data and clinical analyses of patient biospecimens showed that BRAF plus MEK inhibitors prime the TME early after treatment initiation, by increasing T cell infiltration and PD-1/ PD-L1 expression and decreasing immunosuppressive cytokines7–12. These changes to the TME may help to enhance antitumor responses driven by checkpoint blockade7,13,14. Moreover, patients with meta- static melanoma who derive clinical benefit from anti-PD-1 thera- pies have a median duration of response (DOR) of >4 years1,3, while BRAF and MEK inhibitors are associated with high response rates in patients with BRAF V600-mutant metastatic melanoma, with > 90% of patients achieving at least stable disease (SD)2,4,15. Therefore, the combination of a checkpoint inhibitor with targeted therapy may lead to enhanced antitumor activity compared with either type of treatment alone, increasing the depth and durability of responses. Consistently, treatment with an anti-PD-1 monoclonal antibody in combination with dabrafenib and trametinib demonstrated supe- rior antitumor activity compared with dabrafenib and trametinib
alone in preclinical studies7.

1University Hospital Zürich Skin Cancer Center, Zurich, Switzerland. 2APHP Hôpital Saint-Louis, Dermatology and CIC, Université de Paris, Paris, France. 3Greenslopes Private Hospital, Gallipoli Medical Research Foundation, University of Queensland, Greenslopes, Queensland, Australia. 4Papa Giovanni XXIII Cancer Center Hospital, Bergamo, Italy. 5Mount Vernon Cancer Centre, Northwood, UK. 6Hospital Clinic of Barcelona, Barcelona, Spain. 7Medical University of Graz, Graz, Austria. 8National Cancer Center Hospital, Tokyo, Japan. 9Gustave Roussy and Paris-Sud-Paris-Saclay University, Villejuif, France. 10University Hospital Essen, Essen and German Cancer Consortium, Heidelberg, Germany. 11The University of Texas MD Anderson Cancer Center, Houston, TX, USA. 12Istituto Nazionale Tumori IRCCS Fondazione ‘G. Pascale’, Naples, Italy. 13University of California, Los Angeles, Los Angeles, CA, USA. 14Massachusetts General Hospital Cancer Center and Harvard Medical School, Boston, MA, USA. 15Novartis Healthcare Private Limited, Hyderabad,
India. 16Novartis Institutes for BioMedical Research, Inc., Cambridge, MA, USA. 17Novartis Pharmaceuticals Corporation, East Hanover, NJ, USA. 18Novartis Pharma AG, Basel, Switzerland. 19Melanoma Institute Australia, The University of Sydney and Royal North Shore and Mater Hospitals, Sydney, New South
Wales, Australia. ✉e-mail: [email protected]

Clinical evidence to date also supports the premise that combin- ing checkpoint inhibitors with BRAF and MEK inhibitors may pro- long survival in patients with BRAF V600-mutant melanoma16–18. In a randomized phase 2 trial of 120 patients with BRAF-mutant metastatic melanoma, the anti-PD-1 antibody pembrolizumab in combination with dabrafenib and trametinib (KEYNOTE-022, NCT02130466) led to numerically longer PFS than the targeted therapies alone. Despite a favorable hazard ratio (0.66) and P value (0.043), the protocol-specified criteria for statistical significance were not met at the time of this primary analysis16. However, a subsequent update with approximately 20 months of additional follow-up demonstrated sustained and increasing advantages of PFS, DOR and OS with pembrolizumab plus dabrafenib and tra- metinib versus dabrafenib plus trametinib alone19. More recently, the phase 3 IMspire150 trial (NCT02908672) of the anti-PD-L1 antibody atezolizumab plus the BRAF and MEK inhibitors vemu- rafenib and cobimetinib in patients with BRAF-mutant metastatic melanoma met its primary endpoint, with significant improvement in investigator-assessed PFS (hazard ratio = 0.78, P = 0.0249) and clinically meaningful improvement in DOR compared with tar- geted therapy alone20.
Spartalizumab is an anti-PD-1 monoclonal antibody shown to have antitumor activity in patients with advanced noncutaneous and cutaneous melanoma, many of whom had previously received multiple lines of treatment, demonstrating an ORR of 27.9% and a median PFS of 4.7 months21, consistent with other trials of anti-PD-1 monotherapy in previously treated patients22–24. Here, we report findings from parts 1 (safety run-in) and 2 (biomarker cohort) of the COMBI-i trial (NCT02967692) of spartalizumab in combination with dabrafenib and trametinib in patients with unre- sectable or metastatic BRAF V600-mutant melanoma (Extended Data Fig. 1). In the safety run-in, nine previously untreated patients with an Eastern Cooperative Oncology Group performance status (ECOG PS) of ≤1 and alanine and aspartate aminotransferase lev- els <2.5-fold the upper limit of normal were enrolled to determine the recommended phase 3 regimen (Methods). The initial dose level was spartalizumab 400 mg every 4 weeks plus the approved doses of dabrafenib (150 mg twice daily) and trametinib (2 mg once daily). The DLT observation period was 56 days; one event of dose-limiting hepatotoxicity (grade 3) was observed. This dosing was thus estab- lished as the recommended phase 3 regimen, and an additional 27 previously untreated patients with an ECOG PS of ≤2 were enrolled in the biomarker cohort.
Data from COMBI-i parts 1 and 2 were pooled at the patient level to assess efficacy and safety. At data cutoff (19 August 2019), median follow-up was 24.3 months (range, 20.8–29.5 months) and treatment was ongoing in 10 of 36 patients (28%). Baseline char- acteristics are summarized in Table 1; 56% of patients had stage IV M1c disease, 42% had elevated lactate dehydrogenase (LDH) levels and 56% had three or more organ sites with disease involvement. The investigator-assessed ORR was 78% (95% confidence interval (CI) 61–90%) (Table 2) and 33 patients experienced a reduction in target lesion size (Extended Data Fig. 2). CRs were observed in 16 of 36 patients (44%), while partial responses were observed in 33% (Table 2). Median DOR was not estimable (NE; 95% CI, 17 months-NE), and the 24-month DOR rate was 53% (95% CI, 29–73%) (Table 2 and Extended Data Fig. 3a). In patients with a CR, median DOR was NE (95% CI, 17 months-NE); the estimated 24-month duration of CR was 55% (95% CI, 22–79%). The baseline characteristics of complete and noncomplete responders are sum- marized in Supplementary Table 1.
Progression-free survival events were observed in 19 patients (53%), with a median PFS of 23 months (95% CI, 12 months-NE); 18 patients had disease progression and one died before documented progression. Estimated 24-month PFS was 41% (95% CI, 23–59%; Table 2 and Extended Data Fig. 3b). Nine patients (25%) had died

Table 1 | Baseline characteristics of patients enrolled in parts 1 and 2 of COMBI-i
Characteristic Part 1
(n = 9) Part 2
(n = 27) Parts 1 and 2
(n = 36)
Age, median (range)
(years)
Age, n (%) 45 (35–69)

7 (78)/
2 (22) 61 (23–76)

18 (67)/
9 (33) 55.5 (23–76)

25 (69)/11 (31)
Male/female, n (%) 7 (78)/
2 (22) 15 (56)/
12 (44) 22 (61)/
14 (39)
White, n (%) 9 (100) 24 (89) 33 (92)
ECOG PS, n (%)
0 7 (78) 19 (70) 26 (72)
1 2 (22) 8 (30) 10 (28)
AJCC 7 stage, n (%)
IIIC 0 2 (7) 2 (6)
IV M1a 2 (22) 6 (22) 8 (22)
IV M1b 3 (33) 3 (11) 6 (17)
IV M1c with normal
LDH level
IV M1c with elevated LDH level 2 (22)

2 (22) 5 (19)

11 (41) 7 (19)

13 (36)
BRAF mutation status, n (%)
V600E 8 (89) 21 (78) 29 (81)
V600K 1 (11) 3 (11) 4 (11)
V600, other 0 3 (11) 3 (8)
LDH level, n (%)
<1 × ULN 6 (67) 13 (48) 19 (53)
≥1 to <2 × ULN 3 (33) 6 (22) 9 (25)
≥2 × ULN 0 6 (22) 6 (17)
Sum of diameters, median (range) (mm) 42 (10–133) 61 (10–255) 57 (10–255)
No. of organ sites with disease, n (%)
<3 4 (44) 12 (44) 16 (44)
≥3 5 (56) 15 (56) 20 (56)
Table 1 includes baseline characteristics summarized for patients enrolled in parts 1 and 2 of COMBI-i, as well as the pooled dataset (n = 36). LDH levels from two patients were unavailable. BRAF mutation status is reported based on local testing. A V600K mutation in the presence of another V600 mutation, including V600E, was combined into the ‘V600K’ category. The ‘V600, other’ category includes V600 mutations other than V600E or V600K. AJCC 7, American Joint Committee on Cancer 7th edn.; ULN, upper limit of normal.

at the time of these analyses: eight due to progression and one due to cardiac arrest that was not considered related to study treatment. Median OS was NE, and the estimated 24-month OS rate was 74% (95% CI, 56–86%; Table 2 and Extended Data Fig. 3c). Subgroup analyses were conducted for patients with elevated baseline LDH levels (Table 2). Of 15 patients, 67% had an objective response and CRs were observed in 27%. Median PFS was 11 months (95% CI, 5–19 months), with an estimated 24-month PFS rate of 27% (95% CI, 8–50%). Median OS was NE (95% CI, 7 months-NE), and the estimated 24-month OS rate was 53% (95% CI, 25–74%).
Median exposure to the treatment regimen was 13.4 months (range, 2.5–29.4 months). All patients experienced at least one adverse event (AE), with 81% experiencing at least one AE of grade
≥3. TRAEs were observed in all patients, with 72% experiencing TRAEs of grade ≥3 (Supplementary Tables 2 and 3)—most com- monly pyrexia (17%), increased lipase (11%), neutropenia (11%),

