A randomized, non-comparative phase 2 study of neoadjuvant immune checkpoint blockade in retroperitoneal dedifferentiated liposarcoma and extremity/truncal undifferentiated pleomorphic sarcoma

Abstract

Based on the demonstrated clinical activity of immune checkpoint blockade (ICB) in advanced dedifferentiated liposarcoma (DDLPS) and undifferentiated pleomorphic sarcoma (UPS), we conducted a randomized, non-comparative phase 2 trial (NCT03307616) of neoadjuvant nivolumab or nivolumab/ipilimumab in patients with resectable retroperitoneal DDLPS (n=17) and extremity/truncal UPS (+ concurrent nivolumab/radiation therapy (RT); n=10). The primary endpoint of pathologic response (percent hyalinization) was a median of 8.8% in DDLPS and 89% in UPS. Secondary endpoints were the changes in immune infiltrate, the radiographic response, the 12- and 24-month relapse-free survival and overall survival. Lower densities of T-regulatory cells before treatment were associated with major pathologic response (hyalinization>30%). Tumor infiltration by B-cells was increased following neoadjuvant treatment and associated with overall survival in DDLPS. B-cell infiltration was associated with higher densities of T-regulatory cells before treatment which was lost upon ICB treatment. Our data demonstrate that neoadjuvant ICB is associated with complex immune changes within the tumor microenvironment in DDLPS and UPS and that neoadjuvant ICB with concurrent RT has significant efficacy in UPS.

Introduction

Of the ~13,000 patients diagnosed with soft tissue sarcoma (STS) every year in the United States, more than one-third are expected to die of their disease after current standard management^1,2. Although radiation therapy (RT) and/or chemotherapy reduce recurrence risk, systemic therapy options are limited, highlighting the need for novel treatments². As STS are rare and heterogenous with >100 histologic types and subtypes³, large, randomized clinical trials are difficult, even in collaboration with other large-volume sarcoma centers. Therefore, novel clinical trial designs are necessary to evaluate potential therapies in a timely manner.

Within the past decade, major advances have been made in cancer therapy through the use of immune checkpoint blockade (ICB). Recent evidence suggests that ICB targeting programmed cell death protein 1 (PD-1) and cytotoxic T-lymphocyte antigen-4 (CTLA-4) has activity in metastatic dedifferentiated liposarcoma (DDLPS) and undifferentiated pleomorphic sarcoma (UPS) with variable responses (DDLPS 10–20%, UPS 20–40%)^4–7. Our group and others have previously demonstrated improved clinical benefit from ICB in STS with high pre-treatment immune infiltration^8,9, particularly by B-cells in the context of tertiary lymphoid structures (TLS)^10,11.

Based on these results, we tested the clinical activity of and evaluated immune responses to neoadjuvant ICB in patients with resectable DDLPS and UPS. In this randomized, noncomparative phase 2 trial, patients with treatment-naïve primary or locally recurrent resectable retroperitoneal DDLPS (arms A/B, Fig. 1a) and extremity/truncal UPS (arms C/D, Fig. 1b) were randomized 1:1 to neoadjuvant nivolumab or nivolumab/ipilimumab. UPS patients (arms C/D) were treated with concurrent 50Gy RT (day [D]15–D50). Following completion of neoadjuvant therapy, all patients underwent surgical resection of the tumor.

Fig. 1: Trial schema of immune checkpoint inhibition in retroperitoneal dedifferentiated liposarcoma and extremity/truncal undifferentiated pleomorphic sarcoma.

a, Patients with resectable, pathologically confirmed, retroperitoneal DDLPS were randomized 1:1 ratio to neoadjuvant nivolumab 3mg/kg IV every 14 days for up to three doses (arm A: DDLPS, D1, D15, D29) or ipilimumab 1mg/kg IV one dose plus nivolumab 3mg/kg IV every 14 days for up to three doses (arm B: DDLPS, ipilimumab on D1 only, nivolumab D1, D15 and D29), followed by surgical resection 2 to 4 weeks after last dose of nivolumab. b, Patients with resectable, pathologically confirmed, extremity/truncal UPS were randomized 1:1 ratio to neoadjuvant nivolumab 3mg/kg IV every 14 days for up to four doses and concurrent RT starting two weeks after first dose with 50Gy in 25 fractions (arm C: UPS, nivolumab D1, D15, D29, D43 + RT) or ipilimumab 1mg/kg IV one dose plus nivolumab 3mg/kg IV every 14 days for up to four doses and concurrent RT starting two weeks after first dose with 50Gy in 25 fractions (arm D: UPS, ipilimumab on D1 only, nivolumab D1, D15, D29, D43 + RT), followed by surgical resection 4 to 6 weeks after completion of RT. For both treatment arms, the primary endpoint was pathologic response as assessed by percent hyalinization at surgery. Select secondary endpoints were percent viable tumor at surgery, change in immune infiltration, objective response rate by RECIST 1.1, RFS, OS, and toxicity as assessed by Common Terminology Criteria for Adverse Events (CTCAE) v4.0. Select exploratory endpoints were change in immunologic genomic markers, presence of intratumoral B-cells and TLS, and microbiome composition and diversity. To explore those secondary and exploratory endpoints, longitudinal tumor, and blood samples were acquired. Tumor specimens were collected with biopsies before start of treatment and before second dose of nivolumab, and on surgical specimens. Blood samples were collected pretherapy, before each injection of immunotherapy, before surgery, and during follow-up. Fecal samples were collected prior to initiation of therapy, before second dose of immunotherapy and at surgery.

The primary endpoint of the trial was pathologic response (percent hyalinization as a continuous variable) at the time of surgical resection within each treatment arm. Secondary and exploratory endpoints are listed in Fig. 1. Correlative endpoints are presented in patients with available samples for biomarker analysis in each arm. The results of the reported comparisons are exploratory in nature and hypothesis generating.

Results

Participants, treatment and toxicity.

From January 2017 to February 2020, 28 patients were screened for eligibility and consented on trial (Fig. 2). One patient was deemed a screen failure. Twenty-seven eligible patients were treated on study: 17 DDLPS (arm A: n=8, arm B: n=9; Fig. 1a) and 10 UPS (arm C: n=6, arm D: n=4; Fig. 1b). Initial planned accrual was 10 patients in each arm, but the trial was stopped early due to slow accrual and in the context of the COVID-19 pandemic. Patient characteristics and treatment disposition are shown in Table 1. Eleven DDLPS patients had recurrent tumors and eight had multifocal disease at baseline; one UPS patient had recurrent disease. DDLPS patients were high-risk based on Sarculator-calculated DFS (DDLPS 2%, Interquartile Range [IQR] 1–14; UPS 22%, IQR 20–24) and OS (DDLPS 23%, IQR 15–38; UPS 81%, IQR 79–82) (Table 2).

Fig. 2: Consolidated Standards of Reporting Trials (CONSORT) flow diagram.

Flow diagram depicts the disposition of patients throughout the phases of the study, including screening, randomization to neoadjuvant treatment and surgery. Reasons for screen failures, delayed surgery, or surgery performed off trial are shown.

Table 1.

Patient characteristics at baseline and clinical endpoints by treatment group

Variable	Category	Retroperitoneal DDLPS			Extremity/trunk UPS
		Overall, n=17	nivolumab, n=8	ipilimumab/nivolumab, n=9	Overall, n=10	nivolumab+ nivolumab/RT, n=6	ipilimumab / nivolumab+ nivolumab/RT, n=4
Age at randomization (years), median (IQR)		68 (50, 70)	70 (65, 71)	52 (49, 70)	68 (59, 70)	62 (55, 70)	70 (68, 72)
Gender, n (%)	Female	2 (12%)	1 (12%)	1 (11%)	4 (40%)	3 (50%)	1 (25%)
	Male	15 (88%)	7 (88%)	8 (89%)	6 (60%)	3 (50%)	3 (75%)
ECOG performance status, n (%)	0	11 (65%)	7 (88%)	4 (44%)	7 (70%)	4 (67%)	3 (75%)
	1	6 (35%)	1 (12%)	5 (56%)	3 (30%)	2 (33%)	1 (25%)
Tumor size (RECIST, cm), median (IQR)		9 (7, 14)	9 (7, 13)	9 (7, 14)	5.6 (5.1, 8.6)	5.7 (5.1, 8.6)	5.6 (5.2, 6.7)
Disease status, n (%)	Primary	6 (35%)	2 (25%)	4 (44%)	9 (90%)	5 (83%)	4 (100%)
	Recurrent	11 (65%)	6 (75%)	5 (56%)	1 (10%)	1 (17%)	-
Multifocal, n (%)		8 (47%)	5 (62%)	3 (33%)	0	0	0
Hyalinization (%), median (IQR)		9 (4, 20)	11.3 (5, 20)	7.5 (0, 27.5)	72 (41, 95)	90 (45, 95)	61.5 (22.5, 96)
Viable tumor (%), median (IQR)		78 (64, 91)	80 (70, 90)	70 (47.5, 95)	4 (0, 14)	2.5 (0, 15)	6 (1, 47.5)
Treatment response (≥ 30% hyalinization), n (%)		3 (19%)	1 (12.5%)	2 (22.2%)	9 (90%)	6 (100%)	3 (75%)
Overall survival at 24 months, median (IQR)		82% (55, 94)	88% (39, 98)	78% (36, 94)	90% (47, 99)	83% (27, 97)	100%
Relapse-free survival at 24 months, median (IQR)		38% (15, 60)	25% (4, 56)	50% (15, 77)	78% (36, 94)	60% (13, 88)	100%
RECIST response, n (%)	PR	1 (6%)	1 (12%)	0	2 (20%)	1 (17%)	1 (25%)
	SD	9 (53%)	5 (62%)	4 (44%)	6 (60%	4 (67%)	2 (50%)
	PD	7 (41%)	2 (25%)	5 (56%)	2 (20%)	1 (17%)	1 (25%)
PD-L1 >1% at baseline*, n (%)	Positive	12 (80%)	4 (57%)	8 (100%)	8 (89%)	6 (100%)	2 (67%)
	Negative	3 (20%)	3 (43%)	0	1 (11%)	0	1 (33%)
	Unknown	2	1	1	1	0	1

^*Assessed by IHC 28–8 clone

Dedifferentiated Liposarcoma, DDLPS; Interquartile Range, IQR; Radiation Therapy, RT; Response Criteria in Solid Tumors, RECIST; Society for Immunotherapy of Cancer, SITC; Undifferentiated Pleomorphic Sarcoma, UPS

Table 2.

