Autologous HER2-specific CAR T-cells after lymphodepletion for advanced sarcoma – A Phase I trial

Abstract

In this prospective, interventional phase I study (NCT00902044) for patients with advanced sarcoma, we infused autologous HER2-specific chimeric antigen receptor T-cells (HER2-CART) after lymphodepletion with cyclophosphamide (Cy) +/− fludarabine (Flu): 1×10⁸ T-cells/m² after Flu (cohort A) or Flu/Cy (cohort B), and 1×10⁸ CAR-positive T-cells/m² after Flu/Cy (cohort C). The primary outcome was assessment of safety of one dose of HER2-CART after lymphodepletion. The determination of antitumor responses was the secondary outcome. Thirteen patients were treated in 14 enrollments with 7 receiving multiple infusions. HER2-CART expanded after 19 of 21 infusions. Nine of 12 patients in cohorts A and B developed grade 1–2 cytokine release syndrome (CRS). Two patients in cohort C experienced dose limiting toxicity with grade 3–4 CRS. Antitumor activity was observed with clinical benefit in 50% of patients treated. Tumor samples analyzed showed spatial heterogeneity of immune cells and clustering by sarcoma type and treatment response. Our results affirm HER2 as a CART target and demonstrate the safety of this therapeutic approach in sarcoma.

Introduction

Metastatic and progressive sarcomas portend a poor prognosis¹. Despite their molecular heterogeneity, there is a paucity of “tumor-restricted” antigens for immunotherapy. Human epidermal growth factor receptor 2 (HER2/ERBB2) is a tumor-associated antigen (TAA) of clinical interest²; however, HER2-targeting drugs have thus far proven ineffective against tumors that overexpress HER2 without gene amplification^3,4. Chimeric antigen receptor T-cells (CART) can effectively engage surface antigens expressed at relatively low density, thereby offering a potential therapeutic avenue against tumors with modest HER2 expression^5–7.

We previously reported the safety of autologous T-cells expressing a CAR with an FRP5-single-chain variable fragment (scFv) and CD28 co-stimulatory domain (HER2-CART) in patients with HER2-positive tumors^8,9. Although some patients benefitted, clinical responses were limited by poor CART expansion and persistence, factors that correlate with efficacy following treatment with CD19-CART^10,11. CART-induced regression has been reported in solid tumors ^12–14, however, sustained objective responses are infrequent, owing to poor infiltration or CART dysfunction.^15,16 Recently, remarkable efficacy of GD2-CART against neuroblastoma was reported in patients with low disease burden¹⁷.

Strategies to enhance and sustain the bioavailability of CART against solid tumors including higher dose, host conditioning, and augmentation of cytokines/cytokine-signaling are under investigation^18,19. We aimed to test whether modulation of the homeostatic space could improve HER2-CART kinetics. Given the potential for on-target, off-tumor toxicities observed with other HER2-targeted therapies^20–22, we incrementally incorporated lymphodepletion prior to adoptive T-cell transfer. Here, we report on the safety of HER2-CART after lymphodepletion and the preliminary antitumor activity in patients with progressive HER2-positive sarcoma. We also describe the CART product characteristics, the effect of two lymphodepletion regimens on peripheral blood CART expansion, and the feasibility of repeat infusion cycles in heavily pre-treated patients. When possible, we evaluated tumor samples for immune cell types/markers, CART homing, and target antigen expression to inform future enhancements to our therapeutic approach.

Results

Patient characteristics

From July 1, 2015 to October 18, 2019, we administered HER2-CART (Figure 1a) to 13 patients with progressive sarcoma in 14 enrollments (Table 1). Patient 5 was re-enrolled as Patient 7 at the time of disease recurrence, 17 months following initial study entry²³. Twelve of 14 (86%) patients were <18 years of age at enrollment (median 14 years; range 4–55 years). Five were female, and 9 were male. Eight patients had osteosarcoma, 4 had rhabdomyosarcoma, and 1 each had synovial sarcoma and primitive neuro-ectodermal tumor (PNET). All patients had metastatic disease, documented progression, or recurrence after 1–3 lines of treatment and had active disease at enrollment. Prior therapy included: investigational treatments in eight (57%), metastatectomies (~2–5 surgeries) in five (36%), radiation therapy in seven (50%) and high-dose chemotherapy with autologous stem cell rescue in one patient. The CONSORT chart (Figure 1b) summarizes participant flow through the study. Of 129 patients screened, 49 (38%) had HER2-positive tumors by immunohistochemistry. Nineteen autologous HER2-CART products were successfully manufactured, of which 14 (74%) were infused. Figure 2a describes study evaluations and treatment. Seven patients received multiple CART infusions (range: 1–7 infusions), with or without lymphodepletion, for a total of 21 infusions with lymphodepletion. All treated participants (n=14, including Patient 5 re-enrolled as Patient 7) were evaluable for safety and tumor response analyses, with none lost to follow-up.

Figure 1. HER2-CAR structure and consort flow chart.

(a) Schema showing the components of the 2^nd-generation HER2-CAR and retroviral vector. (b) Consort flow chart summarizing the number of patients screened, enrolled, treated, and followed on the study. *One patient was reassessed for eligibility at recurrence and re-enrolled on the study.

Table1.

Characteristics of all patients treated on the study

Patient ID (UPN)	Age (years) /Sex*	Diagnosis	Prior Treatment				Time from diagnosis to enrollment (months)
			Chemotherapy	Surgery	XRT	Investigational
1	4/F	Synovial sarcoma, Metastatic	(1) Ifosfamide, Doxorubicin	Limb salvage; amputation	No	(1) Gemcitabine, Pazopanib	33.8
2	55/M	Osteosarcoma, Metastatic	(1) Doxorubicin, Cisplatin (2) IE (3) Cisplatin	Primary en bloc	Yes	(1) Gemcitabine, Docetaxel (2) VZV/GD2 CART	11.3
3	14/M	Osteosarcoma, Metastatic	(1) MAP (2) IE/CE	Limb salvage; lung metastatectomy (x2); amputation; paraspinal mass resection	No		37.5
4	16/F	Osteosarcoma, Metastatic	(1) MAPIE	Limb salvage; amputation	No	(1) Pazopanib	36..4
5**	8/M	Rhabdomyosarcoma, Metastatic	(1) VAC/VDC/IE (2) VC, temsirolimus	Primary en bloc	Yes		15.8
6	16/F	Osteosarcoma, Metastatic	(1) MAPIE (2) CE	Limb salvage; lung metastatectomy (x3)	Yes	(1) Inhaled lipid cisplatin	72.8
7 (previously Patient 5)	9/M	Rhabdomyosarcoma, Metastatic	(1) VAC/VDC/IE (2) VC, temsirolimus	Primary en bloc	Yes	(1) HER2-specific CART	35.3
8	17/M	Rhabdomyosarcoma, Metastatic	(1) VI/VDC/IE/VAC (2) VC/temsirolimus	None	Yes	(1) Tumor antigen specific T cells	23.8
9	9/M	Osteosarcoma, Metastatic	(1) MAP (2) Dinutuximab, sargramostim	Amputation; metastatectomy (x1); CNS mass resection; nephrectomy	Yes (x2)	(1) IE with mifamurtide (2) Sorafenib	19.6
10	10/F	Rhabdomyosarcoma, Metastatic	(1) VAC/VI (2) Trametinib (3) VAC/intrathecal topotecan	None	Yes (x2)		8.1
11	31/F	PNET, Metastatic	(1) VDC/IE (2) Flu/Mel + Autologous HSCT	None	No		13.6
12	12/M	Osteosarcoma, Metastatic	(1) MAP	Limb salvage	No		10.7
13	17/M	Osteosarcoma, Metastatic	(1) MAP	Limb salvage; lung metastatectomy (x2)	No	(1) Lenvatinib + IE	24.5
14	14/M	Osteosarcoma, Metastatic	(1) MAP (2) ICE (3) Trofosfamide, Idarubicin, Etoposide, Sorafenib	Limb salvage; lung metastatectomy (x2)	No		37.0

XRT, Radiation therapy. UPN, Unique Patient Number. M, Male. F, Female. IE, Ifosfamide/Etoposide. MAP, Methotrexate/Doxorubicin/Cisplatin. CE, Cyclophosphamide/Etoposide. VDC, Vincristine/Doxorubicin/Cyclophosphamide. VAC, Vincristine/Adriamycin/Cyclophosphamide. VC, Vincristine/Cyclophosphamide. ICE, Ifosfamide/Carboplatin/Etoposide. VI, Vincristine/Irinotecan. CART, Chimeric antigen receptor T-cells. VZV, Varicella zoster virus. Flu, Fludarabine. Mel, Melphalan. HSCT, Haemopoietic stem cell transplant. CNS, Central nervous system. IT, intrathecal.