Table 2 | Summary of RECIST responses to treatment and time-to-event analyses
Patients with measurable disease at baseline n = 36
Best overall response, n (%)
CR 16 (44)
PR 12 (33)
SD 6 (17)
PD 1 (3)
Unknown 1 (3)
Confirmed ORR (CR + PR), n (%) (95% CI)
DCR (CR + PR + SD), n (%) (95% CI) 28 (78) (61–90)
34 (94) (81–99)
DOR, median (95% CI) (months) NE (17-NE)
12-month rate (95% CI) (%) 80 (59–91)
24-month rate (95% CI) (%) 53 (29–73)
PFS, median (95% CI) (months) 23 (12-NE)
12-month rate (95% CI) (%) 67 (49–80)
24-month rate (95% CI) (%) 41 (23–59)
OS, median (95% CI) (months) NE (NE-NE)
12-month rate (95% CI) (%) 86 (70–94)
24-month rate (95% CI) (%) 74 (56–86)
Patients with LDH level ≥1 × ULN n = 15
Best overall response, n (%)
CR 4 (27)
PR 6 (40)
SD 3 (20)
PD 1 (7)
Unknown 1 (7)
Confirmed ORR (CR + PR), n (%) (95% CI)
DCR (CR + PR + SD), n (%) (95% CI) 10 (67) (38–88)
13 (87) (60–98)
DOR, median (95% CI) (months) NE (4-NE)
12-month rate (95% CI) (%) 65 (25–87)
21-montha rate (95% CI) (%) 52 (16–79)
PFS, median (95% CI) (months) 11 (5–19)
12-month rate (95% CI) (%) 33 (12–56)
24-month rate (95% CI) (%) 27 (8–50)
OS, median (95% CI) (months) NE (7-NE)
12-month rate (95% CI) (%) 67 (38–85)
24-month rate (95% CI) (%) 53 (25–74)
Table 2 includes a summary of response evaluation criteria in solid tumors (RECIST) responses and survival analyses for the pooled dataset (n = 36), as well as subgroup analyses in patients with an elevated LDH level (≥1 × ULN; n = 15). aThe 24-month DOR rate was not estimable for patients
with elevated LDH levels because all patients within this subpopulation either had an event or were censored before 24 months.

increased blood creatine phosphokinase (8%) and increased gamma-glutamyltransferase (8%). Adverse events requiring treat- ment with immunosuppressive medication were observed in 83% of patients, including 28% who received steroids for pyrexia management. TRAEs led to dose modification in all patients, including reduction of dabrafenib or trametinib (56% each) and interruption of dabrafenib or trametinib (94% each) or spartali- zumab (64%). Discontinuation of all three study drugs occurred in six patients (17%) (Supplementary Table 2) due to TRAEs, includ- ing immune-mediated hepatitis, paresthesia, hypokalemia, intersti- tial lung disease, increased alanine and aspartate aminotransferases,
increased gamma-glutamyltransferase and generalized exfolia- tive dermatitis, as reported by investigators. These data include 12 patients who received at least one dose of treatment before the introduction of a protocol amendment impacting pyrexia manage- ment (Methods). No treatment-related deaths were reported.
A series of biomarker analyses, including RNA sequencing (RNA-seq) and target DNA sequencing (DNA-seq), were performed using available patient tissue samples (Supplementary Table 4). All patients with samples included in these analyses had >12 months of follow-up. Initially, levels of published GESs in biopsies from patients with early PFS events (≤12 months; available RNA-seq (n = 6) or paired RNA/DNA-seq (n = 5) data) were compared with those in biopsies from patients with PFS >12 months (available RNA-seq (n = 21) or paired RNA/DNA-seq (n = 17) data). Patients with early PFS events had tumors with low expression levels of known genes and GESs associated with immune infiltration (for example, T cell–inflamed and interferon (IFN)-γ; Fig. 1a,b) compared with patients with PFS >12 months (top pathways from unbiased analy- sis (Supplementary Fig. 1)). Early PFS events were associated with low tumor mutational burden (TMB)/T cell–inflamed GES levels (Fig. 1c), high levels of specific immunosuppressive TME signa- tures (for example, cancer-associated fibroblasts) in a subgroup of patients (Fig. 1d) and elevated baseline levels of circulating tumor DNA (ctDNA) (Fig. 2a).
Comparison of biomarker profiles for samples from patients with CRs versus those from all other patients showed that CRs were associated with relatively low or no detectable baseline ctDNA levels (Fig. 2b). Well-known markers for immunotherapy response (for example, TMB and T cell–inflamed GES) were not associated with CR in our dataset (Extended Data Fig. 4), but an unbiased correla- tive analysis showed that patients achieving a CR had tumors with lower baseline levels of immunosuppressive TME signatures (Fig. 2c; M2 macrophage and myeloid signatures shown as examples of top pathways from unbiased analysis (Supplementary Fig. 2)).
T cell–inflamed GES was not associated with best percent change in target lesion size (Extended Data Fig. 5a), but a correlative analysis showed that phosphoinositide 3-kinase (PI3K) signaling was among the pathways most activated in tumors with little shrinkage, suggest- ing that a signaling mechanism compensatory to MAPK inhibition may contribute to lack of early response (Extended Data Fig. 5b).
To identify biomarker modulations following treatment we com- pared tissue immunohistochemistry (IHC), RNA-seq and plasma IFN-γ levels between baseline and on-treatment samples. Unpaired comparisons among different timepoints showed upregulation of checkpoint marker levels over time; 68% of patients had a PD-L1+ tumor based on baseline PD-L1 MEL score, while 91% of those with available PD-L1 results had a PD-L1+ tumor after 2–3 weeks of treat- ment (unpaired summary of all IHC results from part 2 of COMBI-i; Supplementary Table 5). Increased intratumoral CD8+ cells based on exploratory H-score analysis and PD-L1 MEL score were observed in the small sample set for which paired CD8 or PD-L1 results were available (Extended Data Fig. 6a and Supplementary Table 6). Elevated plasma IFN-γ levels were observed in 25 of 27 patients (93%) after 4 weeks of treatment (Extended Data Fig. 6b). An increase in T cell–specific GESs was observed by RNA-seq when paired baseline and 2- to 3-week on-treatment biopsies were compared (Fig. 2d; T cell–inflamed GES shown as example of a top pathway from unbiased analysis (Supplementary Fig. 3)) regard- less of subsequent patient progression, whereas MAPK pathway activity score (MPAS)25 and cell cycle GESs decreased from base- line to the time of biopsy at 2–3 weeks (Fig. 2e; MPAS shown as example of a top pathway from unbiased analysis (Supplementary Fig. 3)). Samples from patients with an early PFS event and avail- able longitudinal data were characterized by a subsequent decrease in T cell–inflamed GES and increase in MPAS based on analysis of 8- to 12-week biopsies (Fig. 2d,e).

PFS >12 months

PFS ≤12 months

PFS(days)
800
700
600
500
400
300
200
100
HAVCR2 PDCD1 LAG3 NKG7 CD8A CD8B IDO1 IFNG CD274
PDCD1LG2 CTLA4 FOXP3

Barplot
CR PR SD PD UNK

Gene expression
2
1
0
–1
–2

⦁ Wilcoxon P = 0.0488 c
Cell cycle (G1S) 3

WNT targets
MPAS 2
Signatures
MDSC 1
Pan-F-TBRS
Cancer-associated fibroblast 0
M2 macrophage 1
Neutrophil lineage
NK cell lineage 2
CD8 T cell
Cytotoxic T cell lineage T cell-inflamed
M1 macrophage B-cell lineage
d

T cell-inflamed GES level (average log2 CPM)
TMB (mutations Mb–1)
6 20
5
10
4

3 5
8.5
Cancer-associated fibroblast GES level (average log2 CPM)
8.0
7.5
7.0
6.5
6.0

5.5

5.0

Shortest PFS
Longest PFS PFS >12 months

PFS ≤12 months PFS >12 months
3 4 5 6
3 4 5 6

T cell-inflamed GES level (average log2 CPM)
T cell-inflamed GES level (average log2 CPM)

Fig. 1 | Baseline biomarker results from biopsy specimens based on PFS. a, Gene expression and GES levels based on timing of progression following treatment (PFS events ≤12 versus >12 months); asterisks indicate censoring. b, Distribution of baseline T cell–inflamed GES levels based on PFS groups; n = 27 independent tumor biopsy specimens (PFS ≤12 months, n = 6; PFS >12 months, n = 21). Box plots show median, first and third quartiles (boxes), and range up to 1.5× IQR from the bounds of the box (whiskers). Points beyond 1.5× IQR from the bounds of the box are plotted individually. Values for each group are reported as median (boxes) (whiskers). PFS ≤12 months, 3.80 (3.47–4.70) (2.88–5.11). PFS >12 months, 4.96 (4.41–5.43) (3.48–6.65). Descriptive P value is based on a two-sided Wilcoxon rank-sum test (W = 29, effect size = −1.08 (95% CI, −1.85 to −0.02)); no adjustments were made for multiple comparisons. c,d, Analysis of TMB (c) and immunosuppressive TME signature levels (for example, cancer-associated fibroblasts) (d) based
on T cell–inflamed GES levels. c,d, n = 22 independent tumor biopsy specimens. CPM, counts per million; IQR, interquartile range; MDSC, myeloid-derived suppressor cell; MPAS, MAPK pathway activity score; NK, natural killer; TBRS, transforming growth factor β response signature; UNK, unknown.

