Surface CD81 supports leukemia stem cell function and reveals a therapeutic vulnerability in acute myeloid leukemia

Abstract

Relapse remains the leading cause of mortality in acute myeloid leukemia (AML), largely due to the persistence of therapy-resistant leukemia stem cells (LSCs). However, surface determinants that sustain LSC function and disease aggressiveness remain incompletely defined. Here, we identify the tetraspanin CD81 as a regulator of LSC function, progression and treatment resistance in AML. Analysis of retrospective patient cohorts revealed that high CD81 surface expression is associated with relapse and adverse clinical outcomes in non-core-binding factor AML. Functional studies demonstrated that elevated CD81 expression promotes chemoresistance and enhances leukemic engraftment in immunodeficient mouse models. In vivo gain- and loss-of-function approaches further established that CD81 drives increased leukemia burden and aggressive disease behavior. Notably, CD81 was enriched within LSC-containing subpopulations, where its expression supported LSC maintenance and resistance to chemotherapy. Mechanistically, CD81 promotes chemoresistance and leukemic aggressiveness through pathways linked to LAPTM4B-mediated STAT3 signaling and enhanced adhesion-dependent cellular interactions. These effects were accompanied by increased migration, invasion, and formation of filopodia-like membrane protrusions. Importantly, therapeutic immunotargeting of CD81 significantly reduced leukemic burden while exhibiting a manageable toxicity profile in preclinical models. Collectively, these findings establish CD81 as a clinically relevant surface marker associated with AML relapse and identify CD81-dependent signaling as a therapeutic vulnerability for targeting LSCs and preventing disease recurrence.

Introduction

Acute myeloid leukemia (AML) is the most common acute leukemia in adults. Treatment typically involves intensive combination chemotherapy, usually consisting of cytarabine and an anthracycline, often followed by hematopoietic stem cell transplantation (HSCT).¹ Recently, various targeted therapies have become available for specific patients,² but despite this progress, mortality in AML patients remains high and poor clinical outcome is mainly due to treatment failure and disease relapse, occurring in up to 40% of younger and 80% of older adult AML cases.^3–5 Even though several relapse risk factors are known,^6,7 a better understanding of the mechanisms leading to refractory or relapsed AML is urgently needed to develop more potent therapeutic strategies.

In this regard, the major cause of disease progression and relapse are therapy-resistant and quiescent LSC,⁸ and strategies to eradicate LSC have recently become the focus of intensive research.⁹ LSC are present within the leukemia bulk at a very low frequency and functionally defined as capable to initiate AML when serially transplanted into immunocompromised mice.¹⁰ While various distinct cell surface proteins on LSC are either enriched such as CD33,¹¹ CD123,¹² CD44,¹³ CD47,^14,15 TIM-3,¹⁶ CLL-1^17,18 and CD45RA¹⁹ or absent such as CD90,²⁰ therapies targeting these markers are lagging behind expectations. This is mainly due to their overlapping expression with non-malignant hematopoietic stem and progenitor cells (HSC) as well as the lack of antigens broadly and consistently expressed on LSC in most AML subtypes. Therefore, the identification of novel druggable and specific LSC markers remains important for the successful development of selective anti-LSC therapies.

CD81 refers to a cell surface protein, also known as “target of an antiproliferative antibody 1” (TAPA-1),²¹ that belongs to the evolutionary conserved tetraspanin superfamily of proteins. CD81 is one of the structural component of specialized membrane microdomains known as tetraspanin-enriched microdomains,^22–24 and plays an important role in signal transduction by interacting with various proteins such as integrins and several distinct membrane proteins.^25,26 In particular, CD81 has been shown to regulate B cell function by interacting with CD19 and CD21 on B-lymphocytes^27,28 and is known to be overexpressed in several hematological malignancies.^21,29,30 Importantly, immunotargeting of CD81 represents a new effective treatment for B cell lymphoma.^29,31 Nevertheless, despite its adverse role in B cell malignancies, its functional implication in AML remains to be characterized.

In this study, we have applied a combination of in vitro and in vivo approaches to demonstrate that CD81 is an important determinant of AML progression, drug resistance and leukemia stemness independent of its expression in lymphocytes. Finally, we provide proof of concept that targeting CD81 may represent a novel and efficacious therapeutic strategy to eradicate LSC in AML.