Patient characteristics at baseline and clinical endpoints by treatment group

Factor	Category	Retroperitoneal DDLPS			Extremity/trunk UPS
		Overall, n=17	nivolumab, n=8	ipilimumab/nivolumab, n=9	Overall, n=10	nivolumab+ nivolumab/RT, n=6	ipilimumab / nivolumab+ nivolumab/RT, n=4
Race, n (%)	Other	1 (6%)	-	1 (11%)	-	-	-
	White	16 (94%)	8 (100%)	8 (89%)	10 (100%)	6 (100%)	4 (100%)
# Previous resections, n (%)	0	6 (35%)	2 (25%)	4 (44%)	9 (90%)	5 (83%)	4 (100%)
	1	7 (41%)	4 (50%)	3 (33%)	-	-	-
	≥2	4 (24%)	2 (25%)	2 (22%)	1 (10%)	1 (17%)	-
Number of dose received, median (IQR)	1	3 (18%)	0	3 (33%)	1 (10%)	1 (17%)	0
	2	2 (11%)	1 (12%)	1 (11%)	2 (20%)	0	2 (50%)
	3	12 (71%)	7 (88%)	5 (56%)	0	0	0
	4	N/A	N/A	N/A	7 (70%)	5 (83%)	2 (50%)
Resection margins	R0	-	-	-	9 (90%)	5 (83%)	4 (100%)
	R1	-	-	-	1 (10%)	1 (17%)	0
	R0/R1	15 (88%)	6 (75%)	9	-	-	-
	R2*	2 (12%)	2 (25%	0	0	0	0
Sarculator Overall Survival, median % (IQR) **		23% (15, 38)	18 (13, 22)	31 (23, 46)	81% (79, 82)	80 (79, 86)	81 (78, 81)
Sarculator Disease-Free Survival, median % (IQR) ***		2% (1, 14)	1% (0, 5)	10% (1, 14)
Sarculator Distant Metastasis at 5 years, median % (IQR)					22% (20, 24)	22 (20, 34)	22 (20, 26)
Relapse-yes, n (%)		9 (53%)	5 (63%)	4 (44%)	3 (30%)	3 (50%)	0
First site of relapse	Locoregional	7 (42%)	5 (63%)	2 (22%)	0	0	0
	Distant	2 (12%)	0	2 (22%)	3 (30%)	3 (50%)	0

Dedifferentiated Liposarcoma, DDLPS; Eastern Cooperative Oncology Group, ECOG; Interquartile Range, IQR; negative microscopic resection margins, R0; positive microscopic resection margins, R1; incomplete gross resection, R2; Radiation Therapy, RT; Undifferentiated Pleomorphic Sarcoma, UPS

^*R2 for residual well-differentiated liposarcoma; all dedifferentiated liposarcoma completely resected

^**at 6 years for recurrent DDLPS, at 7 years for primary DDLPS, and at 5 years for UPS

^***at 6 years for recurrent DDLPS, at 7 years for primary DDLPS

Overall, 70.4% (n=19) of patients completed all doses of neoadjuvant ICB therapy (Table 2). Patients in the nivolumab arms were more likely to complete therapy than those in the ipilimumab/nivolumab arms: DDLPS: 87.5% (n=7/8) nivolumab arm A, 55.6% (n=5/9) ipilimumab/nivolumab arm B; UPS: 83.3% (n=5/6) nivolumab/RT arm C, 50% (n=2/4) ipilimumab/nivolumab + nivolumab/RT arm D. All UPS patients completed preoperative RT.

All 27 patients (intention-to-treat, ITT) who received at least 1 dose of neoadjuvant ICB on trial underwent surgical resection, although 2 patients developed lung metastases preoperatively (7.4%) and 1 underwent delayed surgery due to toxicity. One UPS patient in the nivolumab/RT arm developed lung metastases on preoperative imaging but underwent resection on trial followed by systemic chemotherapy. One DDLPS patient in the ipilimumab/nivolumab arm developed lung metastases, received systemic chemotherapy, underwent resection off trial (D227) and was therefore considered not to have undergone surgery on trial in the ITT analysis. Another DDLPS patient in the ipilimumab/nivolumab arm developed grade 3 leukocytosis and biliary stasis/hyperbilirubinemia, was treated with prednisone, mycophenolate mofetil and tocilizumab and underwent delayed surgery on trial (D84).

Toxicities were consistent with known safety profiles of nivolumab and nivolumab/ipilimumab; there were no new safety concerns (Extended Data 1). Seven of 27 patients (25.9%) experienced grade 3 or higher treatment-related adverse events. One life-threatening event not thought to be related to treatment was observed in the DDLPS nivolumab arm (post-operative reversible renal failure). Post-operative complications were as expected including one anastomotic leak resulting in reversible renal failure and atrial fibrillation requiring temporary hemodialysis, 4 patients with anemia and 1 wound infection. One patient developed post-operative immune-related colitis, requiring steroid treatment for 30 days post-operatively. There were no perioperative deaths within 90 days.

Clinical activity of neoadjuvant ICB +/− RT.

In the ITT analysis of 27 randomized patients, the primary endpoint of percent hyalinization as a continuous variable was observed to be a median of 8.8% (IQR 2.5–20) in DDLPS and 89% (IQR 40–95) in UPS (Table 1, Extended Data 2a) and was similar between nivolumab (DDLPS: 15%, UPS: 90%) and ipilimumab/nivolumab (DDLPS: 7.5%, UPS: 61.5%). Median residual viable tumor was 77.5% (IQR 62.5–92.5%) in DDLPS and 3.5% (IQR 0–15) in UPS and similar between groups (Table 1). There was an inverse correlation between hyalinization and viable tumor (Spearman correlation −0.88, p<0.001) in the whole cohort (Extended Data 2b).

The radiographic objective response rate (ORR) by Response Criteria In Solid Tumors (RECIST) 1.1¹² was 5.9% in DDLPS (1/17 partial responses [PR]; 95% confidence interval [CI] 0.15–28.7%; Fig. 3a) and 20% in UPS (2/10 PR; 95% CI 2.5–55.6%; Fig. 3b). There is no validated cutoff for pathologic response in STS^13,14. However, we identified 30% hyalinization as the optimal cut-off for pathologic response with receiver-operating curves based on landmark analysis of early relapse within 52 weeks after surgery. At least 30% hyalinization was observed in 17.6% of DDLPS patients (n=3/17; 95% CI 3.8–43.4%; Fig. 3c) and 90% of UPS patients (n=9/10; 95% CI 55.5–99.8%; Fig. 3d) and was used for correlative analyses to discriminate between pathologic responders and non-responders (Fig. 3e-f). There was no correlation between pathologic and radiographic response (Spearman, Rho=−0.34, p=0.086; Extended Data 2c).

Fig. 3: Clinical responses to neoadjuvant nivolumab and ipilimumab/nivolumab in resectable retroperitoneal dedifferentiated liposarcomas and extremity/truncal undifferentiated pleomorphic sarcomas.

a,b, Waterfall plots of radiographic percentage change in overall tumor size from baseline at least 14 days after last dose in (a) DDLPS patients treated with nivolumab or ipilimumab/nivolumab, and (b) UPS patients treated with nivolumab and concurrent RT or ipilimumab/nivolumab followed by nivolumab and concurrent RT. Asterisk indicates surgical samples with pathologic treatment response (≥30% hyalinization). Dashed black line at 20% point depicts cutoff for PD by RECIST 1.1. Dashed black line at −30% depicts cutoff for PR by RECIST 1.1. c,d, Proportion of pathologic response (≥30% hyalinization) in (c) DDLPS patients treated with nivolumab or ipilimumab/nivolumab and (d) UPS patients treated with nivolumab and concurrent RT or ipilimumab/nivolumab followed by nivolumab and concurrent RT who underwent on-trial surgery. e,f, Representative images of RECIST and pathologic evaluation (H&E slides) (e) pre- and post- nivolumab/ipilimumab in a DDLPS patient who had a radiographic PD, but had a pathologic response and has no evidence of disease (NED) after 36 months of follow-up and (f) pre- and post- nivolumab/RT in a UPS patient who had radiographic PR, a pathologic response, and who remains NED after 12 months of follow-up. Pre-therapy H&E slides are 100X and scale indicates 100μm, post-therapy H&E slides are 10X and scales indicate 200μm. g, Kaplan-Meier curves of probability of RFS in DDLPS patients treated with nivolumab (n=8) and ipilimumab/nivolumab (n=8) from on-trial surgery to sarcoma relapse or death. h, Kaplan-Meier curves of probability of RFS in UPS patients treated with nivolumab + RT (n=5) and ipilimumab/nivolumab followed by nivolumab + RT (n=4) from surgery to sarcoma relapse or death.