In this pooled analysis of COMBI-i parts 1 and 2, a promising CR rate of 44% was observed with spartalizumab plus dabrafenib and trametinib; achieving a CR has been associated with prolonged PFS and OS in patients treated with BRAF plus MEK inhibitors or anti-PD-1-based therapy1,2,26,27. While DOR in COMBI-i parts 1 and 2 was similar for all responders versus complete responders, increased durability of CR may emerge with additional follow-up. Although cross-trial comparisons should be interpreted with
caution, the 24-month OS rate (74%) with spartalizumab plus dab- rafenib and trametinib compares favorably with OS rates from phase 3 trials of either dabrafenib plus trametinib (52%) or anti-PD-1 ther- apy (nivolumab, 59%; pembrolizumab, 58%) alone1,2,26. Because sur- vival benefits with pembrolizumab plus dabrafenib and trametinib in KEYNOTE-022 improved with further follow-up19, it will be impor- tant to track the change in benefits, over time, of spartalizumab plus dabrafenib and trametinib observed at this early timepoint.

a Wilcoxon P = 0.00071 b

Tumor contribution to cfDNA
0.75

0.75

0.50

0.25

0.50

0.25

PFS
≤12 months
>12 months

0
≤12 months >12 months PFS

Tumor contribution to cfDNA
0
CR PR SD UNK
Best overall response

c

3.5

M2 macrophage GES level (average log2 CPM)
3.0
M2 macrophage signature Wilcoxon P = 0.0291

4.0

Myeloid GES level (average log2 CPM)
3.5
Myeloid signature Wilcoxon P = 0.02222

d
7

T cell-inflamed GES level (average log2 CPM)
6

5

4

Baseline 2 to 3 weeks of treatment

2.5

2.0

1.5

No CR

8 to 12 weeks of treatment

CR

Disease progression

e 8.0
7.5

MPAS signature level (average log2 CPM)
7.0

6.5

6.0

5.5

5.0

4.5
3.0

2.5

2.0

1.5

No CR

Baseline

CR

2 to 3 weeks of treatment

8 to 12 weeks of treatment

Disease progression

Adverse events led to temporary dose modifications in all patients, but many of these events were reversible because they were more frequently managed through dose reduction or interruption than by permanent discontinuation. However, the acceptability of the risk–benefit profile associated with checkpoint inhibitor plus targeted therapy combinations remains an important question. Further understanding of the timing and attribution of TRAEs with these regimens may be critical to the development of manage- ment strategies that optimize tolerability. These challenges will be assessed through a larger patient cohort and inclusion of the active comparator of targeted therapy alone in COMBI-i part 3.
In previous studies in patients treated with anti-PD-1 therapy, low baseline TMB and T cell–inflamed GES levels were described as intrinsic resistance markers28–31. However, an analysis in the adjuvant

setting suggested that patients with these characteristics do benefit from dabrafenib plus trametinib32. In our study, early PFS events were more frequent in patients with tumors that had low baseline TMB and T cell–inflamed GES levels. Our observation of on-treatment increases in T cell–inflamed GES levels regardless of subsequent progression is consistent with previous observations with anti-PD-1 therapy;33,34 we also observed on-treatment decreases in MPAS and cell cycle GESs. These data provide evidence of pathway inhibition, cell death and immune activation after treatment and suggest that spartalizumab plus dabrafenib and trametinib may have an early impact on tumor cells and the TME. Because our analyses also sug- gest that an immunosuppressive TME might preclude early CRs, it appears that some patients with a favorable intrinsic tumor bio- marker profile (for example, high levels of T cell–inflamed GES or

Fig. 2 | Baseline and on-treatment biomarker analyses based on PFS and RECIST response. a,b, Circulating tumor DNA detected before therapy based on PFS groups (a) and best overall response (b). Tumor contribution to total cfDNA as estimated by PureCN38,39 is plotted relative to PFS, with samples grouped by PFS (≤ or >12 months), and best overall response. a,b, n = 34 independent tumor biopsy specimens (PFS ≤12 months, n = 11; PFS >12 months, n = 23; CR, n = 16; PR, n = 11; SD, n = 6; UNK, n = 1). Box plots show median, first and third quartiles (boxes), and range up to 1.5× IQR from the bounds of the box (whiskers). Points beyond 1.5× IQR from the bounds of the box are plotted individually. Values for each group are reported as median (boxes) (whiskers). a, PFS ≤12 months, 0.39 (0.15–0.44) (0.00–0.47); PFS >12 months, 0.01 (0.01–0.02) (0.00–0.02). b, CR, 0.01 (0.00–0.02) (0.00–0.02); PR,
0.02 (0.005–0.335) (0.00–0.47); SD, 0.11 (0.005–0.3725) (0.00–0.45); UNK, 0.39. a, Descriptive P value is based on a two-sided Wilcoxon rank-sum test (W = 217, effect size = 0.26 (95% CI, 0.10–0.42)); no adjustments were made for multiple comparisons. c, Correlative analysis of GESs and pathways based on CR. GES levels, including M2 macrophage signature (left) and myeloid signature (right), were analyzed in biopsy specimens from patients with
and without a CR. n = 27 independent tumor biopsy specimens (CR, n = 14; no CR, n = 13). Box plots show median, first and third quartiles (boxes) and range up to 1.5× IQR from the bounds of the box (whiskers). Points beyond 1.5× IQR from the bounds of the box are plotted individually. Values for each group are reported as median (boxes) (whiskers). M2 macrophage: CR, 2.31 (2.19–2.54) (1.71–2.60); no CR, 2.76 (2.38–3.04) (2.04–3.50); myeloid: CR, 2.63 (2.09– 2.75) (1.35–3.20); no CR, 3.22 (2.51–3.51) (1.57–4.24). Descriptive P values are based on a two-sided Wilcoxon rank-sum test (M2 macrophage: W = 136, effect size = 0.433 (95% CI, 0.05–0.89); myeloid: W = 138, effect size = 0.612 (95% CI, 0.06–1.15)); no adjustments were made for multiple comparisons. d,e, Modulation of T cell–inflamed GES levels (d) and MPAS signature levels (e) during treatment. Unpaired and paired sample numbers are summarized in Supplementary Table 4. d,e, n = 11 (baseline), n = 9 (2–3 weeks), n = 7 (8–12 weeks) and n = 2 (progression) independent tumor biopsy specimens. Box plots show median, first and third quartiles (boxes) and range up to 1.5× IQR from the bounds of the box (whiskers). Points beyond 1.5× IQR from the bounds of the box are plotted individually. Values for each group are reported as median (boxes) (whiskers). T cell–inflamed: baseline, 4.79 (4.25–5.04) (3.38–5.27); 2–3 weeks, 5.88 (5.77–6.16) (5.46–6.16); 8–12 weeks, 5.13 (4.93–6.01) (3.84–7.25); progression, 4.26 (4.23–4.30) (4.20–4.33). MPAS: baseline, 7.23
(6.91–7.48) (6.59–7.81); 2–3 weeks, 5.06 (4.87–5.98) (4.57–5.98); 8–12 weeks, 4.91 (4.76–5.26) (4.39–5.45); progression, 7.70 (7.69–7.71) (7.67–7.72).

IFN-γ signal) may still not achieve a CR with checkpoint inhibitor plus targeted therapy combination if there are immunosuppressive components in the TME. New treatment approaches that target specific components of the TME (for example, M2 macrophages) may benefit this subset of patients. A high-CD8+/low-CD163+ immunophenotype has also previously been associated with higher CR rates in patients treated with BRAF ± MEK inhibitors;35 our data provide additional insights into factors that may predict CR. Striking differences in baseline ctDNA levels were also observed between subgroups defined by PFS duration or CR. Most patients with PFS >12 months or who achieved a CR had very low baseline ctDNA levels. This is in line with previous studies that demonstrated associations between baseline ctDNA levels and clinical outcome in patients with metastatic melanoma who received checkpoint inhibi- tors or targeted therapies36,37.
Although the findings from our robust set of biomarker analyses are intriguing, there are some limitations. Using targeted DNA-seq, we assessed baseline genetic alterations in key pathways of inter- est (for example, MAPK, PI3K and WNT) and also attempted to identify acquired resistance mutations, but the frequency of genetic alterations in this small cohort did not allow for solid conclusions to be drawn (data not shown). The sample size also precluded anal- ysis of biomarkers together with clinical variables in multivariate Cox models, nor was adjustment for multiple testing performed. Nevertheless, through unbiased analyses with Wilcoxon rank-sum testing, we identified consistent molecular features of key clinical subgroups of interest; these findings require validation in the large, randomized cohort of part 3.
The ongoing randomized, placebo-controlled, phase 3 portion of COMBI-i will further clarify the potential of spartalizumab plus dabrafenib and trametinib in patients with BRAF V600-mutant metastatic melanoma. The findings from COMBI-i parts 1 and 2 reported here complement the recent positive report from the phase 3 IMspire150 study20 in suggesting that combining check- point inhibitors with targeted therapy as a novel treatment approach for metastatic melanoma is worthy of further exploration.
Online content
Any methods, additional references, Nature Research report- ing summaries, source data, extended data, supplementary infor- mation, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41591-020-1082-2.
Received: 13 April 2020; Accepted: 26 August 2020;