Results

CD81 overexpression is associated with AML relapse and poor clinical outcome

Previously, we have associated high CD81 cell surface expression at diagnosis with poor prognosis in a cohort of 134 de novo AML.³² These initial findings are now validated and elaborated using an extended cohort of 252 diagnostic and 38 relapsed AML samples (CONSORT diagram Supplementary Fig. 1). Similarly, we detected a bimodal distribution of CD81 surface expression on blasts in this larger cohort (Supplementary Methods and Fig. 2; Fig. 1a). In this study, we show that core-binding factor (CBF) AML (n = 26), which have well-known favorable prognosis,³ have lower CD81 cell surface expression compared to non-CBF AML (Fig. 1b, Supplementary Fig. 3a). In order to independently validate this finding, we applied the positive correlation found in our cohort between CD81 mRNA and cell surface abundance determined by flow cytometry (FCM) (Fig. 1c) on two independent cohorts of AML. Consistent with our results, CBF-AML have significant lower CD81 mRNA expression compared to non-CBF AML shown for the Leucegene cohort³³ and the Beat AML dataset³⁴ (Fig. 1d). Despite the exclusion of good prognosis CBF-AML and that CD81 expression was not different among other major AML subtypes (Supplementary Fig. 4), high diagnostic CD81 mRNA expression adversely affected overall survival (OS) in non-CBF AML and in NPM1-mutant AML using the Beat AML dataset (Fig. 1e; non-CBF AML: HR [CI] = 1.74 [0.97–3.12]; NPM1-mutant AML: HR [CI] = 4.18 [1.42–12.36]).^35,36 Of note, AML with NPM1 mutation showed lower CD81 expression, although this did not reach statistical significance; therefore for this subset only, we applied a lower cut-off to define high CD81 in this study.

Fig. 2.

High CD81 membrane expression in AML models enhances chemoresistance and engraftment in immunodeficient mice. a Shown are daunorubicin (DNR, left) and cytarabine (AraC, right) drug-response curves and bar charts of relative IC₅₀ using AML models with either high CD81 (red, CD81⁺) or low CD81 expression (blue, CD81^-) vs. respective controls in U-937 [top, gain-of-function], OCI-AML3 [middle, loss-of-function], HNT-34 [bottom, loss-of function] cells evaluated by in vitro drug resistance assay of at least three independent experiments run in duplicates. b Line graphs of luminescence and in vivo images of xenografts (Supplementary Fig. 7) demonstrating increased tumor burden of U-937 CD81-overexpressing cells (CD81⁺, red, n = 17) compared to controls (WT CD81-, blue, n = 17). Conversely, CD81 silencing (CD81 shRNA, blue, n = 11) in OCI-AML3 cells decreased tumor burden compared to controls (control shRNA, red, n = 11). c AML bone marrow (BM) homing measured in BM of NSG mice at 72 h after IV AML cell injection by FCM analysis, indicating increased homing when CD81 was overexpressed (U-937, n = 3) whereas it decreased when CD81 was silenced (OCI-AML3, n = 5). WT: wild type, *P < 0.05, **P < 0.005, ***P < 0.0005

Fig. 1.

High CD81 surface expression associates with relapse and adverse outcome in non-CBF AML. a Shown are representative flow cytometry histograms of CD81 surface expression measured in AML bone marrow taken at diagnosis with blasts (green) and corresponding hematogones (purple). The top graphs illustrate CD81 Low (n = 3) and the bottom graphs CD81 High AML (n = 3). The violin plot in the right panel summarizes low vs. high CD81 in the total AML study group (n = 290). b Violin plot comparing CD81 cell surface expression in CBF vs. non-CBF AML (n = 26, n = 226) and c scatterplot correlating CD81 z-score normalized mRNA vs. CD81 surface expression in the study cohort (n = 47, r² = 0.51). d Violin plots of CD81 z-score normalized mRNA expression in non-CBF vs. CBF-AML in the Leucegene project (n = 15, n = 85, left panel)³³ and the Beat AML dataset (n = 29, n = 335, right panel).³⁴ e Kaplan–Meier curves of Beat AML patients stratified by median CD81 mRNA expression at AML diagnosis, showing worse OS for patients with AML overexpressing CD81 (top panel, n = 83) and within NPM1-mutant AML (bottom panel, n = 33). f Kaplan–Meier curves showing OS and RFS (top and bottom panel) for AML patients stratified by median CD81 surface expression at diagnosis in non-CBF AML (n = 150) and g within NPM1-mutant AML patients stratified by low 25% CD81 surface expression (n = 54). h Violin plot of CD81 surface expression in normal bone marrow (n = 16), de novo AML (n = 252) and AML at relapse (n = 38). i Line graph of corresponding paired diagnostic-relapse AML samples (n = 19; red: increase, blue: no change, green: decrease; paired t-test). *P < 0.05, **P < 0.005, ***P < 0.0005