After a planned data safety monitoring review of the first 5 patients/group that completed therapy, it was noted that there was increased toxicity in the arms with combination nivolumab and ipilimumab. Therefore, a protocol addendum was made to change dosing of ipilimumab and nivolumab on week 1, which was done for the remaining participants in the trial. There was no association between dose schedule and pathologic response (Extended Data 2d-e).

Survival outcomes after neoadjuvant ICB +/− RT.

All 27 patients were followed for relapse and survival. At database lock (February 28, 2022), median follow-up was 38.5 months (IQR 26–42.5) and thirteen patients (n=13/25, 52%; DDLPS: n=11/16, 69%; UPS: n=2/9, 22%) relapsed after surgery. RFS at 24 months was 38% in DDLPS (IQR 15–60; arm A 25%, IQR 4–56; arm B 50%, IQR 15–77) and 78% in UPS (IQR 36–94; arm C 60%, IQR 13–88; arm D 100%; Fig. 3g-h). All UPS relapses were distant lung metastases whereas DDLPS relapses were primarily locoregional (n=9/11, 82%, Table 2).

OS at 24 months was 82% in DDLPS (IQR 55–94; arm A 88%, IQR 39–98; arm B 78%, IQR 36–94) and 90% in UPS (IQR 47–99; arm C 83%, IQR 27–97; arm D 100%; Extended Data 3a-b). Five patients died of recurrent sarcoma. One patient with recurrent DDLPS (arm A) developed abdominal recurrence 6.6 months post-operatively and died from disease 14.5 months after first treatment. Three primary DDLPS patients (arm B) developed abdominal recurrence 4.8, 6.5, and 7.3 months post-operatively and died 29, 12.7, and 25 months after first treatment, respectively. One UPS patient in arm C developed lung metastases during neoadjuvant treatment, underwent resection of the primary tumor followed by systemic chemotherapy, and died 18.6 months after first treatment.

To address the impact of radiographic and pathologic responses to ICB on relapse, we performed exploratory analyses of RFS (Extended Data 3c-d). There was one relapse among the 3 patients with RECIST PR (33%), 9 relapses among the 15 patients with stable disease (SD; 60%), and 5 among the 9 patients with progressive disease (PD; 56%). RECIST was not associated with RFS or OS. Percent hyalinization at surgery was not associated with RFS (Cox Hazard Ratio [HR]=0.98, p=0.12) nor OS (Cox HR=0.99, p=0.60). Notably, neither the dose of ipilimumab nor the disease status (primary vs recurrent) and focality were associated with survival (Extended Data 3e-h).

Programmed Death Ligand 1 is not associated with outcomes.

At baseline, 12 DDLPS and 8 UPS tumors were positive for Programmed Death Ligand 1 (PD-L1) expression (membranous expression by tumor cells ≥ 1%; Table 1). Baseline tumor PD-L1 positivity was not associated with any treatment outcomes, including hyalinization or viable tumor at surgery, pathologic treatment response (≥30% hyalinization), early relapse (metastatic progression before surgery or relapse within 52 weeks following surgery), ORR (Extended Data 4a-b) or RFS (log-rank DDLPS p=0.13, UPS p=0.58; Extended Data 4c-d). Percent expression of PD-L1 as a continuous variable at baseline was not associated with percent hyalinization as a continuous variable at surgery (Spearman Rho=0.21, p=0.32). The number of PD-L1 positive tumors tended to decrease among both DDLPS and UPS patients with treatment (Kruskal-Wallis DDLPS p=0.23, UPS p=0.08).

In an unplanned subgroup analysis of DDLPS, all primary DDLPS were PD-L1 positive (Extended Data Table 1), and baseline PD-L1 status was associated with RFS in recurrent DDLPS (log-rank, p=0.022; Extended Data 4e). PD-L1 status was not associated with focality (p>0.9; Extended Data Table 1) nor with RFS in subgroup analysis by focality (Extended Data 4f).

Immune infiltration is associated with outcomes.

In both cohorts, patients with the highest overall lymphocytic infiltration at baseline did not experience early relapse (Fig. 4a-b; Extended Data 5a). As we have previously demonstrated an impact of cytotoxic T-cells and T-regulatory cells (T-regs) on survival with ICB treatment in advanced STS⁸, we focused our analysis on these two lymphocyte populations. Patients with higher baseline infiltration of cytotoxic T-lymphocytes (CD3+CD8+/CD3+>17% determined by optimal cut-off) had better RFS (log-rank, DDLPS p<0.01, UPS p=0.46; Fig. 4c). There was no association between response and infiltration by CD3+CD8+ lymphocytes (p=0.34). In contrast, baseline infiltration by T-regs (CD3+FoxP3+CD8-/CD3+>15% determined by optimal cutoff) was associated with shorter RFS (log-rank, DDLPS p<0.01, UPS p=0.25; Fig. 4d). Pathologic non-responders had higher densities of T-regs at baseline (CD3+FoxP3+CD8-, Wilcoxon p=0.037; Fig. 4e), with consistent trends in both histologies. Additionally, higher densities of effector T-regs (CD3+CD45RO+FoxP3+CD8-) on-treatment were associated with absence of pathologic response (p=0.012; Fig. 4f), with consistent trends in both histologies.

Fig. 4: Immune infiltration association with early relapse, survival, and pathologic treatment response.

a,b, Supervised hierarchical clustering of intratumoral immune densities of selected immune cell populations assessed by mIF at baseline in (a) DDLPS and (b) UPS patients. Absolute densities of immune cell populations have been scaled and are presented separately in early relapsing patients and non-early relapsing patients. Early relapse was defined as progression before surgery or relapse within 52 weeks following surgery. c,d, Kaplan-Meier curves of probability of RFS from surgery to relapse or death, according to baseline intratumoral relative density of (c) cytotoxic T-cells (CD3+CD8+) and (d) T-regs (CD3+FoxP3+CD8-) in DDLPS (left panels) and UPS patients (right panels). P-values indicate log-rank comparison of survival curves. e,f, Absolute densities (/mm²) of (e) intratumoral T-regs (CD3+FoxP3+CD8-) and (f) effector T-regs (CD3+CD45RO+FoxP3+CD8-) in pathologic responders (≥30% hyalinization) and non-responders (n=12 vs n=14), at each time point (baseline, on-treatment, and surgery). Left panels are for the overall cohort and right panels are subgroup analyses by histology. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Empty circles indicate DDLPS, solid dots indicate UPS. Two-sided p-values are Wilcoxon-rank sum test comparisons of densities of cells. g, Example of mIF images of cytotoxic T-cells (CD3+CD8+), T-regs (CD3+FoxP3+CD8-), and effector T-regs (CD3+CD45RO+FoxP3+CD8-). Scale shows 50μm.

Tumor-associated macrophages have been associated with resistance in other studies¹⁵, however, there was no association between macrophages and pathologic response in this trial (Extended Data 5b-g).

To evaluate the impact of heterogeneous clinical factors in our DDLPS cohort, we performed subgroup analyses based on disease status (primary vs recurrent) and focality for infiltration by cytotoxic T-cells and T-regs. There was no difference in infiltration by cytotoxic T-cells in these subgroup analyses, and baseline infiltration by cytotoxic T-cells displayed either consistent trends or significant associations with improved RFS across subgroup analyses (Extended Data 6a-d). Likewise, tumors with higher baseline infiltration with T-regs showed consistent trends of impaired RFS across subgroups of DDLPS. Notably, primary DDLPS had significantly higher infiltration with T-regs at baseline compared to recurrent DDLPS (Extended Data 6e-h), although low sample numbers limit drawing significant conclusions.

The changes in immune infiltrate are histotype specific (Extended Data 7a).

Clustering of all samples across time points displayed histology-specific trends (Extended Data 7b-c). In DDLPS, samples tended to cluster by timepoint rather than by patient; there was a distinct cluster of highly infiltrated samples at surgery, not present in the matched baseline samples. UPS samples tended to cluster on a patient level: tumors which were highly infiltrated at baseline tended to remain so, while lesser infiltrated tumors at baseline continued to exhibit lower immune densities. Intriguingly, the only UPS patient with marked increase in tumor immune infiltration (patient 3) between baseline and surgery experienced early relapse.

Heterogeneity in immune infiltration is specific of cell type.

Intratumor heterogeneity has been studied across tumor types and it is now established that there are different tumor clones across regions of the same tumor, although less understood in sarcoma¹⁶. In order to study the degree of intratumor heterogeneity of the immune infiltration, we performed IHC on five separate regions of surgically resected tumors on a select group of patients with either high or low infiltration based on multiplex IF (mIF) analyses. Overall, after ICB, although there was some intratumor variation in densities of CD8+ and FoxP3+ cells, patients with high cellular densities tended to display higher densities in all regions studied (Fig. 5a-d). However, there seemed to be more intratumor differences in densities of CD163+ cells (macrophages; Fig. 5e-f). This is reflected by the standard deviations in densities of cells, with a maximum standard deviation of 405, 228, and 1776 cells/mm² in CD8+, FoxP3+, and CD163+ densities, respectively but there was significant inter-patient variations (Levene’s p<0.001) for all three markers.