Published: xx xx xxxx
References
⦁ Robert, C. et al. Pembrolizumab versus ipilimumab in advanced melanoma (KEYNOTE-006): post-hoc 5-year results from an open-label, multicentre, randomised, controlled, phase 3 study. Lancet Oncol. 20, 1239–1251 (2019).
⦁ Robert, C. et al. Five-year outcomes with dabrafenib plus trametinib in metastatic melanoma. N. Engl. J. Med. 381, 626–636 (2019).
⦁ Larkin, J. et al. Five-year survival with combined nivolumab and ipilimumab in advanced melanoma. N. Engl. J. Med. 381, 1535–1546 (2019).
⦁ McArthur, G. et al. 5-Year survival update of cobimetinib plus vemurafenib in BRAF V600 mutation-positive advanced melanoma: final analysis of the coBRIM study. Pigment Cell Melanoma Res. 33, 148–255 (2020).
⦁ Ugurel, S. et al. Survival of patients with advanced metastatic melanoma: the impact of MAP kinase pathway inhibition and immune checkpoint inhibition
– update 2019. Eur. J. Cancer 130, 126–138 (2020).
⦁ Ugurel, S. et al. Survival of patients with advanced metastatic melanoma: the impact of novel therapies – update 2017. Eur. J. Cancer 83,
247–257 (2017).
⦁ Hu-Lieskovan, S. et al. Improved antitumor activity of immunotherapy with BRAF and MEK inhibitors in BRAF(V600E) melanoma. Sci. Transl. Med. 7, 279ra241 (2015).
⦁ Kakavand, H. et al. PD-L1 expression and tumor-infiltrating lymphocytes define different subsets of MAPK inhibitor-treated melanoma patients. Clin. Cancer Res. 21, 3140–3148 (2015).
⦁ Wilmott, J. S. et al. Selective BRAF inhibitors induce marked T-cell infiltration into human metastatic melanoma. Clin. Cancer Res. 18, 1386–1394 (2012).
⦁ Wilmott, J. S. et al. Dynamics of chemokine, cytokine, and growth factor serum levels in BRAF-mutant melanoma patients during BRAF inhibitor treatment. J. Immunol. 192, 2505–2513 (2014).
⦁ Ascierto, P. A. & Dummer, R. Immunological effects of BRAF + MEK inhibition. Oncoimmunology 7, e1468955 (2018).
⦁ Mandalà, M. et al. Immunomodulating property of MAPK inhibitors: from translational knowledge to clinical implementation. Lab. Invest. 97, 166–175 (2017).
⦁ Yu, C. et al. Combination of immunotherapy with targeted therapy: theory and practice in metastatic melanoma. Front. Immunol. 10, 990 (2019).
⦁ Kelderman, S., Schumacher, T. N. & Haanen, J. B. Acquired and intrinsic resistance in cancer immunotherapy. Mol. Oncol. 8, 1132–1139 (2014).
⦁ Ascierto, P. A. et al. Update on tolerability and overall survival in COLUMBUS: landmark analysis of a randomised phase 3 trial of encorafenib plus binimetinib vs vemurafenib or encorafenib in patients with BRAF
V600-mutant melanoma. Eur. J. Cancer 126, 33–44 (2020).
⦁ Ascierto, P. A. et al. Dabrafenib, trametinib and pembrolizumab or placebo in BRAF-mutant melanoma. Nat. Med. 25, 941–946 (2019).
⦁ Ribas, A. et al. Combined BRAF and MEK inhibition with PD-1 blockade immunotherapy in BRAF-mutant melanoma. Nat. Med. 25, 936–940 (2019).
⦁ Sullivan, R. J. et al. Atezolizumab plus cobimetinib and vemurafenib in BRAF-mutated melanoma patients. Nat. Med. 25, 929–935 (2019).

⦁ Ferrucci, P. F. et al. Updated survival in patients with BRAF-mutant melanoma administered pemobrlizumab, dabrafenib, and trametinib. Pigment Cell Melanoma Res. 33, 148–255 (2020).
⦁ McArthur, G. A. et al. Evaluation of atezolizumab (A), cobimetinib (C), and vemurafenib (V) in previously untreated patients with BRAFV600
mutation-positive advanced melanoma: primary results from the phase 3 IMspire150 trial. In American Association for Cancer Research Virtual Annual Meeting I, abstract CT012 (2020).
⦁ Lin, C.-C. et al. Phase I/II study of spartalizumab (PDR001), an anti-PD1 mAb, in patients with advanced melanoma or non-small cell lung cancer. Ann. Oncol. 29, VIII413 (2018).
⦁ Larkin, J. et al. Overall survival in patients with advanced melanoma who received nivolumab versus investigator’s choice chemotherapy in CheckMate 037: a randomized, controlled, open-label phase III trial. J. Clin. Oncol. 36, 383–390 (2018).
⦁ Hamid, O. et al. Final analysis of a randomised trial comparing pembrolizumab versus investigator-choice chemotherapy for
ipilimumab-refractory advanced melanoma. Eur. J. Cancer 86, 37–45 (2017).
⦁ Ribas, A. et al. Pembrolizumab versus investigator-choice chemotherapy for ipilimumab-refractory melanoma (KEYNOTE-002): a randomised, controlled, phase 2 trial. Lancet Oncol. 16, 908–918 (2015).
⦁ Wagle, M. C. et al. A transcriptional MAPK Pathway Activity Score (MPAS) is a clinically relevant biomarker in multiple cancer types. NPJ Precis. Oncol. 2, 7 (2018).
⦁ Long, G. V. et al. Characteristics of long-term survivors and subgroup analyses with combination nivolumab plus ipilimumab for advanced melanoma (CheckMate 067). Pigment Cell Melanoma Res. 33, 148–255 (2020).
⦁ Osgood, C. et al. FDA analysis of depth of response (DpR) and survival across 10 randomized controlled trials in patients with previously untreated unresectable metastatic melanoma (UMM) by therapy type. J. Clin. Oncol. 37, 9508 (2019).
⦁ Cristescu, R. et al. Pan-tumor genomic biomarkers for PD-1 checkpoint blockade-based immunotherapy. Science 362, eaar3593 (2018).
⦁ Hodi, F. S. et al. Genomic analyses and immunotherapy in advanced melanoma. Cancer Res. 79, CT037 (2019).

⦁ Ribas, A. et al. Tumor mutational burden (TMB), T cell-inflamed gene expression profile (GEP) and PD-L1 are independently associated with response to pembrolizumab (pembro) in patients with advanced melanoma in the KEYNOTE (KN)-006 study. Cancer Res. 79, 4217 (2019).
⦁ Weber, J. S. et al. Adjuvant nivolumab (NIVO) versus ipilimumab (IPI) in resected stage III/IV melanoma: 3-year efficacy and biomarker results from the phase III CheckMate 238 trial. Ann. Oncol. 30, V533–V534 (2019).
⦁ Dummer, R. et al. Adjuvant dabrafenib plus trametinib versus placebo in patients with resected, BRAF(V600)-mutant, stage III melanoma
(COMBI-AD): exploratory biomarker analyses from a randomised, phase 3 trial. Lancet Oncol. 21, 358–372 (2020).
⦁ Gide, T. N. et al. Distinct immune cell populations define response to anti-PD-1 monotherapy and anti-PD-1/anti-CTLA-4 combined therapy. Cancer Cell 35, 238–255 (2019).
⦁ Vilain, R. E. et al. Dynamic changes in PD-L1 expression and immune infiltrates early during treatment predict response to PD-1 blockade in melanoma. Clin. Cancer Res. 23, 5024–5033 (2017).
⦁ Massi, D. et al. The density and spatial tissue distribution of CD8(+) and CD163(+) immune cells predict response and outcome in melanoma patients receiving MAPK inhibitors. J. Immunother. Cancer 7, 308 (2019).
⦁ Lee, J. H. et al. Circulating tumour DNA predicts response to anti-PD1 antibodies in metastatic melanoma. Ann. Oncol. 28, 1130–1136 (2017).
⦁ Santiago-Walker, A. et al. Correlation of BRAF mutation status in circulating-free DNA and tumor and association with clinical outcome across four BRAFi and MEKi clinical trials. Clin. Cancer Res. 22, 567–574 (2016).
⦁ Riester, M. et al. PureCN: copy number calling and SNV classification using targeted short read sequencing. Source Code Biol. Med. 11, 13 (2016).
⦁ Oh, S. et al. Reliable analysis of clinical tumor-only whole-exome sequencing data. JCO Clin. Cancer Inf. 4, 321–335 (2020).
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
© The Author(s), under exclusive licence to Springer Nature America, Inc. 2020