Importantly, high median CD81 cell surface expression at diagnosis adversely affected not only overall survival (OS) but also relapse-free survival (RFS) in non-CBF AML patients within our study cohort (Table 1, Fig. 1f; n = 150, OS: HR [CI] = 1.88 [1.07–3.32]; RFS: HR [CI] = 1.92 [1.09–3.38]), which retained prognostic impact in a multivariate analysis, which included age, diagnostic white blood cell count (WBC) and ELN2022 risk (Table 2). In addition, we also show that high CD81 surface expression at diagnosis negatively affects OS within a larger cohort of NPM1-mutant AML,³² a molecular subtype with favorable clinical response (OS: HR [CI] = 7.71 [1.00–59.2]; Fig. 1g). Notably, multivariate analysis indicates that this effect is independent of co-occurring FLT3-ITD mutations known to attenuate the favorable effect of the NPM1 mutation (n = 54, Table 3) and a similar trend was obtained for RFS (Fig. 1f; RFS: HR [CI] = 1.71 [0.81–4.12]). Finally, given the prognostic effect of CD81 surface expression at diagnosis, we hypothesize that naïve AML blasts exhibit increased expression of this membrane protein compared to non-malignant bone marrow (BM) and that CD81 expression would be even higher at relapse compared to diagnosis, a disease state frequently refractory to treatment. As expected, CD81 surface protein expression is higher in AML compared to mononucleated BM cells (BMNC) obtained from healthy individuals (44 vs. 35%) and AML at relapse exhibited even higher CD81 membrane expression when compared to diagnostic samples (71 vs. 44%, Fig. 1h, Supplementary Fig. 3b). Specifically, when analyzing paired diagnostic-relapsed samples available for 19 patients, we confirmed progressive enrichment of high CD81 surface expression (67 vs. 20%, Fig. 1i).

Table 1.

Diagnostic parameters of the study cohort treated with standard chemotherapy

Parameter	Total	CD81 low	CD81 high	P-value
No. of patients (%)	150 (100)	68 (45)	82 (55)	–
Age [y], median (min-max)	55.6 (17.3-77.6)	53.6 (17.3-77.6)	56.3 (17.8-75.7)	0.71
Gender (M/F)	72/78	31/37	41/41	1.0
WBC [×109/L], median (min-max)	27.2 (1-405)	13.4 (1-249)	46.1 (1-405)	0.010*
BM blast [%], median (min-max)	70 (5-98)	61.5 (5-97)	72 (5-98)	0.114
Cytogenetics				0.86
Favorable, n (%)	–	–	–
Normal, n (%)	85 (56.7)	41 (60.3)	44 (53.6)
Abnormal, n (%)	35 (23.3)	13 (19.1)	22 (26.8)
Complex, n (%)	26 (17.3)	12 (17.6)	14 (17.1)
N/A, n (%)	4 (2.7)	2 (3)	2 (2.5)
ELN2022 risk				0.61
Adverse, n (%)	27 (17.3)	13 (19.1)	14 (17.1)
Favorable, n (%)	26 (17.3)	15 (22.1)	11 (13.4)
Intermediate, n (%)	93 (62.7)	38 (55.9)	55 (67.1)
N/A, n (%)	4 (2.7)	2 (2.9)	2 (2.4)

^*P < 0.05

N/A not available, WBC white blood cell count, ELN European LeukemiaNet, BM bone marrow

Table 2.