Fig. 5: Intratumor Heterogeneity of Immune Cells at Surgery.

a, Images of CD8 IHC staining of CD8 positive cells (left) and image analysis example (right) in high infiltration (top) and low infiltration (bottom) regions. b, Patient-level representation of the densities of CD8 positive cells (cells/mm2) in different regions of the surgical resection specimens. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Red dots are patients without pathologic response and blue dots are patients with pathologic response. Two-sided p-values are Levene’s test for homogeneity of variance, p<0.0001. c, Images of FoxP3 IHC staining of FoxP3 positive cells (left) and image analysis example (right) in high infiltration (top) and low infiltration (bottom) regions. d, Patient-level representation of the densities of FoxP3 positive cells (cells/mm2) in different regions of the surgical resection specimens. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Red dots are patients without pathologic response and blue dots are patients with pathologic response. Two-sided p-values are Levene’s test for homogeneity of variance, p<0.0001, e, Images of CD163 IHC staining of CD163 positive cells (left) and image analysis example (right) in high infiltration (top) and low infiltration (bottom) regions. f, Patient-level representation of the densities of CD163 positive cells (cells/mm2) in different regions of the surgical resection specimens. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Red dots are patients without pathologic response and blue dots are patients with pathologic response. Two-sided p-values are Levene’s test for homogeneity of variance, p<0.0001.

Intratumoral B-cell and TLS are associated with survival.

In DDLPS, there was an increase in B cell infiltration with suggestion of TLS as assessed by CD20/CD21 immunohistochemistry (IHC; Fig. 6a) after ICB treatment, with 3 tumors having B-cell infiltrates suggestive of TLS at baseline and 8 tumors with B-cells at surgery, although only one tumor had B-cells present with suggestion of TLS at both timepoints (McNemar test; p=0.056). In contrast, in UPS intratumoral B-cells with TLS features were observed at baseline in 2 patients which were lost upon treatment, presumably due to the radiosensitivity of B-cells (Fig. 6b). Similarly, using bulk tumor RNA-sequencing data and a 290 immune gene signature by single sample gene set enrichment analysis (ssGSEA; Fig. 6c), there was an increase TLS signature in DDLPS. Tumor samples with strong B-cell and TLS signatures tended to have intratumoral B-cells with TLS features by IHC (Fig. 6c). In both cohorts, the presence of B-cells with TLS features by IHC at baseline was associated with higher intratumoral densities of T-cells (CD3+, Wilcoxon false discovery rate [FDR] p.adjusted=0.02; Extended Data 8a), cytotoxic lymphocytes (CD3+CD8+ FDR adjusted p=0.02), and T-regs (CD3+FoxP3+CD8-, Wilcoxon FDR p.adjusted=0.02; CD3+CD45RO+FoxP3+CD8-, Wilcoxon FDR p.adjusted=0.02; Fig. 6d). In contrast, the presence of intratumoral B-cells with TLS features was not associated with higher densities of T-regs at surgery, which was also seen in using the RNA-sequencing B-cell signature (Fig. 6e). This association was specific to T-regs, in contrast to most other immune cell densities. The presence of B-cells at surgery was significantly associated with higher infiltration by multiple immune cells populations at surgery (Extended Data 8a) but the baseline association with T-regs was lost.

Fig. 6: Intratumoral B-cells and TLS association with other intratumoral immune cells and oncological outcomes.

a, Images of CD20 and CD21 IHC staining of B-cell aggregates in a DDLPS (left panels) and a UPS (right panels) at 200x. b, Proportion of specimens showing presence of TLS features by IHC in DDLPS (top panels) and UPS (bottom panels) patients. Samples which failed quality control are not represented (baseline n=4 DDLPS, n=1 UPS; on-treatment n=2 DDLPS, n=2 UPS, surgery n=0 DDLPS, n=3 UPS). c, Changes in RNAseq ssGSEA scores of B-cells (left panels) and TLS (right panels) in DDLPS (top panels) and UPS (bottom panels) patients. Data are presented as minima from maxima, solid line in the violin plot indicates median. Samples with (blue dots) or without (red dots) TLS features by IHC are represented. Grey dots indicate samples without IHC data available. Dashed grey lines indicate matched samples. P-values are Kruskal-Wallis tests. d, Absolute densities (/mm²) of intratumoral T-regs (left panel) and effector T-regs (right panel) in samples with (blue circles and dots) and without (red dots and circles) TLS features by IHC, at each time point. Data are presented as minima from maxima, solid line in the violin plot indicates median. Empty circles indicate DDLPS, solid dots indicate UPS. Two-sided p-values are Wilcoxon-rank sum test, e, Correlation between T-regs and B-cells RNAseq ssGSEA scores in samples, at each time point. Empty circles indicate DDLPS, solid dots indicate UPS. Two-sided p-values are Spearman rank-order correlation tests, grey area is 95% CI. f, Kaplan-Meier curves of RFS (left panel) and OS (right panel) according to presence of TLS features by IHC in DDLPS specimens at surgery. P-values indicate log-rank comparison. g, Kaplan-Meier curves of RFS (left panels) and OS (right panel) according to presence of TLS features by IHC in specimens of UPS patients at baseline. P-values indicate log-rank comparison. h, Kaplan-Meier curves of RFS (left panel) and OS (right panel) according to high (top quartile) or low (remaining quartiles) RNAseq ssGSEA scores of TLS in specimens of the whole cohort at baseline. P-values indicate log-rank comparison.

In DDLPS, although the presence of intratumoral B-cells with TLS features by IHC at baseline was not associated with RFS or OS (Extended Data 8b), patients with intratumoral B-cells with TLS features by IHC at surgery had significantly better OS (median: 29.1 months vs. NR, n=5/8 vs. 0/8, log-rank p=0.045) and a trend towards longer RFS (median: 13.4 months vs. 40, n=7/8 vs. 4/8, log-rank p=0.14; Fig. 6f). In UPS, neither of the two patients with baseline infiltration by B-cells with TLS features by IHC have relapsed and both are alive at last follow-up (Fig. 6g). Using RNA-sequencing data, we found that patients with higher TLS immune signature (top quartile) had non -significant longer RFS (log-rank p=0.32) and OS (log-rank p=0.14; Fig 6h), which was consistent in each histotype group (Extended Data 8c-d).

In subgroup analyses based on disease status at baseline (primary vs recurrent) and focality, there was a numerical increase in the number of tumors containing B-cells with TLS features in all subgroups. Additionally, patients with presence of B-cells with TLS features at surgery had a non-significant improved OS across all the subgroup analyses (Extended Data 9a-f).

To assess for intratumor heterogeneity in B-cells with TLS features, we evaluated surgical resection blocks for a subset of patients and counted the number of lymphoid aggregates for one slide on each available block of the surgical resection (mean= 13 slides/tumor, minimum =3 slides/tumor, maximum= 49 slides/tumor). Overall, patients that had tumors positive for B-cells with TLS features had significantly higher number of lymphoid aggregates (p<0.01; Extended Data 9g-h). Additionally, all slides analyzed for UPS patients had very low numbers of lymphoid aggregates, with a median of 0 lymphoid aggregate per slide.

Discussion

Here we report a randomized neoadjuvant trial of ICB in patients with resectable DDLPS and UPS with pathologic response (hyalinization) as the primary endpoint. The toxicity profile was overall manageable with no new safety concerns. Across tumor types, neoadjuvant ICB trials have shown increased activity and these trial designs have gained a lot of attention^17–22, suggesting that ICB may be more effective in the earlier, localized setting than in the advanced, metastatic setting^23,24.

In the UPS cohort, striking pathologic responses were observed with concurrent ICB and RT. Historic data from our institution of pre-operative RT in 17 extremity/truncal UPS demonstrated a median hyalinization of 17.5% and the pathologic complete response rate (0% viable tumor) of 9% ²⁵. The European Organization for Research and Treatment of Cancer assessed pathologic response and survival after preoperative RT in 100 patients with STS. The median hyalinization was 10% for the whole cohort and for the unclassified sarcoma cohort (n=34), and 5% for the pleomorphic sarcoma cohort (n=6)¹³. Median viable tumor at surgery was 30% in the the unclassified sarcoma cohort and 73% in pleomorphic sarcoma¹³. Although our pathologic response data compares very favorably with these reports, a formal prospective comparison between preoperative RT and preoperative combination of RT with concurrent ICB has not been reported.

While pathologic response to ICB was not as robust in DDLPS, 1-year RFS was 71%. This is notable as DDLPS patients in this study are characterized by unfavorable prognostic factors, including high grade, recurrent DDLPS, and multifocal disease. Such patients historically have poor overall oncologic outcomes with estimated 6-year DFS of 6.5%, 6-year OS of 32.2%, and 1-year DFS after a second surgery for relapse of 50%²⁶. In contrast, in primary retroperitoneal liposarcoma, a recent phase 3 trial (STRASS) reported a 3-year abdominal RFS of 60.4% and an estimated 1-year abdominal RFS of 70%²⁷ including low grade histology with more favorable prognostic factors. Thus, further investigations of neoadjuvant ICB in DDLPS patients may be warranted, with combination treatments and biomarker-based selection of patients.

The higher clinical benefit seen in our trial in UPS compared to DDLPS patients may not be solely attributed to histology-specific immune-sensitivity. RT has known immune-modulating effects²⁸ and the addition of RT to ICB has demonstrated increased response rates compared to ICB alone in other cancer types²⁹. Neoadjuvant RT may be challenging in retroperitoneal diseases, however, the STRASS trial has shown its feasibility with modest benefit in unplanned subgroup analysis when used alone²⁷. A French multicenter neoadjuvant trial with sequential ICB and RT is enrolling patients with STS (NCT03474094). The SARC032-SU2C trial (NCT03092323) is a multicenter randomized trial of neoadjuvant RT with or without ICB in resectable DDLPS and UPS³⁰. Data from these trials will provide further insight regarding the benefit of RT in combination with ICB in STS.