^*Sex was determined by patient/guardian self-reporting; F, female. M, male.

^**Patient 5 was re-enrolled at disease recurrence and became Patient 7

Figure 2. Study schema and safety assessment.

(a) Schematic representation of study screening, enrollment, treatment, and follow-up for safety and disease response assessment and other study evaluations. (b) Pro-inflammatory cytokines (IFN-γ, IL-6, IL-2, TNF-α, sIL-2R shown) measured from serum (n=8 patients) after the onset of fever following CART infusion (day +1 to +3) arranged by cytokine release syndrome (CRS) grade. Data shown as individual values. Dotted line represents the upper limit of the reference range derived from healthy donor serum in Cytokine 13 panel. UPN, Unique Patient Number. (c) Coronal view of chest computerized tomography (CT) of Patient 14 demonstrating near complete obstruction of the right middle lobe bronchus and mass effect on the bronchus intermedius and right lower lobe bronchus from the perihilar tumor post-treatment (day +4). (d) Longitudinal assessment of serum human anti-mouse antibody (HAMA) in treated patients (n=13). Dotted line represents the threshold for positive antibody activity.

Safety

Cytokine release syndrome (CRS) was observed in 11 patients (79%) and was of higher grade after Flu/Cy lymphodepletion than Flu alone (Table 2). Fever consistently occurred within 24 hours of CART infusion, and no patients had recurrence of symptoms at the time of peak CART expansion. Of the 12 patients infused with 1×10⁸ total-cells/m² in cohorts A (Flu) and B (Flu/Cy), 5 developed grade 1 and 4 developed grade 2 CRS. The two patients in cohort C, Patients 13 and 14, received 1×10⁸ CAR-positive cells/m² after Flu/Cy and developed grade 3 and grade 4 CRS with respiratory involvement, respectively. These adverse events (AEs) were considered dose limiting toxicities (DLT), and enrollment to cohort C ceased. Therefore, we determined 1×10⁸ total cells/m² infused after Flu/Cy to be the maximum tolerated dose (MTD).

Table 2.

Clinicopathological findings at study enrollment and clinical outcomes

Dose Level	Patient ID (UPN)	Diagnosis	Disease at Enrollment	Maximum CRS Grade	Best Response	Time to Progression (months from study treatment initiation)	Survival (months from study treatment initiation)	Vital Status
Cohort A Flu + 1×108 total T cells/m2	1	STS	B/L lung nodules (~5×8 mm)	0	PD	1.7	14.2	DOD
	2	OS	R skull base (3.9×2.9×6.9 cm)	1	PD	1.6	4.6	DOD
	3	OS	Distal pancreas (6.7×6.6 cm)	2	SD	4.4	8.2	DOD
Cohort B Flu/Cy + 1×108 total T cells/m2	4	OS	B/L lung nodules (~ 3 mm)	1	SD	9.0	14.9	DOD
	5*	RMS	Bone marrow	1	CR	18.0	71.7	Alive
	6	OS	Indeterminate CR - positive surgical margins after resection of a pleural-based lung mass (~10 cm)	0	CR	42.4	65.9	Alive
	7 (previously Patient 05)	RMS	Bone marrow	1	CR	-	-	Alive
	8	RMS	Bone, mediastinal, hilar and juxta aortic/caval adenopathy	0	SD	3.7	17.4	DOD
	9	OS	L parietal enhancement, B/L lung nodules (~11 mm), Rt iliac bone (1.4 cm)	1	PD	1.6	7.1	DOD
	10	RMS	Diaphragm (9 mm), B/L pelvic bone & T7 (ill defined), bone marrow	2	PD	1.9	8.5	DOD
	11	PNET	Liver (up to 1.3 cm), pancreatic (2.9 cm), pelvic (3.4 cm), ovary (2.9 cm)	2	SD	3.4	6.1	DOD
	12	OS	Multiple B lung nodules (up to 4 mm)	2	PD	1.5	30	Alive
Cohort C Flu/Cy + 1×108 CAR-positive T cells/m2	13	OS	Multiple R lung nodules (up to 6 mm)	3	PD	1.5	3.8	DOD
	14	OS	Multiple B lung nodules (up to 2.5 cm), perihilar lymphadenopathy (4.8 cm)	4	PD	1.2	1.5	DOD

UPN, Unique Patient Number. B/L, bilateral. R, right. L, left. Flu, fludarabine. Flu/Cy, fludarabine/cyclophosphamide. CRS, cytokine release syndrome. STS, soft tissue sarcoma. OS, osteosarcoma. RMS, rhabdomyosarcoma. PNET, primitive neuroectodermal tumor. PD, progressive disease. SD, stable disease. CR, complete response. DOD, died of disease.

^*Patient 5 was re-enrolled at disease recurrence and became Patient 7

Clinical measurement of serum cytokines performed in a subset of patients (n=8) with CRS showed consistently elevated sIL-2R and variably elevated IL-6, IFN-γ, IL-2, and TNF (Figure 2b) Supportive care resulted in recovery in all without lasting organ dysfunction. Patient 13 received respiratory support with CPAP for one day. Patient 14 experienced respiratory failure (grade 4 CRS) and was found to have near complete obstruction of bronchi adjacent to metastatic tumor on imaging (Figure 2c); direct bronchoscopy confirmed severe mass effect. Methylprednisolone treatment led to respiratory improvement over the following three days and discontinuation of mechanical ventilation.

Upon retrospective analysis, there were no neurological AEs suggestive of immune effector cell-associated neurotoxicity syndrome (ICANS)²⁴. Post-treatment echocardiograms done 6-weeks after each HER2-CART infusion demonstrated no change in cardiac function from baseline. We detected low baseline human anti-mouse antibody (HAMA) positivity in Patients 6 and 9 without a clinically relevant increase in titer by 6-weeks post-infusion (Figure 2d); none developed a new HAMA response to the murine HER2-CAR-scFv despite multiple infusions.

We observed cytopenias (Extended Data Figure 1a to 1d) secondary to chemotherapy in all; absolute lymphocyte count on day 0 was significantly lower in patients receiving Flu/Cy compared to Flu alone (p<0.0001, two-tailed t test), but the pattern of lymphocyte recovery was similar. All but one patient treated with Flu/Cy developed grade 4 neutropenia (p<0.0001, two-tailed t test), with a median of 14 days (range: 7–28 days) to recovery. No patients developed infections or required growth factor support. Baseline grade 2 thrombocytopenia in Patient 11 progressed to grade 3 after Flu/Cy. All AEs considered at least possibly study-related are listed in Extended Data Table 1. AEs attributable to the investigational agent were not observed during long-term follow-up.

Cellular kinetics

The total cell dose infused ranged from 1×10⁸/m² to 1.4 × 10⁸/m², and the median “CAR-positive” T-cell dose was 0.81×10⁸/m² (range: 0.62×10⁸/m² to 1×10⁸/m²; Extended Data Table 2). CART expansion was observed after all but two of 21 total infusions given after lymphodepletion. In serum samples obtained pre- and post-CART, we observed a variable trend in homeostatic cytokines after Flu and Flu/Cy (Figure 3a; Extended Data Figure 1e). Following the first T-cell infusion, we detected a 2–3 log increase in HER2-CAR transgene levels in the peripheral blood by qPCR (Figure 3b), with peak expansion most often observed at week one. We found a significant increase in area under the curve (AUC) following the first infusion upon comparing cohort A (n=3) and cohort B (n=9) with historical controls (n=7) who received the same total T-cell dose without lymphodepletion⁸ (Figure 3c; p=0.006, two-way ANOVA). Mean log₂ AUCs were significantly different between cohort B (Flu/Cy) and no lymphodepletion (p=0.002; two-way ANOVA) but not between cohort A (Flu) and no lymphodepletion (Figure 3d; p=0.114; two-way ANOVA). We confirmed these findings using nonlinear mixed-effects modeling of the cellular kinetics which estimated that the maximum CAR transgene levels in patients who received lymphodepletion was approximately 3.5-fold higher than patients who did not (Extended Data Figure 2). In the Visual Predictive Check of the final model, the 5^th, 50^th and 95^th percentiles of the prediction-corrected observations were in close agreement with the confidence intervals of the 5^th, 50^th, and 95^th percentiles of the prediction-corrected simulated data, indicating a robust predictive ability of the model to describe CAR-transgene concentrations.