Methods
Patients. Parts 1 (dose-finding safety run-in) and 2 (biomarker cohort) of COMBI-i (NCT02967692) enrolled adult patients (≥18 years of age) with
histologically confirmed, BRAF V600-mutant unresectable or metastatic melanoma (stage IIIC/IV). No previous systemic therapy for unresectable or metastatic melanoma (for example, checkpoint inhibitors, targeted therapy, chemotherapy, biologic therapy, tumor vaccine therapy or investigational treatment) was allowed; previous intralesional, adjuvant or neoadjuvant therapy was allowed if completed
≥6 months before the start of study treatment, and previous radiation therapy was allowed if completed ≥4 weeks before the start of study treatment. For part 1, eligible patients had an ECOG PS of ≤1, alanine and aspartate aminotransferase levels <2.5× the upper limit of normal and no history of central nervous system metastases. For part 2, eligible patients had an ECOG PS of ≤2, no active central nervous system metastases and at least two cutaneous, subcutaneous or nodal lesions accessible for tumor sample collection. Across parts 1 and 2, patients were enrolled between February and October 2017.
Trial design and treatment. COMBI-i is a phase 3 trial consisting of three parts: a safety run-in (part 1), biomarker cohort (part 2) and randomized, placebo-controlled trial (part 3). The present report describes efficacy, safety and biomarker analyses from parts 1 and 2; part 3, in which spartalizumab plus dabrafenib and trametinib is evaluated against placebo plus dabrafenib and trametinib, is ongoing. Details of participating study sites and investigators for parts 1 and 2 are found in Supplementary Table 7.
Part 1 was designed to determine the recommended phase 3 regimen of spartalizumab plus dabrafenib and trametinib using a Bayesian logistic regression model based on the escalation with overdose control principle. The primary endpoint was the incidence of DLTs as defined per protocol occurring within the first 8 weeks of treatment, and monitored by a review team consisting of at least one clinician, safety representative and biostatistician from the study sponsor (Novartis Pharmaceuticals) and at least one investigator participating in the study who enrolled at least one patient in part 1. At least 6 and up to 18 patients could be enrolled and were treated with an initial specified dose of spartalizumab (400 mg) intravenously every 4 weeks, in combination with dabrafenib (150 mg) orally twice daily and trametinib (2 mg) orally once daily. Additional dose levels were explored using a Bayesian logistic regression model guided by escalation with overdose control criteria to determine the recommended phase 3 regimen. Newly enrolled patients were to be treated with the recommended dosing regimen if the risk of excessive toxicity was <25%. If the initial dose was not tolerated, fixed-dose combinations were to be explored in two parallel cohorts (Extended Data Fig. 1a).
In part 2 of COMBI-i, blood and tissue samples were collected at baseline,
after 2–4 and 8–12 weeks during treatment and at disease progression. Sample collection during treatment and at progression was optional for patients enrolled in part 1 but mandatory for patients enrolled in part 2 (Extended Data Fig. 1b). The primary endpoint of part 2 was change in PD-L1 expression levels and CD8+ cell levels in the tumor after treatment. Correlative analysis of biomarker data was performed using best overall response and PFS >/≤12 months as measurements of clinical outcome. Biomarker values were compared between outcome groups using Wilcoxon rank-sum tests.
During enrollment of parts 1 and 2, the study protocol was amended to modify guidelines for the management of pyrexia. Dabrafenib and trametinib should be interrupted promptly at the first symptom of pyrexia or its associated prodrome. Treatment with dabrafenib and trametinib should be restarted at the same dose if the patient is symptom free for ≥24 h (refs. 40,41).
Efficacy and safety assessments. Tumor assessments were conducted by the investigator using RECIST v.1.1, beginning at baseline. Postbaseline imaging assessments occurred on day 1 of cycle 4 (28-day cycle) and continued every
8 weeks. Beginning on day 1 of cycle 22, assessments continued every 12 weeks until disease progression, death or withdrawal from the study. All responses were confirmed by repeat scan ≤4 weeks after the initial RECIST response. Descriptive statistics in part 2 were summarized by visit.
The pooled safety set included all patients who were assigned to receive spartalizumab plus dabrafenib and trametinib and who received at least one dose of any study treatment. Adverse events were collected at every visit and graded according to the National Cancer Institute Common Terminology Criteria for Adverse Events v.4.03. Adverse event follow-up was conducted until 150 d after the last dose of spartalizumab. If the patient continued treatment with either dabrafenib or trametinib >150 d after the last dose of spartalizumab, monitoring
was conducted until 30 d after the last dose of dabrafenib or 120 d after the last dose of trametinib.

DNA/RNA extraction. Sections of thickness 5 µm (±1 µm) were cut from all blocks received. A pathologist visually inspected archival formalin-fixed,
paraffin-embedded (FFPE) slides and freshly cut slides from the tumor blocks to identify and notate the approximate percentage of tumor content in the region of interest and total tumor area (mm2). Depending on the tumor cell content,
4–12 slides were macrodissected and used for DNA/RNA isolation. If the region of

interest contained <10% tumor content, further processing was canceled. DNA/ RNA was coextracted from all samples available using the AllPrep RNA/DNA Extraction from FFPE Tissue Kit (Qiagen).

DNA- and cfDNA-seq. Both tumor DNA and ctDNA were sequenced following targeted capture on the same panel of 579 genes reported to be altered in cancer. For tumor DNA-seq, extracted DNA was sheared by ultrasonication (Covaris) and then underwent end repair, A-tailing, adapter ligation and PCR amplification (TruSeq Nano Library Preparation Kit, Illumina). For cfDNA sequencing, cfDNA was extracted from approximately 4 ml of plasma (QIAamp Circulating Nucleic Acid kit, Qiagen) and then constructed into sequencing libraries with end repair, A-tailing, unique molecular identifier (UMI) adapter ligation (custom adapters, IDT) and PCR amplification (TruSeq Nano Library Preparation Kit, Illumina).
The constructed DNA and cfDNA libraries were then hybridized to RNA baits (SureSelect, Agilent) targeting the 579 cancer-relevant genes. The captured libraries were sequenced to achieve a mean unique coverage of 563× for tumor DNA-seq and 2,226× for cfDNA-seq, using Illumina v.4 chemistry and paired-end 100-base pair (bp) reads (HiSeq, Illumina).

DNA-seq and cfDNA-seq data processing and analysis. DNA-seq data from tissue were aligned to the human reference genome (build hg38) using the Burrows–Wheeler Aligner v.0.7.10 (BWA-MEM, http://bio-bwa.sourceforge. net/)42. The resulting alignments were cleaned to remove PCR duplicates (Picard v.1.130, http://broadinstitute.github.io/picard/) and local realignment and base quality score recalibration was performed (Genome Analysis Toolkit (GATK) v.3.4.46, https://gatk.broadinstitute.org/hc/en-us)43,44. For cfDNA-seq data, UMIs were trimmed from the reads using UMI-Toolkit v.1 (https://github.com/angadps/ UMI-Toolkit) and the reads were then aligned to the human reference genome (build hg38) using BWA-MEM42. The alignments were then locally realigned and base quality scores recalibrated (GATK)43,44. Consensus reads were created using the UMI and alignment position to remove PCR-duplicate reads and sequencing artifacts (UMI-Toolkit). Single-nucleotide variants (SNVs) were identified with MuTect v.1.1.7 (https://software.broadinstitute.org/cancer/cga/mutect)45. Short insertion/deletion (indels) events were identified using Pindel v.1.0 (http://gmt. genome.wustl.edu/packages/pindel/)46. Structural variants were identified using PureCN v.1.8.1 (https://bioconductor.org/packages/release/bioc/html/PureCN. html)38. Chromosomal rearrangements were called using Socrates v.1 (https:// github.com/jibsch/Socrates)47.
Tumor DNA-seq libraries were included in downstream analyses if mean coverage was ≥100×, had GC/AT dropouts of <20% and there was evidence of tumor content in the sequenced sample (inferred tumor purity >0). Circulating free DNA libraries were included in the downstream analysis if the coverage was ≥500× and GC/AT dropouts were <20%. Potential sequencing artifacts and germline genetic variants were removed from downstream analyses. For DNA-seq, probable artifactual SNVs were identified by low coverage (<50×), low read support (<5 reads supporting the mutant allele), low allelic fraction (<0.01 unless known or probable oncogenic mutation), low average base quality (<25 unless known hotspot mutation) or a high proportion of reads with poorly supported alignments (>10% mapping quality equal to 0 (MQ0)). Probable
artifactual indels were identified by low coverage (<50×), low read support (<4), low allelic fraction (<0.04) or overlap with repetitive regions of the genome. For cfDNA-seq, a position-specific error rate was calculated based on the sequencing of plasma from 24 healthy controls, and mutations were retained only if they had support significantly greater than the position-specific error rate. Additional potential artifacts were removed using similar filters to the tumor DNA-seq: low allelic fraction (<0.005 unless known or probable oncogenic), poorly supported alignments (>50 MQ0 reads), low base quality (<20), low coverage (<100×) or in repetitive regions. Probable germline SNVs and indels were identified by their presence in the databases dbSNP 147 (https://www.ncbi.nlm.nih.gov/snp/), the Exome Sequencing Project (ESP6500SI-V2-SSA137.GRCh38-liftover, http://evs. gs.washington.edu/EVS/) and the Exome Aggregation Consortium (release 0.3;
now part of gnomAD, https://gnomad.broadinstitute.org/) at appreciable frequency (ESP minor allele frequency >0.001 or ExAC count >3 unless a known hotspot mutation). SNVs and indels were assigned a functional significance based on their presence in the Catalog of Somatic Mutations in Cancer (COSMIC v.83, https:// cancer.sanger.ac.uk/cosmic) and functional effect, with mutations reported in COSMIC in five or more tumors considered as ‘known’ oncogenic, mutations with COSMIC count <5 but predicted to lead to premature truncation of the protein considered as ‘likely’ oncogenic, and all others considered to have ‘unknown’ oncogenic status. Copy number variations were considered as amplifications if
the estimated copy number was ≥7, or as homozygous deletions if the estimated copy number was ≤0.5. PureCN uses a combination of the B allele frequency of single-nucleotide polymorphisms in copy number variants and the allele frequency of somatic point mutations to determine the proportion of cfDNA derived from the tumor38,39. The same approach was used to estimate tumor content (purity) in tumor DNA-seq. TMB was also calculated by PureCN, using the tumor content and allelic fraction information to remove germline variants and artifacts. TMB was then calculated as the number of somatic mutations per megabase of ‘callable’ coding sequence (that is, with sufficient coverage and quality).