The optimal endpoint for pathologic response after neoadjuvant therapy in sarcoma is evolving and ill-defined. At the time of protocol activation, there were no data regarding pathologic response after neoadjuvant ICB. Others as well as our group had identified percent hyalinization as a reasonable surrogate marker for outcomes after neoadjuvant RT^13,25; this was thus chosen as the primary endpoint. In the current trial, we were unable to identify an optimal cutoff for percent hyalinization that was strongly associated with oncologic outcomes, but the trial was not powered to do so. We identified 30% hyalinization as a reasonable surrogate endpoint for major pathologic response for our correlative analysis. Larger studies of histotype-tailored criteria may be key to define “optimal” cutoffs for future neoadjuvant ICB studies.

Previously reported RECIST response rates to ICB in the metastatic setting for DDLPS and UPS range from 7–29%^4,5,31 and similar in our study. However, we found no correlation between pathologic and radiographic response, highlighting the need for better assessments of response to neoadjuvant therapy in sarcoma. Imaging evaluation in sarcomas has remained a challenge, despite several attempts to address this, including a dedicated neoadjuvant prospective trial with standard of care treatments, comparing several imaging modalities and criteria^32,33. Additionally, imaging evaluation can be more challenging in the neoadjuvant setting compared to metastatic setting. For instance, the immune-related RECIST evaluation requires confirmation of response or progression with repeat imaging four weeks after initial evaluation in some cases³⁴, which is not feasible neoadjuvant trials.

Immunologic correlative studies recapitulated several known immune biomarkers of response to ICB³⁵. Tumor PD-L1 expression in sarcomas is overall associated with worse prognosis and more advanced disease^36,37. However, PD-L1 is a dynamic biomarker affected by treatments such as pre-operative radiation²⁵ and chemotherapy³⁸, or combination of pembrolizumab and talimogene laherparepvec in advanced diseases³⁹. In our trial, we found that PD-L1 expression numerically decreased with neoadjuvant ICB in DDLPS and UPS; this may be due to a direct pharmacologic effect of anti-PD1 or this could also be due to fewer tumor cells present at time of surgery. The predictive impact of baseline expression of PD-L1 with ICB treatment is controversial and no trial to date has displayed a significant association in STS, although it has been noted that responders are more likely to express PD-L1 in the advanced setting⁸, and another trial demonstrating that PD-L1 status may be more informative on treatment rather than before ICB treatment⁹. Likewise, our data may suggest that PD-L1 positive tumors have a longer ”tail” on the survival curves compared to PD-L1 negative tumors; however, these results are largely non-significant.

Presence of cytotoxic T cells and absence of T-regs at baseline were associated with RFS and OS, particularly in DDLPS patients. In contrast, data from the SARC028 trial reported that the presence of T-regs within tumors at baseline in advanced disease was associated with improved RFS with use of pembrolizumab⁸. The contrasting results between these trials is in line with data from other tumor types, where T-reg infiltration was a negative prognostic marker in most situations but a positive predictive marker in others³⁵. This observation warrants further evaluation into the different phenotypes of T-regs and their interaction with other cells in the tumor microenvironment^40–43.

Our group previously showed that patients with advanced STS expressing a B-cell gene expression signature and characterized by presence of intratumoral TLS had a 50% response to ICB¹⁰. In the current neoadjuvant trial, we found that the presence of intratumoral B-cells at surgery after neoadjuvant ICB treatment was associated with improved OS whereas their presence at baseline was not associated with prognosis in DDLPS. Notably, baseline tumor infiltration by B-cells was associated with higher T-reg densities but this association was lost upon ICB treatment. Several studies have shown an interaction between T-regs and TLS including murine models of fibrosarcoma in which T-regs impede TLS formation⁴⁴ and lung adenocarcinoma which showed improved tumor control after depletion of T-regs, predominantly present in TLS⁴⁵. In melanoma improved responses to ICB were reported after depletion of follicular T-regs, which are T-regs located in TLS⁴¹. Interestingly, the recently published PEMBROSARC trial, also found a prognostic implication of infiltration of TLS by T-regs⁴⁶.

In conclusion, our findings provide evidence that neoadjuvant ICB in combination with RT in UPS is safe with significant pathologic responses and promising survival benefit. Neoadjuvant ICB in DDLPS and UPS is associated with complex immune changes in the tumor microenvironment including stimulation of TLS and B-cells and disruption of associations between B-cells and T-regs. Further studies are needed to optimize these regimens, determine the long-term benefits, and fully elucidate the mechanisms of response and resistance.

Methods

Inclusion and ethics.

This research complies with all relevant ethical regulations. Written informed consent was provided by all study participants before treatment and this trial adhered to all relevant ethical considerations. The study was approved by The University of Texas MD Anderson Cancer Center’s Institutional Review Board and monitored by Data and Safety Monitoring Board provided. Data were collected and analyzed by the investigators, and all authors approved and agreed to submit the final manuscript for publication. The authors vouch for the accuracy and completeness of the data and for the fidelity of the trial to the study protocol. This trial was pre-registered on clinicaltrials.gov on 10/04/2017, https://clinicaltrials.gov/study/NCT03307616.

Statistics and Reproducibility.

This is an investigator-initiated, randomized, open-label, single institution, non-comparative phase II study designed to detect pathologic and immunologic biomarkers of response to ICB in resectable, treatment-naive primary or locally recurrent DDLPS of the retroperitoneum (arms A/B) and UPS of the trunk or extremities (arms C/D). CONSORT guidelines were followed⁴⁷ and the study protocol is included in the Supplementary Information. Initial intended accrual was 10 patients in each arm, however the trial was terminated early due to slow accrual and in the context of the COVID-19 pandemic.

Patients with treatment-naïve primary or locally recurrent retroperitoneal DDLPS were randomized in a 1:1 ratio. In arm A, patients with treatment-naïve primary or recurrent retroperitoneal DDLPS received 3 doses of nivolumab 3mg/kg every 2 weeks on weeks 1, 3 and 5 prior to surgical resection. In arm B, patients with treatment-naïve primary or recurrent retroperitoneal DDLPS received 1 dose of ipilimumab 3mg/kg combined with nivolumab 1mg/kg on week 1 followed by 2 doses of nivolumab 3mg/kg every 2 weeks on week 3 and 5 prior to surgical resection. Subject randomization was implemented by the Clinical Trial Conduct website maintained by the Department of Biostatistics at the University of Texas M.D. Anderson Cancer Center (https://biostatistics.mdanderson.org/ClinicalTrialConduct).

Patients with treatment-naïve primary or recurrent extremity or truncal UPS were randomized in a 1:1 ratio between arm C and arm D, both of which include combination nivolumab 3mg/kg every 2 weeks on weeks 3, 5 and 7 + 50 Gy RT in 25 fractions. In arm C, UPS patients received 1 dose of nivolumab 3mg/kg on week 1 followed by combination nivolumab + RT, as described above. In arm D, UPS patients received 1 dose of combination ipilumumab 3mg/kg + nivolumab 1mg/kg, followed by combination nivolumab + RT.

After a planned data safety monitoring review of the first 5 patients/group that completed therapy, it was noted that there was increased toxicity in the arms with combination nivolumab and ipilimumab. Therefore, a protocol addendum was made to change dosing to 1 dose of ipilimumab 1mg/kg combined with nivolumab 3mg/kg on week 1 in both DDLPS and UPS, which was done for the remaining participants in the trial.

The primary endpoint was defined as the percent of hyalinization in the surgical resection specimen in each arm^13,25. With a sample size of 26 (13/arm) for the DDLPS cohort and a sample size of 14 (7/arm) for the UPS cohort, the trial would have had 80% power to detect an effect size of 1.145 and 1.632, respectively. Other secondary endpoints were to assess the change in immune infiltrate in response to neoadjuvant nivolumab monotherapy and neoadjuvant nivolumab and ipilimumab combination therapy, to assess the ORR (defined as rate of patients achieving partial and complete response) of nivolumab monotherapy and nivolumab and ipilimumab combination therapy administered in the neoadjuvant setting as assessed by imaging (RECIST1.1 and immune-related Response Criteria), to assess the 12- and 24-month RFS (defined as the time from surgery to recurrence) and OS (defined as the time from initiation of treatment to death of any cause), and to evaluate the safety of nivolumab monotherapy and combination ipilimumab and nivolumab in the neoadjuvant setting and peri-operatively by CTCAE version 4.0 criteria.

As pathologic response was not associated with survival, we performed a landmark analysis at one year after surgery to define clinically meaningful response criteria for analysis of correlates. Early relapsing patients included all patients who had either progressed before surgery or within 52 weeks following surgery: 7 patients were considered early-relapsing, including 5 DDLPS and 2 UPS. Hyalinization at surgery as a continuous variable was not associated with RFS (Cox p=0.70) nor OS (Cox p=0.61). In order to select an optimal cut-off point of hyalinization to define pathologic response, we ran sensitivity analyses. Receiver-operating curves of one-year RFS prediction found an area under the curve of 0.465 when using hyalinization as a predictor. Two cut-off points were deemed optimal using the criteria that minimized the difference between sensitivity and specificity: 20% and 30% hyalinization. To select one of these two cutoff points, we ran optimal cut-off points by Log-Rank analysis of RFS: the optimal cut-off point for the whole cohort was 45%, for the DDLPS group was 5% and for the UPS cohort was 30%. Based on these observations, we selected the optimal cut-off point of 30% hyalinization. Additionally, the analyses by the graphical and numerical methods of Lin, Wei, and Ying indicate that hyalinization can be analyzed as a linear functional term and proportional hazards assumption was not violated neither for OS nor RFS.