Figure 3. Homeostatic cytokines and HER2-CART cellular kinetics.

(a) Heatmaps demonstrating fold change in serum IL-15 and IL-7 levels during the first week compared to pre-CART infusion (day 0) baseline in treated patients (n=13). UPN, Unique Patient Number. (b) HER2-CART levels in the peripheral blood after the first infusion given with lymphodepletion, determined by quantitative polymerase chain reaction (qPCR) for HER2-CAR transgene detection. Flu, fludarabine. Flu/Cy, fludarabine and cyclophosphamide. Data are shown as individual values from independent samples analyzed overtime from total 14 patients. Solid lines represent the mean. (c) HER2-CART detected in peripheral blood following infusion of 1×10⁸ cells/m² after Flu in cohort A (n=3) and after Flu/Cy in cohort B (n=9), in comparison to the same dose of T-cells infused without lymphodepletion (LD; n=7) in a previous cohort of patients (historical controls). (d) Areas under the curve (AUC) were log-transformed to stabilize variances and analyzed using two-sided ANOVA with linear contrasts. Box plot (upper and lower quartiles with horizontal line at median) showing the difference in mean log₂ AUC for 6 weeks post-CART between patients in cohort A (n=3), and historical (no LD; n=7) controls (p=0.114), cohort B (n=9) and historical (no LD) controls (**p=0.002), cohort A and cohort B (p=0.264). Dots represent individual values, and diamond represents the mean. (e) Comparison of AUC between first and second treatment cycles in 5 patients receiving multiple cycles of LD and CART infusion (*p=0.029, paired two-tailed t-test).

Despite their initial expansion, HER2-CART in peripheral blood declined substantially in most patients by week-6. Five patients received repeat cycles of lymphodepletion and HER2-CART (Extended Data Figure 3a), with a significant increase in AUC in the second treatment cycle (p=0.029; paired t-test; Figure 3e). Peripheral blood CART levels in relation to serum pro-inflammatory cytokines measured by multiplex assay are shown in Extended Data Figure 3b. HER2-CAR transgene was detected at the tumor site(s) in six of seven patients who underwent clinically indicated surgical intervention (Extended Data Figure 4a). At 4 weeks post-CART, a metastatic lung nodule from Patient 12 and a malignant pleural effusion from Patient 13 contained HER2-CART (114 and 907 transgene copies per μg DNA, respectively). In Patient 7 (previously Patient 5), we detected HER2-CAR transgene in a right lung lesion resected 16 months after the last CART infusion (299 copies/μg DNA or 0.0029 copies/cell) that showed benign lymphoid tissue on histopathology (Extended Data Figure 4b).

Antitumor Activity

Among the 14 enrollments, we observed a best response of complete response (CR) in 3 (21%) and stable disease (SD) in 4 (29%). Seven (50%) had progressive disease (PD) by the end of the 6-week disease evaluation period (Table 2). Patient 5, a boy with metastatic rhabdomyosarcoma, achieved CR after repeat CART infusions²³. He re-enrolled (Patient 7) 19 months after the initial enrollment due to disease recurrence 6 months off-therapy, and the second CR is ongoing at over six years of follow-up. The second patient with CR, Patient 6, was a 16-year-old girl with osteosarcoma who had experienced four recurrences over 2 years. She had undergone multiple pulmonary metastatectomies, including resection of a 10 cm pleural-based metastasis with positive surgical margins eight weeks prior to study enrollment. She remained without tumor progression following a single infusion of HER2-CART after Flu/Cy. A persistent left lung nodule was resected 13 months after the CART, but no malignant cells were identified (Figure 4a), and CD3 immunostain was positive in the inflammatory infiltrate surrounding the fibrotic nodule with probable treated/necrotic tumor. Tumor remission persisted for 42 months until relapse in the right lung adjacent to the prior resection site with HER2-negative tumor (Figure 4b). Four other patients had stable disease lasting up to 8.7 months (range: 3.1–8.7 months; Figure 4c), and 7 had disease progression. In cohort C, both patients had rapid disease progression. Four weeks after complete recovery from CRS, Patient 13 developed sudden right-sided chest pain, shortness of breath, and an acute drop in hemoglobin; chest imaging showed significant interval tumor growth, a large right hemorrhagic pleural effusion, and multiple new bilateral lung nodules. Patient 14 had extensive progression of metastatic disease involving bilateral lung parenchyma and perihilar/mediastinal lymph nodes at 4 weeks post-infusion. As a post hoc analysis, median overall survival (OS) for all patients was 8.2 months (n=13, 95% CI, 4.4–17 months; Figure 4d), and median progression-free survival (PFS) was 2.4 months (n=14, 95% CI: 1.3–8.7 months; Figure 4e), at a median follow-up of 8.2 months (range: 1.3–70.6 months).

Figure 4. Histopathological findings, clinical responses, and survival outcomes.

Post-treatment hematoxylin and eosin (H&E)-stained biopsy of the left lung nodule in Patient 6 showing (a) dense central sclerosis with inflammatory infiltrates in the periphery (yellow arrows) and adjacent lung parenchyma (left panel) and subpleural nodule with sclerosis, inflammatory cells, and cholesterol clefts (right panel) without viable tumor identified. (b) Recurrent osteosarcoma lung metastasis in Patient 6 showing absence of HER2 expression (left panel) by immunohistochemistry (repeated independently with similar results) and positivity for vimentin (right panel). Vimentin staining was used as a control to account for the potential antigen loss during tissue processing. Panels (a) and (b) show representative microscopic images; scale bar 100 μm. (c) Swimmer plot showing disease status and treatment response for all study patients. *Denotes time point when Patient 5 is re-enrolled as a new patient (Patient 7). Kaplan-Meier estimate of (d) overall survival (OS; n=13) and (e) progression-free survival (PFS; n=14) for all study patients from the time of first HER2-CART infusion. For OS analysis, Patient 5 who was re-enrolled as Patient 7 at disease recurrence was counted according to the time from first study treatment only. UPN, unique patient number. SD, stable disease. PD, progressive disease. CR, complete response. LD, lymphodepletion.

Tumor HER2 expression and immune cell profiling

The grade and intensity of tumor HER2-expression assessed at eligibility screening (Extended Data Figure 5a) were tabulated in relation to the best response achieved (Extended Data Table 3). Following HER2-CART infusion, we examined tumor samples from a subset of patients in whom adequate tissue was available from a clinically indicated surgery and observed absence of HER2 in one (Patient 6; Figure 4b), present but decreased HER2 grade/intensity in tumor samples from Patient 4 (Extended Data Figure 5b) and Patient 5/7²³, and potential antigen loss during tissue processing in Patient 12 (Extended Data Figures 5c to 5f).

We characterized the tumor microenvironment (TME) for spatially resolved immune-related protein expression (Supplementary Table 1) in tumor samples available (n=12, 8 osteosarcoma, 3 rhabdomyosarcoma, 1 PNET) using GeoMx^® Digital Spatial Profiler (DSP). Guided by hematoxylin & eosin (H&E; Extended Data Figure 6a), we examined 187 different tumor regions of interest (ROI; average~15/slide) from 9 pre-treatment (5 primary, 4 metastases) and 3 post-treatment (metastases) biopsies and observed spatial heterogeneity in abundance of immune cell markers by sarcoma type (CD45, CD8, CD4, CD56, PD-1, CTLA4; Figure 5a to 5c; Extended Data Figure 6b to 6e), CRS grade (CD20, CD68; Figure 5d, 5e), and treatment response (CD68; Figure 5f). Principal component analysis (PCA) of 18 proteins from pre- and post-treatment samples revealed clustering by diagnosis and by best response achieved (Extended Data Figure 6f). Hierarchical clustering was used to analyze the expression of immune markers in pre-treatment samples, and contrasts were performed by best treatment response (CR vs. PD) and by CRS grade (grade ≤1 vs. grade ≥2). Analysis of all ROIs (n=143) generated 3 main clusters (Extended Data Figure 6g), demonstrating similarities in expression of CD3, CD4, Beta-2-microglobulin and granzyme B in those with CR. Upon analysis of pre-defined tumor compartments, TME-1 (95 ROIs) exhibit high CD68 and low granzyme B in PD in contrast to low CD68 and high granzyme B in CR (Figure 5g). When compared by CRS grade, 3 major patterns were noted in TME-1 based on the expression of fibronectin, CD11c, CD68, CD56, and granzyme B (Extended Data Figure 6h). In TME-2, 4 clusters were observed, with tumors from patients achieving CR showing high CD4, CD11c, and granzyme B expression (Figure 5h). Comparison by timepoint revealed upregulation of HLA-DR in TME-2 from post-treatment samples (Figure 5i), independent of the treatment response observed.