Univariate and multivariate analysis

		Univariate analysis		Multivariate analysis
	Variable	HR	(95% CI)	p-value	HR	(95% CI)	p-value
RFS	CD81 high	1.92	1.09–3.38	0.025*	1.56	0.83–2.93	0.0489*
	Age >60	2.18	1.22–3.88	0.0085*	2.37	1.29–4.35	0.0065*
	Log(WBC)	1.07	0.87–1.31	0.53	2.63	1.29–5.36	0.28
ELN2022 risk	Intermediate	–	–	–	–	–	–
	Adverse	1.52	0.57–4.02	0.40	1.75	0.65–4.68	0.31
	Favorable	0.50	0.26–0.99	0.0473*	0.61	0.30–1.22	0.088
	N/A	6.12	1.33–28.3	0.0203*	8.19	1.01–66.3	0.063
OS	CD81 high	1.88	1.07–3.32	0.029*	1.96	1.06–3.62	0.0325*
	Age >60	3.93	2.08–7.40	<0.0001****	3.75	1.95–7.19	0.0001****
	Log(WBC)	1.04	0.86–1.26	0.69	1.10	0.91–1.31	0.31
ELN2022 risk	Intermediate	–	–	–	–	–	–
	Adverse	2.31	1.13–4.72	0.0219*	2.33	1.10–4.91	0.0332*
	Favorable	0.26	0.11–0.63	0.0028**	0.32	0.13–0.79	0.0161*
	N/A	1.57	0.21–11.7	0.66	2.67	0.92–20.8	0.41

^*P < 0.05, **P < 0.005, ***P < 0.0005, ****P ≤ 0.0001

N/A not available, WBC white blood cell count, ELN European LeukemiaNet, RFS relapse-free survival, OS overall survival

Table 3.

Univariate and multivariate analysis within the NPM1-mutant AML subset

		Univariate analysis			Multivariate analysis
	Variable	HR	(95% CI)	p-value	HR	(95% CI)	p-value
RFS	CD81	1.71	0.81–4.12	0.146	1.60	0.65–3.93	0.31
	FLT3-ITD	1.54	0.28–1.51	0.078	1.38	0.30–1.73	0.47
OS	CD81	7.71	1.00–59.2	0.021*	6.58	0.84–51.6	0.073
	FLT3-ITD	2.64	0.65–1.11	0.317	1.98	0.17–1.51	0.22

^*P < 0.05

Collectively, these findings demonstrate that CD81 surface overexpression on AML blasts adversely affects AML outcome and may contribute to the aggressiveness characterizing blasts at relapse.

CD81 promotes leukemia initiation and disease aggressiveness in vivo

To further strengthen the deleterious role of CD81 in AML, we modulated its expression in three human AML cell lines: OCI-AML3, HNT-34 and U-937. Overexpression of CD81 (CD81⁺) was induced in U-937 cells, which originally have undetectable CD81 expression (WT CD81^-), whereas silencing of CD81 was achieved using the shRNA approach in HNT-34 and OCI-AML3 cells, which by contrast highly express CD81 at baseline. CD81 expression of these engineered models have been validated by RT-PCR, immunoblotting and flow cytometry (FCM); and in accordance with primary AML data, CD81 membrane expression and CD81 mRNA expression also correlated in our in vitro models (Supplementary Fig. 5). Since we associated diagnostic CD81 surface overexpression to AML relapse, we first assessed whether modulating CD81 expression levels affected chemoresistance. We found that U-937-CD81⁺ cells were more resistant to daunorubicin, whereas CD81-depleted cells (OCI-AML3, HNT-34) were more sensitive to daunorubicin and cytarabine (Fig. 2a) and this effect was not attributable to reduced proliferation (Supplementary Fig. 6). Then we tested xenoengraftment efficacy in immunocompromised NSG mice, an indicator of AML aggressiveness, of two independent AML models, and evaluated leukemia engraftment over time by in vivo imaging of leukemia cells exhibiting both either high or low CD81 expression.^37,38 As shown in Fig. 2b and Supplementary Fig. 7, xenoengraftment rate was higher in U-937 CD81⁺ compared to WT CD81-negative control cells, whereas CD81-depleted OCI-AML3 cells exhibited reduced xenoengraftment capacity. Consistent with this, AML cell homing evaluated by human AML blast accumulation in murine BM after intravenous (IV) injection was similarly affected by CD81 surface expression level (Fig. 2c).