Descriptive statistics [frequency distribution, median (range)] were used to summarize patient’s characteristics. The primary efficacy endpoint, pathologic response, assessed at time of surgical resection by percentage hyalinization, was estimated by study cohort. The McNemar test was used to determine if there are differences on a dichotomous dependent variable between two related groups. Non-parametric unpaired tests (Wilcoxon rank sum test and Kruskal Wallis) were used to compare continuous variables between groups and adjusted for multiple comparison by FDR, as required. Comparison between categorical variables were done using chi-squared or Fisher’s exact tests, as required. The distributions of RFS, and OS were estimated by the Kaplan-Meier method. For events that have not occurred by the time of data analysis, times were censored at the last contact at which the patient was known to be progression or recurrence free for RFS or the last time the patient was known to be alive for OS. Log-rank test was performed to test the difference in survival between groups. The linear functional form and proportional hazards assumption for of hyalinization in the Cox model for RFS and OS were assessed using the graphical and numerical methods of Lin, Wei, and Ying⁴⁸. Levene’s test was used to test homogeneity of variance across patients⁴⁹. SAS version 9.4 and R version 4.1.3 are used to carry out the computations for all analyses. Data was transformed using dplyr R package version 1.0.8⁵⁰, analyzed using rstatix R package⁵¹, and plots were generated using ggpubr R package version 0.4.0⁵² and ggplot2 R package version 3.3.5⁵³.

Study data were collected and managed using REDCap electronic data capture tools hosted at The University of Texas MD Anderson Cancer Center^54,55. REDCap (Research Electronic Data Capture) is a secure, web-based software platform designed to support data capture for research studies, providing 1) an intuitive interface for validated data capture; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for data integration and interoperability with external sources. Data for correlative studies were collected using Microsoft Excel (v2016) spreadsheets.

Participants.

Adult subjects (>18 years) of all sex and genders with treatment-naïve primary or locally recurrent DDLPS of the retroperitoneum or UPS of the trunk or extremity were eligible for inclusion in this study only if disease was determined to be surgically resectable and candidates for upfront surgery as agreed upon by a multidisciplinary consensus (Surgical Oncology, Medical Oncology, Radiation Oncology) after presentation at sarcoma multidisciplinary conference. Patients must have recent imaging (CT or MRI, as appropriate) within 4 weeks of trial enrollment, demonstrating measurable disease as defined by RECIST1.1 and at least one tumor amenable to serial biopsy in clinic or be willing to undergo serial biopsies through image-guided procedures during the neoadjuvant phase of the protocol. Patients had to be medically fit to undergo surgery as determined by the treating medical and surgical oncology team, have Eastern Cooperative Oncology Group (ECOG) performance status 0–2, and intact cardiopulmonary and organ functions. Major exclusion criteria were prior intra-abdominal surgery within 4 weeks of enrollment, prior chemotherapy or targeted therapy for the current sarcoma, prior radiation for sarcoma in the same area, prior or concurrent immunotherapy, prior active malignancy within the 2 previous years, active autoimmune disease, or current immunosuppressive medication use.

Pathologic assessment.

Hematoxylin and eosin (H&E) slides of pre-treatment and post-treatment specimens were reviewed by pathologists who specialize in bone and soft tissue tumors (WLW, AJL). Pathologic response was determined on treated surgical specimens by recording percentage of hyalinization (decreased cellularity with dense collagen deposition), necrosis and residual viable tumor^13,56.

PD-L1 staining.

IHC study for PD-L1 was performed on four-micron whole section formalin-fixed paraffin-embedded unstained slides using the PD-L1 28–8 pharmDx kit (Aglient Dako Carpinteria, CA) on the Dako Autostainer Link 48, according to the manufacturer’s instructions. The percentage of viable tumor cells with any membranous staining and of any intensity was assessed. For tumors to be considered positive for PD-L1, a cut-off of 1% expression was used. The results were plotted using Rstudio v.3.5.3.

Single IHC staining for B-cells: CD20/CD21.

Four-micron (μm) unstained slides were prepared from representative whole section formalin-fixed paraffin-embedded tumor blocks (baseline, on-treatment, and surgical samples). IHC staining was performed for CD20 (1:1400, Clone L-26, Agilent Dako) using an autostainer (Bond Max, Leica Biosystems, Buffalo Grove, IL). If CD20 staining revealed any B-cells, CD21 (1:20, Clone 2G9, Leica Biosystems) IHC staining was performed to highlight reticular networks where labeling of cells within lymphoid aggregates were suggestive of tertiary lymphoid structures.

Multiplex Immunofluorescence.

Multiplex immunofluorescence staining was performed using similar methods as previously described and optimized^57,58 . Briefly, four mm-thick formalin fixed, paraffin embedded sample sections were stained using a mIF panel contained antibodies against: CD3 (clone D7A6E, Cell Signaling Technology), CD8 (clone C8/144B, Thermo Fisher Scientific), CD45RO (clone UCHL1, Leica Biosystems), FOXP3 (clone D2W8E, Cell Signaling Technology), PD-1 (clone [EPR4877(2)], ABCAM), KI67 (clone MIB-1, DAKO), PD-L1 (clone E1L3N, Cell Signaling Technology), and CD68 (clone [PG-M1 (M)], DAKO). All the markers were stained in sequence using their respective fluorophore contained in the Opal 7 kit (catalogue #NEL797001KT; Akoya Biosciences, Waltham, MA) and the individual tyramide signal amplification fluorophores Opal Polaris 480 (catalog #FP1500001KT) and Opal Polaris 780 kit (catalog #FP1501001KT, Akoya Biosciences)⁵⁷. The slides were scanned using the Vectra/Polaris 3.0.3 (Akoya Biosciences) at low magnification, 10x (1.0 µm/pixel) through the full emission spectrum and using positive tonsil controls from the run staining to calibrate the spectral image scanner protocol⁵⁹. A pathologist selected all the tumor area using regions of interest (ROIs) for scanning in high magnification by the Phenochart Software image viewer 1.0.12 (931 × 698 µm size at resolution 20x) in order to capture various elements of tissue heterogeneity. Each ROI was analyzed by a pathologist using InForm 2.4.8 image analysis software (Akoya Biosciences). Marker co-localization was used to identify specific cells phenotypes in the tumor. Densities of each cell phenotype were quantified, and the final data was expressed as number of cells/mm2⁵⁹. All the data was consolidated using R studio 3.5.3 (Phenopter 0.2.2 packet, Akoya Biosciences).

Intratumor Heterogeneity.

FFPE tumor tissue blocks from surgical resections of 9 UPS and 9 DDLPS patients were used to build a tissue microarray (TMA) block using the ATA-100 Advanced Tissue Arrayer (Chemicon International). Selection of cases in the DDLPS cohort was based on the density of CD3+ cells calculated by mIF: four patients with the lowest infiltration and the five patients with the highest infiltration by mIF at surgery were selected for this TMA.

In the UPS cohort, the nine patients who did not progress prior to surgery were included in this analysis. The TMA block included a total of 90 cores, each measuring 1 mm in diameter. Multi-sampling of the tissue block was performed in order to account for intratumoral heterogeneity, (3 random intratumoral areas, one area with high tumor infiltrating lymphocytes (TILs) and other with low TILs. (5 cores per sample)).

IHC studies were performed on 4-μm FFPE sections using a Leica BOND RXm autostainer. Slides were stained with antibodies targeting human CD8 (clone C8/144B; Thermo Fisher #MS457S), CD163 (clone 10D6; Leica #NCL-L-CD163), and FOXP3 (clone 206D; BioLegend #320102) using a modified version of the standard Leica Bond DAB “F” IHC protocol. Slides stained for CD8, CD163 and FOXP3 were scored by a board-certified pathologist, using digital image analysis software HALO v3.5 with a modified cytonuclear algorithm to detect the presence of positive cells per area of tissue analyzed, the results were exported as cell density (cells/mm²).

Lymphoid aggregates count.

To evaluate the intratumor heterogeneity of TLS evaluation, we evaluated the presence of lymphoid aggregates on all the surgical pathology blocks of a select group of patients. For the DDLPS cohort, we selected the same nine patients at the oens included in the intratumor heterogeneity TMA. For the UPS cohort, we selected the eight patients who had not progressed before surgery and for which TLS evaluation had been conclusive at surgery, as three patients had inconclusive evaluation due to abundant tumor necrosis. An experienced pathologist (RL) reviewed H&E whole-section slides for each surgical pathology block to count the lymphoid aggregates in each slide, which were defined as a group of more than 50 lymphoid cells located in the tumor area.

RNA extraction and Quality Control.

RNA was extracted by NORGEN Total RNA Purification Kit (Cat. 37500) (NORGEN BIOTEK CORP). The extracted RNA was treated with DNase I. The treated RNA then was cleaned-up using the AMPure XP beads (Beckman Coulter Life Sciences) and eluted into 1x TE buffer. The purified RNA was quantified using Quant-iT™ RiboGreen™ RNA Assay Kit (ThermoFisher SCIENTIFIC) and the RNA quality was accessed using Agilent RNA 6000 Nano Kit and the 2100 Bioanalyzer Instrument (Agilent Technologies).

cDNA Synthesis.

The cDNA was prepared from the extracted total RNA using Ovation® RNA-Seq System V2 (NuGEN). Amplification is initiated at the 3’ end as well as randomly throughout the transcriptome in the sample. The prepared cDNA was quantified using Quant-iT™ PicoGreen™ dsDNA Assay Kit (ThermoFisher SCIENTIFIC) and quality was accessed using Genomic DNA ScreenTape and Reagents on the Tapestation 4200 (Agilent Technologies)

RNA Library preparation.