Figure 5. Spatial profiling of immune markers in tumor microenvironment (TME).

Total immune cell (CD45⁺) infiltrates (a), cytotoxic (CD8⁺) T-cells (b), and helper (CD4⁺) T-cells (c) in pre-treatment tumors from 9 patients grouped by sarcoma type. Protein expression shown as Signal-to-Noise Ratio (SNR). (d) Expression of B cell marker CD20 in pre-treatment samples by cytokine release syndrome (CRS) grade. Monocyte-macrophage marker CD68 by CRS grade (e) and by best response (f). In panels (a) to (f), individual data point represents a distinct region of interest (ROI) in a tumor sample (n=9 patients). UPN, unique patient number. OS, osteosarcoma. RMS, rhabdomyosarcoma. PNET, primitive neuroectodermal tumor. Gr, grade. CR, complete response. SD, stable disease. PD, progressive disease. Hierarchical clustering of immune cell markers by best treatment response in TME-1 (g) and TME-2 (h). **contrasts were performed by comparing CR vs. PD. (i) Hierarchical clustering of immune-related protein expression in TME-2 from pre- and post-treatment samples. **feature used for performing contrast comparison by two-sided t-tests with the Benjamini-Hochberg FDR (BH-FDR) adjustment for multiple comparisons.

HER2-CART product characteristics

The average product manufacturing time was 10 days (range: 8–14 days); there were no manufacturing failures. Infused products (n=14), assessed by flow cytometry prior to cryopreservation, had a median CAR transduction efficiency of 78% (range:61–95%; Extended Data Figure 7a) and showed robust HER2-specific cytotoxicity (Extended Data Figure 7b). Cellular products were ≥98% CD3⁺ lymphocytes, containing both cytotoxic (CD8⁺; median 51.3%, range:28.1%−72.7%) and helper (CD4⁺; median 40%, range:21.5%−60.8%) subsets of effector memory (TEM; median 43.8%, range:3%−70%) and central memory (TCM; median 18.4%, range:4.8%−69.1%; Figure 6a; Supplementary Table 2) phenotypes. The total duration of ex vivo T-cell expansion showed a negative and positive linear correlation with the proportion of CD3⁺CD45RA⁺ (r = −0.688, 95%CI −0.892 to −0.248, p=0.0065; Pearson’s correlation) and CD3⁺CD45RO⁺ subsets (r = 0.648, 95%CI 0.179 to 0.877, p=0.012; Pearson’s correlation), respectively (Extended Data Figure 7c, 7d). Despite inter-patient variability in product composition and phenotype, there were no marked differences among sarcoma subtypes (Figure 6b; p>0.05, two-tailed t-test). When analyzed by clinical response, CART products contained a similar proportion of central and effector memory T-cell pools (Figure 6c, p>0.05, two-tailed t-test). CD3⁺CD8⁺ subsets had a lower frequency of terminally differentiated CD45RA⁺CD27⁻ effector T-cells in CART products from patients achieving CR (Figure 6d; p=0.022, two-tailed t-test).

Figure 6. HER2-CART product characteristics.

(a) Heatmap shows the CART product (n=14) immunophenotype assessed prior to cryopreservation using flow cytometry. UPN, unique patient number. (b) CART product composition in CD8⁺ and CD4⁺ T-cell subsets shown by sarcoma type. OS, osteosarcoma. RMS, rhabdomyosarcoma. T_N/T_SCM, naïve/stem cell memory (CD45RO⁺CD62L⁺); T_CM, central memory (CD45RO⁺CD62L⁺ ); T_EM, effector memory (CD45RO⁺CCR7⁻CD62L⁻). Data are shown as individual values for autologous CART products (n=14). Error bars represent the median with 95% CI. two-tailed Mann-Whitney test, ns p >0.05. (c) Floating bars (min to max) showing CD8⁺ and CD4⁺ T-cell immunophenotype in CART products from patients achieving complete response (CR; n=3)) compared to others (n=11). SD, stable disease. PD, progressive disease. Lines represent the median. Two-tailed Mann-Whitney test, ns, p >0.05. (d) Proportion of CD8⁺CD45RA⁺CD27⁻ effector T-cells in CART products shown by best treatment response (*p=0.022, two-tailed Mann-Whitney test). Data are shown as individual values for each product tested (n=14). Lines represent median. (e) Single cell analysis of cryopreserved CART products (n=9) using mass cytometry (CyTOF) showing % of CD8⁺PD-1⁺ (*p=0.018, two-tailed unpaired t-test) and CD4⁺PD-1⁺ (**p=0.004, two-tailed unpaired t-test) T-cells from patients with osteosarcoma (n=4), in comparison to rhabdomyosarcoma (n=4) and primitive neuroectodermal tumor (not compared as n=1). Data are shown as individual values. Lines represent median. (f) Single-cell analysis (CyTOF) of cryopreserved CART products (n=9) showing the co-expression of immune checkpoint receptors in CD8⁺ T-cell subsets: % of CD8⁺PD-1⁺TIM-3⁺ (p=0.261, two-tailed Mann-Whitney test), % of CD8⁺PD-1⁺LAG-3⁺ (p=0.095, two-tailed Mann-Whitney test), and % of CD8⁺PD-1⁺CD39⁺ (p=0.166, two-tailed Mann-Whitney test) in patients with CR (n=3) vs. patients with SD/PD (n=6). Data are shown as individual values in box plots (min to max) with a horizontal line at the median.

Single-cell analysis of unstimulated CART products (n=9) using mass cytometry demonstrated a lower percentage of CD8⁺PD-1⁺ (p=0.018, two-tailed t-test) and CD4⁺PD-1⁺ (p=0.004, two-tailed t-test) T-cells in products from patients with RMS compared to osteosarcoma (Figure 6e; Extended Data Figure 7e). The percentage of CD8⁺ T-cells that were PD-1⁺TIM-3⁺, PD-1⁺LAG-3⁺, or PD-1⁺CD39⁺ was comparable among patients achieving CR, while there was a larger variation amongst patients with SD/PD (Figure 6f; p>0.05, two-tailed t-test).

Discussion

In this study, patients with sarcoma safely tolerated up to 1×10⁸ total T-cells/m² (~0.8×10⁸ CAR-positive T-cells/m²) of HER2-CART infused after Flu/Cy. Among these heavily pre-treated patients, we observed disease stabilization or remission in 7 of 14 (50%), with 66.6% (6 of 9) having clinical benefit in cohort B. The 1-year OS was 46% with a median follow-up 8.2 months for all patients and 5.4 years for surviving patients; one patient remains in long-term remission.

Sarcomas are a diverse group of malignancies of mesenchymal origin. Although their natural history differs by subtype, the overall outcomes of patients with metastatic/recurrent sarcoma are poor^25,26. Despite their biologic and antigenic heterogeneity which poses a considerable challenge to systematic testing of new therapies, several TAAs including HER2, NY-ESO-1, GD2, B7-H3, GPC3, and EGFR are being targeted in clinical studies^8,23,27–29. Though sarcoma TAA expression data is limited ^29,30, the HER2-positivity rate in this study (~38%) aligns with existing literature^4,31. HER2-negativity rate amongst other factors contributed to the high attrition, a major constraint in early trials of targeted drugs/biologics^32,33.

We previously reported the safety of HER2-CART alone in lymphoreplete patients with sarcoma⁸. In the cohorts reported here, early-onset CRS occurred with HER2-CART given after lymphodepletion but was manageable with supportive care. The two patients with osteosarcoma treated in cohort C developed DLT with grade 3–4 CRS; both patients recovered from the severe respiratory AEs but died from progressive pulmonary metastatic disease shortly afterwards, raising the possibility that rapidly progressing disease contributed to this toxicity. Although others did not experience severe toxicity, the high rate of lung metastases in this population and the inflammation seen in some merits further investigation to improve patient selection and interventions. CART products, particularly those utilizing a CD28 co-stimulatory domain, have been associated with ICANS^34,35, which was not observed in this study. Other toxicities observed with HER2-directed therapies including decreased cardiac function^36–38 and gastrointestinal bleeding²² were not seen, though differences in patient population, the short persistence of CART in the peripheral blood, and the FRP5 rather than trastuzumab-derived scFv all limit this comparison. The long-term safety data in our study is limited to a few surviving patients.