RNA-seq. Ribosomal RNA from extracted total RNA was depleted using RNAseH. The rRNA-depleted sample was then fragmented, converted to complementary DNA and carried through the remaining steps of next-generation sequencing library construction—end repair, A-tailing, indexed adapter ligation and PCR amplification—using the TruSeq RNA v.2 Library Preparation kit (Illumina). The captured library was pooled with other libraries, each having a unique adapter index sequence, and applied to a sequencing flow cell. The flow cell underwent cluster amplification and massively parallel sequencing by synthesis using Illumina
v.4 chemistry and paired-end 100-bp reads (Illumina).
Sequence data were aligned to the reference human genome (build hg19) using STAR v.2.4.0e (https://github.com/alexdobin/STAR)48. Mapped reads were then used to quantify transcripts with HTSeq v.0.6.1p1 (https://htseq.readthedocs.io/ en/master/)49 and RefSeq (https://www.ncbi.nlm.nih.gov/refseq/) GRCh38 v.82 gene annotation. Gene expression data were normalized using the trimmed mean of M-value normalization as implemented in the edgeR R/Bioconductor package
v.3.20.9 (ref. 50). Hierarchical clustering was performed using Euclidean distance for samples and Pearson correlation for genes and gene sets; features were also ordered using the optimal leaf-ordering algorithm as implemented in the R package cba v.0.2.19. Pathway/gene set expression was derived using the geometric mean expression of all genes in each set. For pathway analyses, we used 1,329 gene
sets from MSigDB C2 Canonical Pathways v.6.2 (https://www.gsea-msigdb. org/gsea/msigdb)51–53 and added in-house and published gene sets as shown in Supplementary Table 8.

IHC analysis. Formalin-fixed, paraffin-embedded tissue blocks or unstained slides were used for IHC staining. Blocks were sectioned in a nuclease-free manner into 4-μm-thick FFPE slides, which were baked (30–120 min, depending on the antibody, at 60 ± 2 °C) at a vendor selected by the study sponsor. Tumor evaluation was performed by a certified pathologist. PD-L1 was assessed using the IHC 28-8 PharmDx assay on a Dako Autostainer Link 48. CD8 clone C8/144B (diluted
1:75, final concentration 2.1 μg ml–1; Dako), CD163 clone MRQ-26 (ready-to-use; Ventana), TIM-3 clone D5D5R (R1262) (diluted 1:125; Cell Signaling) and LAG-3 clone 17B4 (R1231) (diluted 1:600; Novus Biologicals) were analyzed using a Ventana Benchmark XT. FOXP3 clone 236A/E7 (R1096) (diluted 1:50, final concentration 10 μg ml–1; eBioscience/ThermoFisher Scientific) was analyzed using IHC staining on a Labvision autostainer. A negative control slide (rabbit and/or mouse IgG) was included for each patient sample. For PD-L1, the fraction of viable tumor cells expressing PD-L1 (discernible membrane staining of any intensity) was scored; cytoplasmic staining was not included in the scoring. In addition, MEL score was determined, which is based on a combination of the percentage of
PD-L1+ tumor cells and the percentage of PD-L1+ tumor-associated immune cells;54 these immune cells are present in the contiguous stroma that surrounds cancer
cell nests. Levels of intratumoral CD8+ cells were assessed using frequency bins of CD8+ density and a semiquantitative approach based on H-score. For all other IHC assays, the amount of IHC staining was used by measuring the area (marker area) of IHC-stained cells or structures in the center and periphery of a tumor.

Cytokine testing. IFN-y levels in plasma samples were quantified using Meso Scale Diagnostics (MSD) Proinflammatory Panel 1 (human), a multiplex sandwich electrochemiluminescent immunoassay. The assay was validated at a clinical research organization selected by the study sponsor and plasma samples were subsequently analyzed. Briefly, samples were diluted twofold with Diluent 2 as recommended by the assay kit manufacturer (MSD). Fifty microliters of calibrators and diluted samples was used for each replicate. Three levels of control were included in each run. Standards, controls and samples were all tested in duplicate. Assay signal, which is proportional to the amount of analyte present in the sample, was read on an MSD instrument. A four-parameter logistic curve fit was used to construct the standard curve, and IFN-y levels in test samples were determined from the standard curve. Results from controls were checked before acceptance of patient sample results.

Statistical analysis. Efficacy and safety analyses were conducted on the pooled data collected from parts 1 and 2 using SAS v.9.4. The Kaplan–Meier method was used to generate curves for DOR, PFS and OS. Biomarker analyses were performed in R v.3.4.3 and Bioconductor v.3.6. Correlations with tumor shrinkage were assessed by calculating the Spearman correlation coefficient. Comparisons between groups were made using two-sided Wilcoxon rank-sum tests; no adjustments were made for multiple comparisons. Nominal P values, test statistics (W), differences
in the locations of the two groups under comparison (effect size) and 95% CIs are reported. Tumor contribution to baseline cfDNA was compared between patients with PFS ≤12 months and those with PFS >12 months. DNA-seq was also used to estimate baseline TMB, as compared between patients with DNA-seq data who did and did not achieve a CR. RNA-seq was used to compare baseline expression of genes and pathways of interest in patients with PFS ≤12 months versus those with PFS >12 months, and in patients who did versus those who did not achieve a CR.
For all unbiased RNA-seq analyses, only those pathways with nominal P < 0.05 were selected and were then ranked by the absolute values of log2(fold change) between the two groups under comparison. Comparisons were made between complete and noncomplete responders (the latter category including patients who

had a best overall response of PR, SD or progressive disease) and between those who had a PFS event within the first 12 months of treatment versus all others (all patients in the biomarker cohort had a follow-up of ≥12 months without any censoring). To compare changes from baseline to 2–3 weeks on treatment, we
used the R/Bioconductor package limma v.3.34.5, including only patients who had samples taken at both timepoints and explicitly modeling patients as a covariate. The full gene sets for which P < 0.05 in these unbiased analyses are available in Excel format as Supplementary Tables 9–11. All biomarker experiments were performed once and analyzed using R v.3.4.3.
Study oversight. This study was funded by the sponsor and conducted in accordance with the provisions of the Declaration of Helsinki and Good Clinical Practice guidelines. The protocol was approved by the institutional review
board or human research ethics committee at each participating study center (Supplementary Table 7). The study was designed by the authors and sponsor. Data were collected by the study site staff and monitored by the sponsor and independent data monitoring committees. The sponsor was involved in data analysis and interpretation as well as writing the report; all authors had full access to all data in the study and had final responsibility for the decision to submit for publication. All participants provided signed written informed consent.

Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Data availability
Novartis is committed to sharing, with qualified external researchers, access to patient-level data and supporting clinical documents from eligible studies.
Requests are reviewed and approved by an independent review panel on the basis of scientific merit. All data provided are anonymized to respect the privacy of patients who have participated in the trial, in line with applicable laws and
regulations. This trial data availability is in accordance with the criteria and process described on ClinicalStudyDataRequest.com. Publicly available databases utilized for the biomarker analyses in this study include RefSeq (https://www.ncbi.nlm.nih. gov/refseq/), dbSNP (https://www.ncbi.nlm.nih.gov/snp/), MSigDB C2 Canonical Pathways (https://www.gsea-msigdb.org/gsea/msigdb), the Exome Sequencing Project (http://evs.gs.washington.edu/EVS/), the Exome Aggregation Consortium (now part of gnomAD, https://gnomad.broadinstitute.org/) and COSMIC (https:// cancer.sanger.ac.uk/cosmic).

References
⦁ Atkinson, V. et al. Optimizing combination dabrafenib and trametinib therapy in BRAF mutation-positive advanced melanoma patients: guidelines from Australian melanoma medical oncologists. Asia Pac. J. Clin. Oncol. 12, 5–12 (2016).
⦁ Menzies, A. M. et al. Characteristics of pyrexia in BRAFV600E/K metastatic melanoma patients treated with combined dabrafenib and trametinib in a phase I/II clinical trial. Ann. Oncol. 26, 415–421 (2015).
⦁ Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows– Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
⦁ McKenna, A. et al. The genome analysis Toolkit: a MapReduce framework for analyzing next-generation DNA sequencing data. Genome Res. 20, 1297–1303 (2010).
⦁ DePristo, M. A. et al. A framework for variation discovery and genotyping using next-generation DNA sequencing data. Nat. Genet. 43, 491–498 (2011).
⦁ Cibulskis, K. et al. Sensitive detection of somatic point mutations in impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013).
⦁ Ye, K., Schulz, M. H., Long, Q., Apweiler, R. & Ning, Z. Pindel: a pattern growth approach to detect break points of large deletions and medium sized insertions from paired-end short reads. Bioinformatics 25, 2865–2871 (2009).
⦁ Schroder, J. et al. Socrates: identification of genomic rearrangements in tumour genomes by re-aligning soft clipped reads. Bioinformatics 30, 1064–1072 (2014).
⦁ Dobin, A. et al. STAR: ultrafast universal RNA-seq aligner. Bioinformatics 29, 15–21 (2013).
⦁ Anders, S., Pyl, P. T. & Huber, W. HTSeq—a Python framework to work with high-throughput sequencing data. Bioinformatics 31, 166–169 (2015).
⦁ Robinson, M. D., McCarthy, D. J. & Smyth, G. K. edgeR: a Bioconductor package for differential expression analysis of digital gene expression data. Bioinformatics 26, 139–140 (2010).
⦁ Subramanian, A. et al. Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles. Proc. Natl Acad. Sci. USA 102, 15545–15550 (2005).
⦁ Liberzon, A. et al. The Molecular Signatures Database (MSigDB) hallmark gene set collection. Cell Syst. 1, 417–425 (2015).
⦁ Liberzon, A. et al. Molecular signatures database (MSigDB) 3.0.
Bioinformatics 27, 1739–1740 (2011).
⦁ Daud, A. I. et al. Programmed death-ligand 1 expression and response to the anti-programmed death 1 antibody pembrolizumab in melanoma. J. Clin. Oncol. 34, 4102–4109 (2016).