Up to 200 ng of each cDNA sample based on the PicoGreen quantification was sheared (mechanically fragmented) using the E220 Focused-ultrasonicator Covaris (Covaris). The sonication was performed under the following conditions: Peak Incident Power 200, Duty Cycle 25%, Cycles per Burst 50, and duration 10 seconds for 120 iterations. To ensure the proper fragment size, samples were examined on TapeStation 4200 using the DNA High Sensitivity kit (Agilent Technologies). The sheared cDNA was proceeded to library preparation using SureSelect XT Low Input Reagent Kit with indexes 1–96 (Agilent Technologies) as automated method on the Sciclone G3 NGSx Workstation (PerkinElmer, Inc.).

This protocol consists of 3 enzymatic reactions for end repair, A-tailing and Adaptor ligation, followed by barcode insertion by PCR using Herculase II Fusion DNA Polymerase (8 to 14 cycles, based on input DNA quality and quantity). PCR primers were removed by using 1x volume of Agencourt AMPure PCR Purification kit (Agencourt Bioscience Corporation). The quality and quantity of the prepared libraries were evaluated using TapeStation 4200 and the DNA High Sensitivity kit (Agilent Technologies) to verify correct fragment size and to ensure complete removal of primer dimers.

Hybridization and Capture.

Subsequently, the prepared libraries were individually hybridized to Agilent SureSelect Human All Exon v.4 probes (Agilent Technologies). The hybridization steps were automated on the Sciclone G3 NGSx Workstation (PerkinElmer, Inc.). Agilent captures were hybridized as single sample reactions using 500–1000 nanograms of prepared library as input. All hybridization and post-hybridization capture & washes were performed according to Agilent’s protocol.

Briefly, the capture reagents and probes were added to the prepared libraries, the mixture was incubated at 65^oC on thermocycler with heated lid on for up to 24 hours. The targeted regions were captured using streptavidin beads and the streptavidin-biotin-probe-target complex was washed and the captured libraries were enriched by PCR amplification according to manufacturer’s protocol. The quality and quantity of each captured sample was analyzed on TapeStation 4200 using the DNA High Sensitivity kit.

RNA-Sequencing and Data Analysis.

The captured libraries were sequenced on Illumina NovaSeq 6000 platform for 2 × 150 paired end reads with an 8nt read for indexes using Cycle Sequencing v3 reagents (Illumina). Raw RNA Sequence data was processed by an in-house RNA Seq data analysis pipeline, which, among other tools uses the STAR aligner⁶⁰ to align raw reads to hg19 version of Human reference genome, featureCounts⁶¹ to quantify aligned reads with to produce raw counts, Oncofuse⁶² to filter and prioritize fusion candidate generated by the STAR aligner, and FastQC and QualiMap⁶³ to evaluate quality of raw reads and feature counts. We used VirusFinder⁶⁴ (version 2.0) to align reads that did not map to the human reference genome to a viral database that contains viruses of 32,102 known classes⁶⁵, and in particular all different variations of the Human Papilloma Virus and Polyoma Virus. RNA sequencing reads of the samples were mapped to the hg19 reference genome using the STAR aligner⁶⁰. For calculation of gene expression, the raw count of genes in the samples were converted to transcripts per million that normalize counts for library size and gene length. Similar to the Charoentong’s studies⁶⁶, the composition of the immune infiltrated cells of samples were then generated using the single sample ssGSEA enrichment scores of 34 immune gene signatures (geneset signature available in Supplementary Table 2).

Extended Data

Extended Data Fig. 1.

Frequencies of maximum-grade treatment-related adverse events (TRAEs) in the whole cohort of patients in ascending shades of blue (nivolumab arms) and red (ipilimumab/nivolumab arms) for grades 1 to 3. There were no grade 4 or 5 TRAEs in this trial. All adverse events that were possibly or probably related to treatment are reported. All patients (n = 27) were included in this analysis.

Extended Data Fig. 2.

a, Patient-level pathological data showing percent hyalinization, percent viable tumor, and percent necrosis of surgical specimens in the ITT population (n = 27) in DDLPS (left panels) and UPS (right panels) patients. Patient 29 had clinical progression with lung metastasis treated by chemotherapy, followed by surgery off trial. b, Correlation between percent viable tumor and percent hyalinization at surgery. Two-sided p-value is Spearman rank-order correlation test. Gray zone is 95% CI. c, Correlation between overall change in tumor size from baseline at least 14 days after last dose assessed by RECIST1.1 and percent hyalinization at surgery. Dots are in ascending size of percent viable tumor at surgery. Two-sided p-value is Spearman rank-order correlation test. Gray zone is 95% CI. d, Table reporting pathologic response and early relapse in patients treated in arm B, according to the dose of ipilimumab received. P values are two-sided. e, Table reporting pathologic response and early relapse in patients treated in arm D, according to the dose of ipilimumab received. P values are two-sided.

Extended Data Fig. 3.

a, Kaplan-Meier curves of OS in DDLPS patients treated with nivolumab (n = 8) and ipilimumab/nivolumab (n = 9) from surgery (on-trial and off trial) to death. b, Kaplan-Meier curves of OS in UPS patients treated with nivolumab + RT (n = 6) and ipilimumab/nivolumab followed by nivolumab + RT (n = 4). c, Kaplan-Meier curves of RFS according to radiographic response (PD, SD, PR) as assessed by RECIST1.1 from baseline at least 14 days after last dose in DDLPS patients (left panel) and UPS patients (right panel). Two-sided p values indicate log-rank comparison. d, Kaplan-Meier curves of OS according to radiographic response (PD, SD, PR) as assessed by RECIST1.1 from baseline at least 14 days after last dose in DDLPS patients (left panels), and UPS patients (right panels). Two-sided p values indicate log-rank comparison. e, Kaplan-Meier curves of RFS according to dose of ipilimumab and nivolumab received in DDLPS patients (arm B; left panel) and UPS patients (arm D; right panel). Two-sided p values indicate log-rank comparison. f, Kaplan-Meier curves of OS according to dose of ipilimumab and nivolumab received in DDLPS patients (arm B; left panel) and UPS patients (arm D; right panel). Two-sided p values indicate log-rank comparison. g, Kaplan-Meier curves of RFS (left panel) and OS (right panel) according to disease status at initiation of treatment (primary versus recurrent) in DDLPS. Two-sided p values indicate log-rank comparison. h, Kaplan-Meier curves of RFS (left panel) and OS (right panel) according to focality at initiation of treatment in DDLPS patients. Two-sided p values indicate log-rank comparison.

Extended Data Fig.4.

a, Number of samples with positive (≥1%) and negative (<1%) PD-L1 IHC staining on malignant cells by clone 28–8 at each time point in DDLPS (left panel) and UPS (right panel) patients. b, Table displaying oncological efficacy end points in DDLPS and UPS patients with baseline PD-L1 positive and negative tumors. PD-L1 positive tumors were defined as those with ≥1% on expression on malignant cells by IHC with clone 28–8. Two-sided p values are Wilcoxon rank-sum tests and Fisher’s exact tests. c, d, Kaplan-Meier curves of probability of RFS from surgery to relapse or death according to positivity of PD-L1 staining on tumor cells at baseline, in (c) DDLPS and (d) UPS patients. P values indicate log-rank comparison of survival curves. e, Kaplan-Meier curves of probability of RFS from surgery to relapse or death according to positivity of PD-L1 staining on tumor cells at baseline, in DDLPS patients with primary tumors (left panel) or recurrent tumors (right panel) at initiation of treatment. Two-sided p values indicate log-rank comparison of survival curves. f, Kaplan-Meier curves of probability of RFS from surgery to relapse or death according to positivity of PD-L1 staining on tumor cells at baseline, in DDLPS patients with unifocal (left panel) or multifocal tumors (right panel) at initiation of treatment. Two-sided p values indicate log-rank comparison of survival curves.

Extended Data Fig.5.

a, Example of mIF images of T cells (CD3+), activated T cells (CD3 + Ki67+), CD8+ cells, activated cytotoxic T cells (CD3 + CD8 + Ki67+), PD-L1 positive cytotoxic T cells (CD3 + CD8 + PD-L1+), antigen-experienced T cells (CD3 + PD-1+), memory T cells (CD3 + CD45RO + CD8-), macrophages (CD68+), and PD-L1 positive macrophages (CD68 + PD-L1+). Scale shows 50 μm. b–d, Absolute densities (/mm²) of (b) tumor-associated macrophages (CD68+), (c) activated macrophages (CD68 + Ki67+), and PD-L1+ macrophages (CD68 + PD-L1+) in pathologic responders (≥30% hyalinization; blue dots and blue circles) and non-responders (red dots and red circles) at each time point (baseline, on-treatment, and surgery) in DDLPS (circles) and UPS (dots) patients. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons of densities of cells. e–g, Absolute densities (/mm²) of (b) tumor-associated macrophages (CD68+), (c) activated macrophages (CD68 + Ki67+), and PD-L1+ macrophages (CD68 + PD-L1+) in pathologic responders (≥30% hyalinization) and non-responders at each time point (baseline, on-treatment, and surgery) in DDLPS (top panels) and UPS (bottom panels) patients. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons of densities of cells.