We found that Flu/Cy conditioning in cohort B significantly improved HER2-CART AUCs compared to historical controls. The effect of lymphodepletion on CART expansion kinetics was endorsed by the nonlinear mixed-effects modeling approach. Though some patients had clinical benefit, the peak expansion achieved was approximately one log-fold lower than cellular kinetics of other clinically successful CART products^12,17,39. At the highest dose level evaluated (cohort C), we observed protocol-defined DLTs early after CART infusion, which precluded cellular kinetics analysis.

We show that employing iterative treatment cycles is an option to improve the HER2-CART bioavailability. Despite the intensity of the Flu/Cy regimen used⁴⁰, repeat cycles were well tolerated, with no evident non-hematological toxicities. Since most of the study participants were children and adolescents, this finding should be extrapolated with caution to older patients. Among patients receiving repeat CART doses, no AEs related to the potential immunogenicity of the CAR were observed.

Detection of the CAR transgene in post-treatment tissue samples and the durable clinical benefit in some patients suggests wide distribution of CART and homing to tumor sites. However, the relationship between CART frequency/activity at disease sites and the peripheral blood level is unknown. The decline in peripheral blood CART levels is likely multifactorial, including rapid contraction of the homeostatic space with endogenous lymphocyte recovery and inadequate signal 1 (antigen-mediated), signal 2 (co-stimulation), and signal 3 (cytokine) to sustain CART proliferation and persistence. Although secretion of cytokines or co-expression of cytokine receptors could improve CART longevity⁴¹, the safety of this approach across targets is unknown but could eliminate the need for lymphodepletion or repeat T-cell infusions. The probability of disease-free survival for patients with recurrent osteosarcoma remains very low ²⁵. Most patients with osteosarcoma on this study had prognostic factor(s) predictive of inferior outcomes including metastatic disease at diagnosis, poor histologic response to therapy, disease recurrence, and extrapulmonary metastasis^25,42–44. Similarly, patients with other refractory/recurrent metastatic sarcomas had poor prognosis²⁶. Therefore, the clinical benefit observed in some provides promising preliminary evidence for HER2-CART bioactivity. However, this phase I basket study was not designed to establish efficacy. CRs in this study were achieved in patients with relatively low disease burden prior to CART infusion. While differences in sarcoma biology make comparisons of extent of disease among patients difficult, our observation is consistent with a recent report of GD2-CART for neuroblastoma wherein patients with lower disease burden had a longer survival¹⁷. Disease recurred during off-therapy follow-up in two patients with CR, underscoring the importance of studying post-treatment/recurrent tumors to sustain therapeutic benefit.

Target antigen downregulation/loss can lead to relapse after CD19-CART ^45,46. In solid tumors, this phenomenon is described but is not consistent across tumor types or target antigens^13,17,33. In this cohort, we found lower tumor HER2 expression in two patients post-treatment (UPN4, UPN5)²³ and complete absence in the osteosarcoma recurrence in Patient 6. The TME and its impact on CART treatment outcomes is of considerable interest in solid tumors^15,16,47. In a comprehensive study of pediatric sarcoma, the TME immune signature in pre-treatment primary tumors was an independent predictor of prognosis⁴⁸. Although differential protein expression analysis did not show statistically significant differences in our relatively small, heterogenous cohort, the TME immune-marker landscape correlated with prior reports⁴⁸. In pre-treatment samples, we found a higher abundance of markers associated with antigen presentation and T-cell cytotoxicity in responding tumors. While enrichment of CD68, a monocyte-macrophage marker, was associated with PD and more severe CRS across sarcoma histology, its expression on osteoclasts may confound these results. Tumor samples showed evidence of immune activation relative to pre-treatment tumors. Due to the small sample size, diversity of sarcoma subtypes, and variability in biopsy timing, these observations are descriptive, and prospective evaluation in a larger disease-specific cohort of patients is needed. Incorporating surgical cohorts into future trials may enable robust profiling of tumor antigens and immune infiltrates.

Intrinsic immune deficits and the cumulative effect of prior chemotherapy affect T-cell quality and CART functionality⁴⁹. While CART product parameters are correlated with clinical outcomes in lymphoid malignancies^50–53, this knowledge is limited in patients with solid tumors^12,15,54. In this study, despite differences in prior therapy and timing of procurement, only the duration of ex vivo expansion correlated with the number of naïve/memory/effector T-cells contained in the final CART product. There were no notable variations in the product immunophenotype when compared by diagnosis. A discrete subset of CD8⁺ cytotoxic T-cells, characterized by the loss of CD27 and re-expression of CD45RA, are thought to be functionally distinct terminally differentiated effector T-cells with low proliferative capacity⁵⁵. We observed that the CART products from patients with CR contained a very small number of CD3⁺CD8⁺CD45RA⁺CD27⁻ T-cells relative to others. Pre-clinical and early clinical research suggest susceptibility of CART to immune-inhibitory signals^23,56. We found a lower proportion of CD4⁺/CD8⁺ T-cells with PD-1 expression in products from patients with RMS and a trend towards a lower CD8⁺ T-cell percentage co-expressing PD-1 with TIM-3, LAG-3, or CD39 in products of patients who attained CR. While definitive conclusions cannot be drawn, these observations are informative and should be examined in relation to toxicity and clinical response in a larger number of patients to guide improvements to cellular product quality and to inform clinical testing of combination immunotherapies⁵².

In conclusion, this phase I study demonstrates the tolerability of autologous HER2-CART after lymphodepletion in patients with advanced sarcoma. The preliminary antitumor activity observed warrants further testing of HER2-CART and measures to enhance their therapeutic index and clinical impact.

Methods

Study design and treatment

This phase I, open-label study (NCT00902044) was conducted at Texas Children’s Hospital (TCH) and Houston Methodist Hospital (HMH). The protocol was approved by the Institutional Biosafety Committee (IBC) and the Institutional Review Board (IRB) of Baylor College of Medicine (BCM) for TCH, the IRB of HMH, the U.S. FDA, and the Recombinant DNA Advisory Committee of the NIH and was conducted in accordance with the ethical principles in Declaration of Helsinki. All thirteen patients/guardians provided written informed consent/assent. The Data Safety Monitoring Committee of the Dan L. Duncan Comprehensive Cancer Center at BCM provided continued study oversight.

Tumor HER2 expression was determined using immunohistochemistry (ab8054 [CB11], Abcam Inc, Cambridge, MA) as previously described^5,8,31. Adult and pediatric patients, both male and female, with HER2-expressing sarcoma (at least grade 1 proportion of tumor cells [1–25%] and intensity of staining score ≥1+) were eligible for study enrollment if they had refractory/recurrent disease following standard first-line treatment. We previously established the safety of monotherapy with autologous HER2-CART at doses up to 1×10⁸ total T-cells/m² (historical controls for comparison of cellular kinetics)⁸ prior to introducing progressive intensification of lymphodepletion. In the lymphodepletion cohort, the first study patient was enrolled on March 18, 2015 and the last patient on September 25, 2019. Patients in cohort A were conditioned with fludarabine (Flu; 25 mg/m²/dose x 5 days) alone, and patients in cohorts B and C received Flu (25 mg/m²/dosex5 days) and cyclophosphamide (Cy; 30 mg/kg/dosex2 days). Treatment consisted of 1×10⁸ total T-cells/m² after Flu in cohort A, 1×10⁸ total T-cells/m² after Flu and Cy in cohort B, and 1×10⁸ CAR-positive T-cells/m² after Flu and Cy in cohort C. Patients who tolerated prior cycle(s) without symptomatic disease progression were eligible to receive additional infusions of HER2-CART, including up to two more with lymphodepletion.

Safety and clinical response assessment

The primary objective was to evaluate the safety of one infusion of HER2-CART. Safety assessments were done in all patients who received at least one dose of HER2-CART. CRS was graded according to published guidelines⁵⁷. All other AEs, including neurologic toxicities, were graded using the National Cancer Institute Common Terminology Criteria for AEs (CTCAE v4). DLT was defined as any Grade ≥3 toxicity primarily related to the HER2-CART occurring within 6-week period following the first CART infusion. All patients are followed up to 15 years or until off-study (death or another reason) for potential toxicities from retroviral gene transfer. We evaluated disease status using imaging (or histopathology, when applicable) within 4 weeks of initiating study treatment and 6 weeks after HER2-CART infusion. Antitumor response, defined as CR (disappearance of all target lesions), PR (at least a 30% decrease in the sum of the longest diameter of target lesions), PD (at least a 20% increase in the sum of the longest diameter of target lesions), or SD (neither sufficient shrinkage to qualify for PR not sufficient increase to qualify for PD) post-treatment, was the secondary objective and was determined by comparing pre- and post-infusion disease assessment using Response Evaluation Criteria in Solid Tumors (RECIST v1.1) in patients with measurable disease⁵⁸.