Acknowledgements
We thank the patients and their families for their participation. We also thank the study site staff, additional investigators, R. Leary, C. Unitt and S. Mahan (Next Generation Diagnostics) for their contributions. We thank K. Gibbs for biomarker sample management, as well as A. Savchenko, J. Choi, C. Wong, B. Fu, G. Gorgun and R. Ramesh for support with biomarker analyses. We thank Navigate Biopharma for DNA and
RNA extraction, as well as HistoGeneX and Bioagilytix for biomarker testing. We also thank M. Voi (Novartis Pharmaceuticals) for guidance and critical review. We thank A. Lytle and A. Ghiretti (ArticulateScience LLC) for providing medical writing support, which was funded by Novartis Pharmaceuticals Corporation in accordance with Good Publication Practice guidelines (http://www.ismpp.org/gpp3). COMBI-i (NCT02967692) is sponsored by Novartis Pharmaceuticals.

Author contributions
C.R., D.S., H.A.T., J.C.B., E.G. and G.V.L. conceived or designed the work. R.D., C.L., V.A., M.M., P.D.N., A.A., E.R., N.Y., C.R., D.S., H.A.T., P.A.A., A.R., N.P., C.D.C., K.T.F.,
D.G., A.M., J.C.B., E.G. and G.V.L. acquired, analyzed or interpreted the data. R.D., C.L., A.A., E.R., N.Y., C.R., D.S., H.A.T., A.R., K.T.F., J.C.B., E.G. and G.V.L. drafted or
substantively revised the work.

Competing interests
R.D. reports intermittent, project-focused consulting and/or advisory relationships with Novartis, Merck Sharp & Dohme (MSD), Bristol Myers Squibb (BMS), Roche, Amgen, Takeda, Pierre Fabre, Sun Pharma, Sanofi, Catalym, Second Genome, Regeneron and Alligator outside the submitted work. C.L. reports research funding from Roche and BMS, speakers bureau for Roche, BMS, MSD, Amgen, Novartis and Pierre Fabre, and consulting and advisory roles for BMS, MSD, Roche, Novartis, Merck Serono, Sanofi and Pierre Fabre. V.A. reports advisory roles for BMS, Merck Serono, MSD, Novartis, Roche, Nektar and Pierre Fabre, and received speaker’s fees from BMS, Merck, MSD and Novartis, and travel support from BMS, MSD and Onco-Sec. M.M. reports consulting or advisory roles for MSD, Roche, BMS and Pierre Fabre and research funding from Roche, Novartis and BMS. P.D.N. reports advisory roles for BMS, Immunocore, Merck, MSD, Novartis and Pfizer, speaker’s bureaux for BMS and Novartis and steering committee membership for Immunocore, Merck and Novartis. A.A. reports personal fees and other from BMS, MSD, Roche, Novartis, Merck, Sanofi, Amgen and Pierre Fabre, outside the submitted work. E.R. reports consulting or advisory roles for Amgen, Bayer, BMS, MSD, Merck, Novartis, Pierre Fabre, Roche and Sanofi; reception of honoraria from Amgen, Bayer, BMS, MSD, Merck, Novartis, Pierre Fabre, Roche and Sanofi; speaker’s bureaux for Amgen, BMS, MSD, Merck, Novartis, Pierre Fabre and Sanofi; research funding from
Amgen, BMS, MSD, Novartis, Pierre Fabre and Roche; travel support from Amgen, BMS, MSD, Merck, Novartis, Pierre Fabre, Roche and Sanofi; and President of the Austrian Cancer Aid/Styria. N.Y. reports consulting and advisory roles for Novartis, Ono, BMS and MSD, honoraria from Novartis, Ono, BMS and MSD and institutional research support from Novartis, Ono, BMS, MSD and Takara-Bio. C.R. reports consulting or advisory roles

for BMS, Roche, Amgen, Novartis, Pierre Fabre, MSD, Sanofi, Biothera, CureVac and Merck. D.S. reports research funding from Novartis and BMS. H.A.T. reports consulting or advisory roles for Novartis, BMS, Roche-Genentech, Merck and Array BioPharma and research funding from BMS, Novartis, Merck, GlaxoSmithKline, Genentech/Roche and Celgene. P.A.A. reports consulting or advisory roles for BMS, Roche-Genentech, MSD, Array, Novartis, Merck Serono, Pierre Fabre, Incyte, NewLink Genetics, Genmab, Medimmune, AstraZeneca, Syndax, Sun Pharma, Sanofi, Idera, Ultimovacs, Sandoz, Immunocore, 4SC, Alkermes, Italfarmaco, Nektar and Boehringer Ingelheim, research funding from BMS, Roche-Genentech and Array and travel support from MSD. A.R. reports serving as a consultant/independent contractor for, and being the recipient of honoraria from, Amgen, Chugai, Merck, Novartis and Sanofi; advisory roles and receipt of honoraria from Arcus, Bioncotech, Compugen, CytomX, ImaginAb, Isoplexis, Merus, Rgenix, Lutris, PACT Pharma and Tango Therapeutics; self-managed stock shareholder in Arcus, Compugen, CytomX and Merus; and research support from Agilent and BMS.
K.T.F. reports advisory roles for Clovis Oncology, Strata Oncology, Vivid Biosciences, Checkmate Pharmaceuticals, X4 Pharmaceuticals, PIC Therapeutics, Sanofi, Amgen, Asana, Adaptimmune, Fount, Aeglea, Stattuck Labs, Tolero, Apricity, Oncoceutics, Fog Pharma, Neon, Tvardi, xCures, Monopteros and Vibliome; consulting roles for Lilly, Novartis, Genentech, BMS, Merck, Takeda, Verastem, Boston Biomedical, Pierre Fabre and Debiopharm; research funding from Novartis and Sanofi; and stock shareholder in Clovis Oncology, Strata Oncology, Vivid Biosciences, Checkmate Pharmaceuticals, X4 Pharmaceuticals, PIC Therapeutics, Fount, Shattuck Labs, Apricity, Oncoceutics, Fog Pharma, Tvardi, xCures, Monopteros and Vibliome. N.P. reports employment at Novartis Healthcare Pvt. Ltd. C.D.C. reports employment with, and stock in, Novartis Pharmaceuticals. D.G. reports employment with Novartis Institutes for BioMedical
Research. A.M. reports employment by Novartis and stock or ownership in Novartis and BMS. J.C.B. reports employment by, and stock ownership in, Novartis, and is a coinventor on a patent application related to reported biomarker subgroups of interest. E.G. reports employment by, and stock ownership in, Novartis. G.V.L. reports consultant advisory roles for Aduro Biotech, Inc., Pierre Fabre Medicament, BMS, Amgen, MSD, Novartis Pharma, Array BioPharma, Syneos and Sandoz Biopharmaceuticals. All authors received support for third-party medical writing and editorial assistance provided by ArticulateScience LLC and funded by Novartis Pharmaceuticals Corporation.

Additional information
Extended data is available for this paper at https://doi.org/10.1038/s41591-020-1082-2.
Supplementary information is available for this paper at https://doi.org/10.1038/ s41591-020-1082-2.
Correspondence and requests for materials should be addressed to R.D.
Peer review information Javier Carmona was the primary editor on this article, and managed its editorial process and peer review in collaboration with the rest of the editorial team.
Reprints and permissions information is available at www.nature.com/reprints.

Extended Data Fig. 1 | Study designs for (a) part 1 and (b) part 2 of COMBI-i. ALT, alanine aminotransferase; AST, aspartate aminotransferase; BID, twice daily; CD, cluster of differentiation; CNS, central nervous system; DCR, disease control rate; DLT, dose-limiting toxicity; DOR, duration of response; ECOG PS, Eastern Cooperative Oncology Group performance status; FFPE, formalin-fixed paraffin-embedded; ORR, overall response rate; OS, overall survival; PD, progressive disease; PD-L1, programmed death ligand 1; PFS, progression-free survival; PK, pharmacokinetics; Q4W, every 4 weeks; Q8W, every 8 weeks; QD, once daily; RECIST, Response Evaluation Criteria in Solid Tumors; RP3R, recommended phase 3 regimen; S, screening; ULN, upper limit of normal.a BRAF V600 mutation was assessed based on local testing (followed by central confirmation using the bioMérieux THxID-BRAF assay). b With
systemic therapy including checkpoint inhibitors, targeted therapy, chemotherapy, biologic therapy, tumor vaccine therapy, or investigational treatment for unresectable or metastatic melanoma; prior intralesional, adjuvant, or neoadjuvant therapy was allowed if completed ≥ 6 months prior to start of study treatment, and prior radiation therapy was allowed if completed ≥ 4 weeks prior to start of study treatment. c DL-1b: DLT observation period starts on cycle 2, day 1 (C2D1 [day 29]). Patients who did not tolerate dabrafenib and/or trametinib and discontinued during the first 4 weeks were to be replaced due to insufficient exposure.