Extended Data Fig.6.

a, Kaplan-Meier curves of RFS, according to baseline intratumoral relative density of cytotoxic T cells (CD3 + CD8+) in primary (left panel) and recurrent (right panel) DDLPS patients. Two-sided p values indicate log-rank comparison. b, Absolute densities (/mm²) of cytotoxic T cells (CD3 + CD8+) in primary and recurrent DDLPS at each time point. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons. c, Kaplan-Meier curves of RFS, according to baseline intratumoral relative density of cytotoxic T cells (CD3 + CD8+) in unifocal (left panel) and multifocal (right panel) DDLPS patients. Two-sided p values indicate log-rank comparison. d, Absolute densities (/mm²) of intratumoral cytotoxic T cells (CD3 + CD8+) in unifocal and multifocal DDLPS at each time point. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons. e, Kaplan-Meier curves of RFS, according to baseline intratumoral relative density of T-regs (CD3+FoxP3 + CD8-) in primary (left panel) and recurrent (right panel) DDLPS patients. Two-sided p values indicate log-rank comparison. f, Absolute densities (/mm²) of intratumoral T-regs (CD3+FoxP3 + CD8-) in primary and recurrent DDLPS at each time point. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons. g, Kaplan-Meier curves of RFS, according to baseline intratumoral relative density of T-regs (CD3+FoxP3 + CD8-) in unifocal (left panel) and multifocal (right panel) DDLPS patients. Two-sided p values indicate log-rank comparison of survival curves. h, Absolute densities (/mm²) of intratumoral T-regs (CD3+FoxP3 + CD8-) in unifocal and multifocal DDLPS at each time point. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Wilcoxon rank-sum test comparisons.

Extended Data Fig. 7.

a, Absolute densities (/mm²) of intratumoral immune cells across treatment time points (baseline [BL], on-treatment [OT], and surgery [Surg]) in DDLPS and UPS patients. Dashed lines indicate paired samples. Colors indicate pathologic response. Two-sided p values are Kruskal–Wallis test for comparisons of immune densities across time points, adjusted for multiple comparison by FDR. b, c, Unsupervised hierarchical clustering of tumor specimens based on intratumoral immune densities of selected immune cell populations assessed by mIF at all time points in (b) DDLPS and (c) UPS patients. Absolute densities of immune cell populations have been scaled. Red shows higher densities of immune cells and blue indicates lower densities of immune cells. Annotations include early relapse status, presence of TLS features by IHC, and time point of tumor specimen collection.

Extended Data Fig.8.

a, Absolute densities (/mm²) of intratumoral T cells and macrophages in samples displaying B cell aggregates with TLS features by CD20 and CD21 IHC (blue circles and blue dots) and samples without B cell and TLS features by IHC (red dots and red circles), at each time point (baseline, on-treatment, and surgery). Data are presented as minima from maxima, and solid line in the violin plot indicates median. Empty circles indicate DDLPS, solid dots indicate UPS. Two-sided p values are Wilcoxon rank-sum test comparisons of densities of cells between samples with or without TLS, adjusted for multiple comparison by FDR. b, Kaplan-Meier curves of probability of RFS (left panels) and OS from surgery to death (right panel) according to presence of B cell aggregates with TLS features by CD20 and CD21 IHC in tumor specimens of DDLPS patients at baseline. Two-sided p values indicate log-rank comparison of survival curves. c,d, Kaplan-Meier curves of probability of RFS (left panels) and overall survival from surgery to death (right panels) according to high (top quartile) or low (remaining quartiles) RNA-seq ssGSEA signature scores of TLS in baseline tumor specimens of (c) DDLPS and (d) UPS patients. Two-sided p values indicate log-rank comparison of survival curves.

Extended Data Fig.9.

a, Proportion of tumor specimens showing presence of B cell aggregates with TLS features assessed by CD20 and CD21 IHC at each time point in DDLPS patients with primary (top panels) and recurrent tumors (bottom panels). Samples which did not pass quality control for analysis by IHC are not represented. b, Proportion of tumor specimens showing presence of TLS features assessed by CD20 and CD21 IHC at each time point in DDLPS patients with unifocal (top panels) and multifocal tumors (bottom panels). Samples which did not pass quality control for analysis by IHC are not represented. c–f, Kaplan-Meier curves of OS according to presence of TLS features by CD20 and CD21 IHC in tumor specimens of DDLPS patients at surgery in (c) primary, (d) recurrent, (e) unifocal, and (f) multifocal tumors. Two-sided p values indicate log-rank comparison. g, Mean number of lymphoid aggregates by H&E per slide at surgery in patients with (blue dots and blue circles; n = 5) and without (red dots and red circles; n = 12) positive evaluation of TLS features by CD20/CD21 IHC. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Empty circles indicate DDLPS (n = 9), solid dots indicate UPS (n = 9). Two-sided p values are Wilcoxon rank-sum test comparisons of densities of cells between samples with or without TLS. h, Patient-level representation of the number of lymphoid aggregates by H&E per slide for each block of the surgical resection in patients with (blue dots; n = 5) and without (red dots; n = 12) positive evaluation of TLS features by CD20/CD21 IHC. Data are presented as minima from maxima, and solid line in the violin plot indicates median. Two-sided p values are Levene’s test for homogeneity of variance.

Extended Data Table 1.

Patient characteristics at baseline and clinical end points in DDLPS patients by disease status and foculty

Factor	Category		Disease status			Disease foculty
		Overall n=17	Primary n=6	Recurrent n=11	p-value*	Unifocal n=9	Multifocal n=8	p-value*
Disease status, n (%)	Primary	6 (35%)	-	-	-	3 (33%)	5 (62%)	>0.9
	Recurrent	11 (65%)	-	-		6 (67%)	3 (38%)
# Previous resections, n (%)	0	6 (35%)	6 (100%)	-	-	3 (33%)	3 (38%)	>0.9
	1	7 (41%)	-	7 (64%)		4 (44%)	3 (38%)
	≥2	4 (24%)	-	4 (36%)		2 (22%)	2 (25%)
Multifocal, n (%)		8 (47%)	3 (50%)	5 (45%)	>0.9	-	-
Tumor size (RECIST, cm), median (IQR)		9 (7,14)	17 (13,27)	8 (5,9)	0.02	8 (3, 12)	11 (8, 15)	0.2
ECOG performance status, n (%)	0	11 (65%)	4 (67%)	7 (64%)	>0.9	5 (56%)	6 (75%)	0.6
	1	6 (35%)	2 (33%)	4 (36%)		4 (44%)	2 (25%)
Treatment arm, n (%)	nivolumab	8 (47%)	2 (33%)	6 (55%)	0.6	3 (33%)	5 (62%)	0.3
	ipilimumab nivolumab	9 (53%)	4 (67%)	5 (45%)	6 (67%)	3 (38%)
Resection margins	R0/R1	15 (88%)	6 (100%)	9 (82%)	0.5	9 (100%)	6 (75%)	0.2
	R2*	2 (12%)	0	2 (18%)		0	2 (25%)
Relapse yes, n (%)		9 (53%)	4 (67%)	5 (45%)	0.6	4 (44%)	5 (62%)	0.6
First site of relapse	Locoregional	7 (41%)	3 (50%)	4 (36%)	>0.9	2 (50%)	5 (100%)	0.2
	Distant	2 (12%)	1 (17%)	1 (9%)		2 (50%)	0
Hyalinization (%), median (IQR)		9 (4, 20)	5 (1, 9)	18 (6, 20)	6.3	12 (0, 20)	6 (5, 24)	0.8
Viable tumor (%), median (IQR)		78 (64,91)	65 (56, 89)	82 (71,89)	0.4	75 (59, 89)	80 (69, 91)	0.7
Treatment response (≥ 30% hyalinization), n (%)		3 (19%)	1 (17%)	2 (20%)	>0.9	1 (12%)	2 (25%)	>0/9
Overall survival at 24 months, median (95%CI)			80% (52, 100)	88% (67, 100)		100% (100, 100)	70% (42, 100)
Progression-free survival at 12 months, median (95%CI)			50% (22, 100)	82% (62, 100)		67% (42, 100)	75% (50, 100)
Sarculator overall survival (median %)***		23% (15, 38)	22% (9, 36)	23% (18, 38)	0.8	29% (23, 59)	16% (7, 22)	0.034
Sarculator disease-free survival (median %)***		2% (1, 14)	6% (1, 12)	2% (1, 14)	>0.9	13% (2, 15)	1% (0, 3)	0.019
RECIST response, n (%)	PR	1 (6%)	0	1 (9%)	0.2	1 (11%)	0	0.2
	SD	9 (53%)	5 (83%)	4 (36%)		3 (33%)	6 (75%)
	PD	7 (41%)	1 (17%)	6 (55%)		5 (56%)	2 (25%)
PD-LI >1% at baseline**, n (%)	Positive	12 (80%)	6 (100%)	6 (67%)	0.2	6 (86%)	6 (75%)	>0.9
	Negative	3 (20%)	0	3 (33%)		1 (14%)	2 (25%)
	Unknown	2	0	2		2	0

^*Fisher’s exact test and Wilcoxon Rank-Sum test

^**Assessed by IHC 28-8 clone

^***at 6 years for recurrent DDLPS, at 7 years for primary DDLPS

Dedifferentiated liposarcoma, DDLPS; Eastern Cooperative Oncology Group, ECOG; interquartile range, IQR; partial response, PR; progressive disease, PD; complete resection, R0/1; incomplete gross resection, R2; Response Criteria in Solid Tumors, RECIST; stable disease, SD

Data Availability

The authors declare that the data supporting the findings of this study are available within the manuscript and its supplementary information files.

RNA–sequencing data that support the findings of this study have been deposited in the Gene Expression Omnibus (GEO) under accession codes GSE202361, (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE202361), which will be released/publicly available by January 1st 2024.

All other relevant de-identified data related to the current study are available from the corresponding author (CLR) upon reasonable academic request (including compelling scientific rationale and preliminary data requiring validation through use of this cohort, this preliminary data should be presented to the authors) and will require the researcher to sign a data access agreement with the University of Texas MD Anderson Cancer Center after approval. Individual patient identifiable clinical data (such as dates) are not publicly available due to concerns with identification of patients.

The hg19 human genome can be found at https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001405.13/

Source data for all figures and extended data have been provided as Source Data files.