Analysis of HER2-CART products

Autologous HER2-CART were manufactured in the current Good Manufacturing Practice (cGMP) facility at TCH. CART were generated from peripheral blood mononuclear cells (PBMCs), collected by a peripheral blood draw⁸, and expanded ex vivo in IL-7 and IL-15. Manufactured cell products were evaluated for purity (CD3⁺/CD19^-), HER2-CAR expression, HLA-identity, viability, sterility, and target-specific cytotoxicity prior to issuance of the Certificate of Analysis. Immunophenotyping of HER2-CART products (n=14) was done prior to cryopreservation at the TCH Center for Cell and Gene Therapy CLIA- and CAP-certified flow cytometry facility according to the institutional standard operating procedure (SOP) for testing all immune effector cells manufactured for clinical use. The flow cytometry panels (Supplementary Table 3) utilized custom-prepared lyophilized monoclonal antibodies (Lot#3165640, BD Biosciences, San Jose, CA), per manufacturer’s instructions without modification, for direct immunofluorescent staining to decrease variability of results in longitudinal studies.⁵⁹ For measuring HER2-CAR expression, autologous CART products and non-transduced T-cells generated in parallel were stained using Alexa Fluor^® 647 AffiniPure Goat Anti-Mouse IgG, F(ab’)2 fragment specific antibody (reconstituted, 1:100 to 1:800 dilution, 0.5–1 μL/100 μL test, polyclonal, Catalogue#115–605-006, Lot#122475, The Jackson Laboratory, Bar Harbor, ME), and at least 50,000 cells were analyzed by flow cytometry on an FACS Canto II flow cytometer (BD Biosciences) using FACSDiva™ software v8.0.1 (BD Biosciences). FlowJo software v10 (Ashland, OR) was used for additional data analysis.

Single cell analysis of cryopreserved CART products (n=14) was performed by mass cytometry (CyTOF)^60,61 at the Flow Cytometry & Cellular Imaging Core Facility at M. D. Anderson Cancer Center using an antibody panel for immune cell markers optimized for research use (Supplementary Table 4). T-cell samples were prepared per manufacturer’s protocol (Fluidigm, San Francisco, CA; now Standard BioTools). The instrument was tuned daily with Tuning Solution (Fluidigm) to meet manufacturer specifications to ensure quality control. For batch corrections, each sample was spiked with 10% EQ Four Element Calibration Beads (Fluidigm), and samples were normalized for intensity differences post-acquisition using CyTOF software version 6.7.1016. Mass cytometry data was analyzed using Cytobank v9.1 (Beckman Coulter, IN)⁶².

HER2-specific cytotoxicity of CART products was tested in a 4-hour chromium-51 release assay using HER2^positive tumor targets: NCI-H1299 (Cat # CRL-5803, ATCC, Manassas, VA) and LM7 (Sa-OS derivative; a gift from Eugenie S. Kleinerman)⁶³, and HER2^negative tumor targets: K562 (Cat # CCL-243, ATCC) and MDA-MB-468 (Cat # HTB-132, ATCC), with autologous non-transduced T cells as a negative control.

Immune monitoring

We evaluated patient serum for the presence of HAMA (IgG) using quantitative ELISA (Alpha Diagnostics International, San Antonio, TX). Serum cytokines were quantified longitudinally using the Luminex^® Multiplex assay (xPONENT, Millipore, Austin, TX). In patients with CRS, serum cytokines were measured in real time by a clinical laboratory test (Cytokine Panel 13, Serum; ARUP Laboratories, Salt Lake City, UT) that uses quantitative multiplex bead assay.

Assessment of HER2-CART kinetics

The HER2-CAR transgene was detected in peripheral blood and tumor tissue using quantitative polymerase chain reaction (qPCR; ABI 7900HT Sequence Detection System; Applied Biosystems)²³. The DNA was extracted (QIAamp DNA Blood Mini Kit, Qiagen, Germany) using FRP5-specific primers (forward primer: 5’-CCACGGTCACCGTTTCCT-3’ 18bp, reverse primer: 5’-GGGTCAGCTGGATGTCAGAAC-3’ 21bp, probe [on reverse strand]: 5’-FAM-CCGCCACCAGAACCG-NFQ-3’ 15bp) and TaqMan probe (Applied Biosystems, Waltham, MA) and qPCR was performed in triplicate using the ABI 7900HT Sequence Detection System (Applied Biosystems)²³.

Cellular kinetics after lymphodepletion were compared to a historical cohort of patients with HER2⁺ sarcoma who received HER2-CART (1×10⁸ total T-cells/m²) without lymphodepletion⁸. AUCs were calculated using the trapezoidal rule for HER2-CART over time for each patient. AUCs were log-transformed to stabilize variances and analyzed using ANOVA with linear contrasts. AUCs for subsequent infusions were compared using paired t-test.

Non-linear mixed effects modeling of cellular kinetics

Nonlinear, mixed-effects modeling was performed in Monolix 2023R1, Lixoft SAS, to assess the impact of lymphodepletion on the CART expansion kinetics. We developed a semi-mechanistic cellular kinetic model which was modified from a previously published model⁶⁴. The final model was a two-compartment model where effector CART are allowed to differentiate in an irreversible way to memory CART controlled by the rate constant r_M. Both effector and memory CART undergo natural turnover at rate constants; δE and δM. To assist with model identifiability, δE was fixed to 0.435 mon-1 per previous reports^64,65. No tissue distribution was integrated in this model due to the sampling scheme. CART in patients who underwent lymphodepletion proliferated at a rate constant ρ for duration T1 to maximum transgene level Cmax. T1 was estimated visually by examining each CART profile to facilitate estimation of the other parameters. During this duration, turnover and memory differentiation was assumed negligible. Contraction at a rate constant α took place after proliferation. The primary focus of this analysis was to quantify the differences in peak transgene levels in patients who underwent lymphodepletion compared with patients who did not undergo such a procedure. Inter-individual variability in model parameters was modelled using an exponential error model. Residual variability was modelled using a combined additive and proportional error model. Model selection was based on evaluation of the goodness of fit plots. The predictive performance of the final model and its usefulness for describing observations was assessed using prediction and variance-corrected visual predictive checks (VPCs), where the final parameter estimates were used to simulate 1,000 replicates of the observed dataset. Both observations and the simulated data were normalized on the typical model prediction for the median independent variable in each bin in order to account for variation in sampling times and predictive covariates introduced by binning of the observations. The median and the 5^th and 95^th percentile concentrations of the simulated datasets were then plotted against the original observations.

Digital spatial profiling of tumor microenvironment

Immune profiling was done on formalin-fixed paraffin-embedded (FFPE) tumors from 10 individual patients (n=12 samples: 9 pre-treatment, 3 post-treatment) using GeoMx^® Digital Spatial Profiler (DSP; NanoString Technologies Inc. Seattle, WA). For multiplex protein profiling of immune markers (Supplementary Table 1), 4 μm thick FFPE tissue slides were prepared following NanoString protocols without modification (Manual 10150–01 Manual Slide Preparation)⁶⁶. Morphological staining (SYTO13, a nuclear stain) and protein detection reagents with UV-photocleavable oligonucleotide containing a unique molecular identifier (UMI) using standard immunohistochemistry protocols were employed. The tissue section was imaged to generate a whole tissue image at single cell resolution. ROIs (600 μm x 600 μm) were identified based on morphology staining and the H&E staining comparison; 192 distinct ROIs were created in strategic locations. TME segments were pre-defined by the presence of viable tumor and density of immune infiltrates as TME-1 (viable tumor with immune infiltrates) and TME-2 (viable and necrotic areas with variable immune infiltrates). Protein probes were sequentially “collected” by photocleaving the oligonucleotides from the tissue and collected in separate wells of a 96 well plate. Collected probes were barcoded, processed, and counted on the NanoString nCounter system (Manual 10089–08 nCounter Readout without modification). Count files were subsequently indexed to the ROI on the tissue scans for analysis.