Extended Data Fig. 2 | Best percent change from baseline in sum of diameters by local investigator review (N = 36). A total of 33 patients experienced a reduction in the size of the target lesion. One patient with SD had a best percent change of 0% in the target lesion. Best percent change in the target lesion was not available for 1 patient with PD. Best percent change in target lesion could not be calculated for 1 additional patient as best overall response was unknown. CR, complete response; PD, progressive disease; PR, partial response; SD, stable disease.

Extended Data Fig. 3 | Time-to-event analyses, including (a) DOR, (b) PFS, and (c) OS, in patients enrolled in parts 1 and 2 of COMBI-i (N = 36). 12- and 24-month DOR rates were 80% (95% CI, 59%-91%) and 53% (95% CI, 29%-73%); 12- and 24-month PFS rates were 67% (95% CI, 49%-80%) and
41% (95% CI, 23%-59%); 12- and 24-month OS rates were 86% (95% CI, 70%-94%) and 74% (95% CI, 56%-86%). DOR, duration of response; PFS, progression-free survival; NE, not estimable; OS, overall survival.

Extended Data Fig. 4 | Markers of response to immunotherapy were not associated with CR. Baseline T-cell–inflamed GES levels and TMB in samples from patients with and without a CR. For T-cell–inflamed GES: n = 27 independent tumor biopsy specimens (CR, n = 14; no CR, n = 13). For TMB:
n = 24 independent tumor biopsy specimens (CR, n = 12; no CR, n = 12). Box plots show median, first and third quartiles (boxes), and range up to 1.5 times IQR from the bounds of the box [whiskers]. Points beyond 1.5 times IQR from the bounds of the box are plotted individually. For T-cell–inflamed GES: CR, 4.90 (4.31-5.27) [3.48-6.09]; no CR, 4.87 (3.99-5.30) [2.88-6.65]. For TMB: CR, 7.196 (5.957-9.205) [3.365-13.479]; no CR, 7.209 (6.639-9.140)
[3.539-10.533]. Descriptive P values are based on a two-sided Wilcoxon rank sum test (T-cell–inflamed GES: W = 89, effect size -0.04 [95% CI, -0.75- 0.83]; TMB: W = 79, effect size 0.05 [95% CI, -2.40-2.85]); no adjustments were made for multiple comparisons. CPM, counts per million; CR, complete response; GES, gene expression signature; IQR, interquartile range; TMB, tumor mutational burden.

Extended Data Fig. 5 | Correlative analysis of GES levels and tumor shrinkage following treatment with spartalizumab in combination with dabrafenib and trametinib. Association between a, T-cell–inflamed GES levels and b, PI3K pathway gene expression and best overall tumor reduction, based on Spearman correlation coefficient. n = 27 independent tumor biopsy specimens. CPM, counts per million; GES, gene expression signature.

Extended Data Fig. 6 | Evidence of immune activation during treatment with spartalizumab in combination with dabrafenib and trametinib. a, Analysis of intratumoral density of CD8+ cells for available paired tumor biopsy specimens (n = 9) at baseline and on treatment using exploratory H-score analysis. Scale bars = 100 μm. b, Modulation of plasma IFN-γ following treatment with spartalizumab plus dabrafenib plus trametinib. Of the 27 independent plasma specimens analyzed, 25 showed elevated IFN-γ levels on treatment, while the other 2 showed a slight decrease. Box plot shows median, first and third quartiles (boxes), and range up to 1.5 times IQR from the bounds of the box [whiskers]. Points beyond 1.5 times IQR from the bounds of the box are plotted individually. Baseline, 1.78 (1.06-2.37) [1.06-4.19]. On treatment, 6.08 (3.65-7.88) [1.69-10.51]. IFN, interferon; IHC, immunohistochemistry;
IQR, interquartile range.

nature research | reporting summary
variants were removed from downstream analyses. For DNA sequencing, likely artifactual SNVs were identified by low coverage (< 50X), low read support (< 5 reads supporting the mutant allele), low allelic fraction (< 0.01 unless known or likely oncogenic mutation), low average base quality (< 25 unless known hotspot mutation), or high proportion of reads with poorly supported alignments (> 10% MQ0). Likely artifactual indels were identified by low coverage (< 50X), low read support (< 4), low allelic fraction (< 0.04), or overlap with repetitive regions of the genome. For cfDNA sequencing, a position-specific error rate was calculated based on the sequencing of plasma from 24 healthy controls, and mutations were retained only if they had significantly greater support than the position-specific error rate. Additional potential artifacts were removed using similar filters to the tumor DNA sequencing: low allelic fraction (< 0.005 unless known or likely oncogenic), poorly supported alignments (> 50 MQ0 reads), low base quality (< 20), low coverage (< 100X), or in repetitive regions. Likely germline SNVs and indels were identified by presence in dbSNP 147 (https://www.ncbi.nlm.nih.gov/snp/), the Exome Sequencing Project (ESP6500SI-V2- SSA137.GRCh38-liftover; http://evs.gs.washington.edu/EVS/) and Exome Aggregation Consortium (release 0.3; http:// exac.broadinstitute.org/) databases at appreciable frequency (ESP MAF > 0.001 or ExAC count > 3 unless known hotspot mutation). SNVs and indels were assigned a functional significance based on presence in the Catalog of Somatic Mutations in Cancer (COSMIC [v83]) and functional effect, with mutations reported in COSMIC in ≥ 5 tumors considered as “known” oncogenic, mutations with COSMIC count < 5 but predicted to lead to premature truncation of the protein considered as “likely” oncogenic, and all others considered to have “unknown” oncogenic status. Copy number variations were considered as amplifications if the estimated copy number was ≥ 7 or as homozygous deletions if the estimated copy number was ≤ 0.5. PureCN uses a combination of the B-allele frequency of single-nucleotide polymorphisms in copy number variants and allele frequency of somatic point mutations to determine the proportion of cfDNA derived from the tumor. The same approach was used to estimate tumor content (purity) in the tumor DNA-seq. TMB was also calculated by PureCN using the tumor content and allelic fraction information to remove germline variants and artifacts. TMB was then calculated as the number of somatic mutations per megabase of “callable” coding sequence (ie, with sufficient coverage and quality).

Ribosomal RNA (rRNA) from extracted total RNA was depleted using RNAseH. The rRNA-depleted sample was then fragmented, converted to cDNA, and carried through the remaining steps of next generation sequencing library construction: end repair, A-tailing, indexed adaptor ligation, and PCR amplification using the TruSeq RNA v2 Library Preparation kit (Illumina). The captured library was pooled with other libraries, each having a unique adaptor index sequence, and applied to a sequencing flow cell. The flow cell underwent cluster amplification and massively parallel sequencing by synthesis using Illumina v4 chemistry and paired-end 100 bp reads (Illumina).
Sequence data were aligned to the reference human genome (build hg19) using STAR v2.4.0e (https://github.com/alexdobin/STAR). Mapped reads were then used to quantify transcripts with HTSeq v0.6.1p1 (https://htseq.readthedocs.io/en/master/) and the Refseq (https:// www.ncbi.nlm.nih.gov/refseq/) GRCh38 v82 gene annotation. Gene expression data were normalized using trimmed mean of M-value normalization as implemented in the edgeR R/Bioconductor package v3.20.9. Hierarchical clustering was performed using Euclidean distance for samples and Pearson correlation for genes and gene sets; features were also ordered using the optimal leaf ordering algorithm as implemented in the R package cba v0.2.19. Pathway/gene set expression was derived using the geometric mean expression of all genes in each set. For pathway analyses, we used 1329 gene sets from MSigDB C2 Canonical Pathways v6.2 (https://www.gsea-msigdb.org/gsea/ msigdb) and added in-house and published gene sets as shown in Supplementary Table 8. To compare changes from baseline to 2 to 3 weeks on treatment, we used the R/Bioconductor package limma v3.34.5, including only patients who had samples at both time points and explicitly modeling patients as a covariate.
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data
Policy information about availability of data
All manuscripts must include a data availability statement. This statement should provide the following information, where applicable:
⦁ Accession codes, unique identifiers, or web links for publicly available datasets
⦁ A list of figures that have associated raw data
⦁ A description of any restrictions on data availability

Field-specific reporting
Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.
Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences
For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

April 2020
Life sciences study design
All studies must disclose on these points even when the disclosure is negative. Sample size

2

nature research | reporting summary
Data exclusions

Replication

Randomization

Blinding

Reporting for specific materials, systems and methods
We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.
Materials & experimental systems Methods

Antibodies
Antibodies used

April 2020
Validation

3

nature research | reporting summary
Human research participants
Policy information about studies involving human research participants Population characteristics

Recruitment

Ethics oversight

Note that full information on the approval of the study protocol must also be provided in the manuscript.

April 2020
Clinical data
Policy information about clinical studies
All manuscripts should comply with the ICMJE guidelines for publication of clinical research and a completed CONSORT checklist must be included with all submissions.
Clinical trial registration Study protocol

4

nature research | reporting summary
Data collection

Outcomes