Statistics and reproducibility

This phase I study was primarily designed to evaluate safety, and thus, most other findings are observational and descriptive. A 3+3 design was used to guide dose-escalation and to determine the MTD of HER2-CART after lymphodepletion. Toxicity outcomes, frequencies of HER2-CART, and cytokine analyses were summarized using descriptive statistics. OS and PFS were calculated from time of first CART infusion to death from any cause or to last follow-up, and to disease progression or death from any cause, respectively. Post-hoc analyses of OS and PFS were done using the Kaplan-Meier method. No patients were excluded from analysis. One patient in cohort B was re-enrolled at disease recurrence to allow for repeat infusion of CART with lymphodepletion; in alignment with the study design, both enrollments are described independently for safety analysis, response, and PFS calculations, and only the first enrollment for OS calculations.

Data were generated using biologically distinct samples from patients, employing technical replicates as indicated. Standard clinical guidelines were followed for imaging, histopathology, and laboratory assessments. When feasible, H&E and HER2 immunohistochemistry on patient samples were independently validated. The sample size for correlative analyses was determined by tissue availability. Experiments were not randomized, were conducted using appropriate positive and negative controls, and, when applicable, validated using donor samples for reproducibility. For data analysis, GraphPad Prism 9.0, Rv4.1.0, SAS 9.4, or Microsoft Excel 2019 were used. Investigators were not blinded. All statistical tests are two-sided unless otherwise specified. P values of <0.05 were considered significant. Adobe Illustrator^® was used to make pictorial images and font editing.

Data from nCounter were passed through robust quality control (QC) procedure to examine standard metrics including ROI nuclei count, surface area, binding density, and positive norm factor. Five ROIs that failed the positive norm factor QC were removed. The remaining data were normalized by signal-to-noise ratio (SNR) per vendor’s standard protocol. Proteins of SNR ≥4.0 in at least three ROIs were included for statistical comparisons. Proteins differentially abundant between groups of designation were detected by linear mixed-effects models (LMM) based on restricted maximum likelihood (REML) in lme4 (v1.1.27.1), with slide ID (unique for each tumor sample) as random effect and group of interest as fixed effect (formula: protein~0+Feature+Diagnosis+(1|Slide, RMEL=TRUE)⁶⁷. Contrasts of group protein abundance differences were performed by two-sided t-tests with the Benjamini-Hochberg FDR (BH-FDR) adjustment for multiple comparisons. Statistical analysis was performed using Rv4.1.1 and Bioconductor (release 3.13). Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Extended Data

Extended Data Figure 1: Study-related cytopenia and pre-infusion serum cytokines.

(a) Absolute lymphocyte count (ALC) on day 0 fludarabine (Flu; n=3 patients) and Flu and cyclophosphamide (Flu/Cy; n=11 patients) lymphodepletion (Flu vs. Flu/Cy, ***p<0.0001, two-tailed unpaired t-test). Error bars represent Mean with SD. (b) Trend in ALC in patients receiving Flu (n=3) compared to Flu/Cy (n=11) during the first 6 weeks after CART infusion. Data are shown as individual values for treated patients. Solid lines represent mean values overtime. D, day. Wk, week. (c) Absolute neutrophil count (ANC) nadir by lymphodepletion received (Flu vs. Flu/Cy, ***p<0.0001, two-tailed unpaired t-test). (d) Time to recovery from severe neutropenia in Flu/Cy group (median time: 14 days, range: 7 to 28 days, **p=0.009, two-tailed unpaired t-test) from CART infusion. Error bars represent Median with range. In panels (c) and (d) data are shown as individual values for Flu (n=3) and Flu/Cy (n=11) groups. (e) Heatmap of serum cytokine concentrations after lymphodepletion and prior to first CAR T-cell infusion (day 0) in patients conditioned with fludarabine (Flu; n=3) or Flu and cyclophosphamide (Flu/Cy; n=11).

Extended Data Figure 2: Nonlinear mixed-effects modeling of cellular kinetics.

(a) and (b) show model fit to individual cellular kinetic profiles. UPN, unique patient number. HC, historical control. (c) Visual Predictive Check of the final cellular kinetic model. Lines represent the 5^th, 50^th and 95^th percentiles of the prediction-corrected observations. The blue bands represent 5^th and 95^th percentiles, and the pink band represents 50^th percentiles of the prediction-corrected simulated data. (d) Final cellular kinetic model parameter estimates.

Extended Data Figure 3: CART kinetics and pro-inflammatory cytokines in peripheral blood.

(a) HER2-CART levels in peripheral blood measured using quantitative polymerase chain reaction (qPCR) after repeat CART infusions given with lymphodepletion. (b) Serum pro-inflammatory cytokine levels (right y-axis) during the first week after the CART infusion, plotted in relation to CART copy numbers (left y-axis) detected in peripheral blood. UPN, unique patient number. Pre, pre-infusion. Hr, hour. Wk, week.

Extended Data Figure 4: Analysis of post-treatment tumor tissue.

(a) HER2-CAR transgene detection at tumor site(s) post-treatment. (b) Hematoxylin and eosin (H&E) staining of right lung biopsy tissue in Patient 5/7 at 16 months off therapy showing benign perivascular lymphoid aggregate (yellow arrow) adjacent to muscularized vessel with intimal hyperplasia (black arrow) markedly obscuring the vessel (left panel; 200X) and area of organizing pneumonia (white arrow) obscuring airway and alveolar architecture of lung (right panel; 200X). Representative microscopic images shown; scale bar 100 μm.

Extended Data Figure 5: Tumor HER2 expression.

(a) Tumor HER2 expression by immunohistochemistry (IHC) prior to study enrollment. (b) HER2 IHC of lung nodule from Patient 4 resected at 5.9 months post first CART infusion. Left lung nodule from Patient 12 resected at 6 weeks post-CART showing (c) viable tumor cells intermixed with osteoid and extensive angiolymphatic invasion on hematoxylin and eosin (H&E) staining, and no detectable HER2 (c) and vimentin (d) on IHC (independently validated by repeat testing). (f) HER2 expression in pre-treatment tumor sample confirmed by repeat IHC, done in parallel with post-treatment tumor tissue. Panels show representative microscopic images; scale bar 20 μm.

Extended Data Figure 6: Spatial profiling of immune markers in tumor microenvironment (TME).

(a) Representative hematoxylin and eosin (H&E) staining showing viable tumor cells and immune infiltrates (TME-1) from different sarcoma histology evaluated; scale bar 200 μm. (b) Representative bubble plots from pre-treatment tumor samples showing heterogenous expression of immune cell markers (CD68 and CD11C shown). Blue denotes DNA staining. Squares represent region of interest (ROIs) selected on GeoMx^® Digital Spatial Profiler. Bubbles represent the density of immune-related protein expression within the corresponding ROI. Expression of NK cell marker CD56 (c), and immune checkpoints PD-1 (d) and CTLA4 (e) in pre-treatment tumors from 9 patients grouped by sarcoma type. Individual data points represent a distinct ROI in a tumor sample. Box plots in (d) and (e) show min to max with horizontal line at the median. Protein expression shown as Signal-to-Noise Ratio (SNR). (f) Principal Component Analysis (PCA) showing immune-related protein expression in pre- and post-treatment samples by diagnosis and by best treatment response achieved. (g) Hierarchical clustering of immune cell markers in all ROIs from pre-treatment samples.** contrast was performed comparing complete response (CR) vs. progressive disease (PD). (h) Hierarchical clustering of immune cell markers in TME-1 by cytokine release syndrome (CRS) grade. Contrast was performed in (g) and (h) using two-sided t -tests with the Benjamini-Hochberg FDR (BH-FDR) adjustment for multiple comparisons. In UPN, unique patient number. OS, osteosarcoma. RMS, rhabdomyosarcoma. PNET, Primitive Neuroectodermal Tumor. CR, complete response. SD, stable disease. PD, progressive disease.

Extended Data Figure 7: HER2-CART product characteristics and gating strategy for mass cytometry analysis.

(a) Transduction efficiency of HER2-CART products manufactured for all study patients (n=14). Data shown as individual values. Horizontal lines represent the median. NT, non-transduced T-cells. (b) 4-hour chromium release assay demonstrating HER2-specific cytotoxic function of infused CART products (n=14 patients). LM7 and NCI-H1299 tumor cell lines expressed HER2. K562 and MDA-MB-468 tumor cell lines were used as negative controls. Data are shown as individual values for autologous CART products tested. Error bars represent the mean+/−SD. Correlation between duration of ex vivo expansion to the proportion of (c) CD3⁺CD45RA⁺ (r = −0.688, 95% CI −0.892 to −0.248, p=0.0065; Pearson’s correlation) and (d) CD3⁺CD45RO⁺ (r = 0.648, 95% CI 0.179 to 0.877, p=0.012; Pearson’s correlation) cells in the final CART products prior to cryopreservation. (e) Mass cytometry (CyTOF; Fluidigm) was performed on cryopreserved CART products and corresponding NT T-cell samples for a subset of patients (n=9). Pre-conjugated metal-tagged antibodies were purchased from Fluidigm. Calibration beads (Fluidigm) were added to all samples. Data were analyzed using Cytobank v9.1 (Beckman Coulter, IN). viSNE high-dimensionality reduction analysis was performed using all 36 metal-tagged antibody parameters. Gating was done to select for intact singlets and to exclude CD19⁺ and CD56⁺ cells. The resultant analysis on T-cells with z-axis coloration for CD4⁺ (left upper panel) and CD8 (right upper panel), respectively, is shown for a representative CART product sample. Further analysis on T-cell subsets was performed to examine co-expression of activation and exhaustion markers on CD8⁺PD-1⁺ T cells (lower panel). First, CD8⁺ T-cells were gated on PD-1⁺ subsets. These were then analyzed for co-expression of TIM-3, LAG-3, and CD39, respectively.

Extended Data Table 1:

Adverse events within the first 6 weeks post HER2-CART infusion

Adverse Event	Cohort A (n=3)		Cohort B (n=9)		Cohort C (n=2)
	Grade 1–2	Grade 3–4	Grade 1–2	Grade 3–4	Grade 1–2	Grade 3–4
Hematological
Anemia	0* (0)**	0 (0)	5 (0)	3 (0)	0 (0)	0 (0)
Neutropenia	1 (0)	0 (0)	0 (0)	9 (0)	0 (0)	2 (0)
Lymphopenia	0 (0)	3 (0)	0 (0)	9 (0)	0 (0)	2 (0)
Thrombocytopenia	1 (0)	0 (0)	4 (0)	1 (0)	1 (0)	0 (0)
Leukopenia	2 (0)	1 (0)	0 (0)	9 (0)	0 (0)	2 (0)
Respiratory
Cough	0 ( 0)	0 ( 0)	1 ( 1)	0 ( 0)	1 ( 1)	0 ( 0)
Dyspnea	0 (0)	0 (0)	0 (0)	1 (1)	1 (1)	1 (1)
Tachypnea	0 (0)	0 (0)	1 (1)	0 (0)	2 (2)	0 (0)
Hypoxia	1 (1)	0 (0)	1 (1)	0 (0)	1 (1)	1 (1)
Respiratory failure	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	1 (1)
Hypercapnia	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	1 (1)
Cardiovascular
Sinus Tachycardia	1 ( 1)	0 ( 0)	1 ( 1)	0 ( 0)	1 ( 1)	1 ( 1)
Hypertension	0 (0)	0 (0)	1 (1)	0 (0)	0 (0)	0 (0)
Hypotension	0 (0)	1 (1)	2 (2)	0 (0)	0 (0)	1 (1)
General
Chills	0 (0)	0 (0)	1 (1)	0 (0)	2 (2)	0 (0)
Fever	2 (2)	0 (0)	7 (7)	0 (0)	1 (1)	1 (1)
Fatigue	1 (1)	0 (0)	3 (1)	0 (0)	2 (2)	0 (0)
Malaise	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
Non-cardiac chest pain	0 (0)	0 (0)	2 (2)	0 (0)	0 (0)	0 (0)
Pain	1 (1)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
Back pain	0 (0)	0 (0)	0 (0)	0 (0)	1 (1)	0 (0)
Myalgia	1 (1)	0 (0)	0 (0)	0 (0)	0 (0)	0 (0)
Anorexia	0 (0)	0 (0)	2 (0)	0 (0)	0 (0)	0 (0)
Generalized muscle weakness	0 (0)	0 (0)	0 (0)	0 (0)	1 (0)	0 (0)
Gastrointestinal
Nausea	2 (0)	0 (0)	5 ( 1)	0 (0)	2 ( 1)	0 (0)
Vomiting	1 (0)	0 (0)	5 (0)	0 (0)	1 (0)	0 (0)
Mucositis, oral	0 (0)	0 (0)	1 (0)	0 (0)	0 (0)	0 (0)
Neurological
Headache	0 (0)	0 (0)	1 (0)	0 (0)	0 (0)	0 (0)
Dizziness	0 (0)	0 (0)	2 (1)	0 (0)	0 (0)	0 (0)
Tremor	0 (0)	0 (0)	1 (1)	0 (0)	0 (0)	0 (0)
Presyncope	0 (0)	0 (0)	0 (0)	0 (0)	1 (0)	0 (0)
Confusion	0 (0)	0 (0)	0 (0)	0 (0)	1 (0)	0 (0)
Metabolic, Renal and Others
Elevated creatinine	0 (0)	0 (0)	1 (0)	0 (0)	0 (0)	0 (0)
Electrolyte abnormalities
Hypermagnesemia	0 (0)	0 (0)	1 (0)	0 (0)	1 (0)	0 (0)
Hypomagnesemia	0 (0)	0 (0)	1 (0)	0 (0)	1 (0)	0 (0)
Hypocalcemia	0 (0)	0 (0)	1 (0)	0 (0)	2 (0)	0 (0)
Hypokalemia	0 (0)	0 (0)	3 (0)	0 (0)	1 (0)	0 (0)
Hyponatremia	0 (0)	0 (0)	4 (0)	0 (0)	1 (0)	0 (0)
Increased aspartate aminotransferase	0 (0)	0 (0)	4 (1)	0 (0)	2 (1)	0 (0)
Increased alanine aminotransferase	1 (0)	0 (0)	1 (0)	0 (0)	0 (0)	0 (0)
Cytokine Release Syndrome (CRS)	2 (2)	0 (0)	7 (7)	0 (0)	0 (0)	2 (2)

^*Study related adverse events

^**( )Adverse events related to the investigational agent

Extended Data Table 2:

CAR transduction efficiency in products and the cell dose infused

Patient ID	HER2-CAR transduction#		Total cell dose infused (per m2)	Total CAR+ cells infused (per m2)	Best tumor response
	CART product (%)	Non-transduced (%)
UPN1	73.1	0.5	1 × 108	7.31 × 107	PD
UPN2	88.2	2	1 × 108	8.82 × 107	PD
UPN3	88.8	0.4	1 × 108	8.88 × 107	SD
UPN4	66.4	3.4	1 × 108	6.64 × 107	SD
UPN5*	77.6	1	1 × 108	7.76 × 107	CR
UPN6	61.9	1.2	1 × 108	6.19 × 107	CR
UPN7*	78	7.87	1 × 108	7.80 × 107	CR
UPN8	79	5.56	1 × 108	7.90 × 107	SD
UPN9	82.6	0.55	1 × 108	8.26 × 107	PD
UPN10	95.6	0.26	1 × 108	9.56 × 107	PD
UPN11	85.2	0	1 × 108	8.52 × 107	SD
UPN12	70.6	1.48	1 × 108	7.06 × 107	PD
UPN13	84	0.1	1.19 × 108	1 × 108	PD
UPN14	71.2	0.29	1.40 × 108	1 × 108	PD

^*Patient 5 (UPN5) was re-enrolled as a new patient (UPN7) at the time of disease recurrence

^#CAR expression assessed by flow cytometry prior to cryopreservation

PD, progressive disease. SD, stable disease. CR, complete response.

Extended Data Table 3:

HER2 grade and intensity in relation to treatment response

	Best Response (n=14)
	Complete Response (n=3)	Stable Disease (n=4)	Progressive Disease (n=7)
HER2 grade, n(%)
1	0 (0)	0 (0)	1 (100)
2	1 (25)	1 (25)	2 (50)
3	2 (40)	1 (20)	2 (40)
4	0 (0)	2 (50)	2 (50)
HER2 intensity, n (%)
2	1 (16.7)	2 (33.3)	3 (50)
3	2 (25)	2 (25)	4 (50)
HER2 total score # , n (%)
4	1 (20)	1 (20)	3 (60)
6	2 (28.6)	2 (28.6)	3 (42.9)
7	0 (0)	1 (50)	1 (50)

^#grade + intensity

Data Availability

The source data for Figures 1–6 and Extended Data Figures 1-7 are provided as Source Data files. All other data supporting the findings of this study are available within the article and supplementary materials. Additional information may be requested from the corresponding authors if in alignment with the study consent and de-identifiable to protect research participant privacy.