As AML engraftment implicates multiple cell-based processes such as adhesion, migration and invasion,³⁹ we also examined the contribution of CD81 on each of these cellular programs (Fig. 3a). Of particular interest, CD81 surface overexpression in the U-937 model enhanced AML cell adhesion to fibronectin-coated plates and during co-culture with human and murine bone marrow fibroblasts (HS-5 and MS-5). In contrast, CD81 depletion in both OCI-AML3 and HNT-34 cell models had the opposite effect. Similar results were observed in cellular migration and invasion experiments using a porous membrane and a semisolid gel matrix (Supplementary Fig. 8). Interestingly, confocal microscopy of phalloidin stained AML cells overexpressing CD81 revealed filopodia-like membrane protrusions, a morphological change reminiscent of the cellular processes investigated (Fig. 3b). As cell adhesion is known to influence the cell cycle,⁴⁰ we investigated whether adhesion to fibronectin alters cell cycle distribution in our models. Interestingly, we observed a G2 arrest in fibronectin-adherent compared with non-fibronectin adherent cells in OCI-AML3 models, independent of CD81 status. In contrast, the proportion of cells in S phase did not differ between conditions.

Fig. 3.

CD81 overexpression enhances adhesion, migration, and invasion and promotes filopodia-like membrane protrusions in AML. a CD81 surface overexpression (CD81⁺, red) enhanced cell adhesion (fibronectin, MS-5 and HS-5 fibroblasts), migration and invasion compared to control (U-937, blue, left panels, n > 4). Conversely, CD81 surface depletion (CD81 shRNA, blue) reduced cell adhesion, migration and invasion in both OCI-AML3 and HNT-34 cells (middle and right panels, n > 3) compared to control (Control shRNA, red). b Confocal microscopy displaying filopodia-like cell membrane protrusions in high CD81 surface expressing AML models (top images, U-937: CD81⁺, OCI-AML3: Control shRNA, HNT-34: Control shRNA) compared to their low CD81 counterparts exempt of membrane protrusions (bottom images, U-937: WT CD81^-, OCI-AML3: CD81 shRNA, HNT-34: CD81 shRNA). Bar charts summarize corresponding cell sizes and circularities determined by confocal microscopy⁷⁶ (U-937, n = 3, 218/434 cells; OCI-AML3, n = 2, 295/170 cells; HNT-34, n = 2, 263/173 cells). c Effect of adhesion to fibronectin on the cell cycle. OCI-AML3 cells after 16 h of adhesion on fibronectin or under non-adherent conditions were stained with PI for detection of G1, S and G2 cell cycle alteration by flow cytometry, a G2 arrest in fibronectin-adherent (green) compared with non-fibronectin adherent cells (white) was detected which was independent of the CD81 status (n = 5). The right panel shows a representative DNA content distribution showing the first G1 peak being higher for with non-fibronectin adherent cells (purple) and the second G2 peak being higher for fibronectin adherent cells (green). *P < 0.05, **P < 0.005, ***P < 0.0005

Finally, we evaluated whether higher CD81 membrane expression also affected leukemia aggressiveness by assessing AML infiltration and progression in xenoengrafted mice. As shown in Fig. 4a, mice engrafted with U-937-CD81⁺ cells have higher leukemia burden in peripheral blood (PB) as well as in BM. In addition, CD81 overexpressing cells infiltrate surrounding muscle tissue as shown using hCD45 immunohistochemistry on sternums and tibias of AML xenografts (Fig. 4b). By contrast, CD81-depletion induced a less aggressive phenotype with reduced tissue infiltration (OCI-AML3, Fig. 4d). Importantly, independent survival experiments, demonstrate that the higher CD81 expression gives rise to more aggressive disease with reduced survival of xenoengrafted mice as shown for the U-937 model (HR [CI] = 16.33 [3.35–79.6]; Fig. 4c; Supplementary Fig. 9a), whereas CD81-depletion results in improved overall survival as demonstrated for the OCI-AML3 model (HR [CI] = 0.08 [0.02–0.36]; Fig. 4e; Supplementary Fig. 9b).

Fig. 4.

High CD81 expression in AML drives increased leukemia burden and aggressive disease in vivo. a High CD81 membrane expression (CD81+, red, n = 12) increased leukemia infiltration in BM and PB in U-937 AML xenografts compared to control (WT CD81-, blue, n = 10). b Immmunohistochemical analysis of the sternal bone of U-937 AML xenografts stained with anti-hCD45 antibody indicate that CD81 surface overexpression (CD81+, left panels) induced higher BM and adjacent muscle tissue AML infiltration (green and red area) compared to control (WT CD81-, right panels). c Kaplan–Meier plot illustrating worse survival of mice xenografted with U-937 CD81 overexpressing cells (CD81+, red, n = 7) compared to control (CD81-, blue, n = 6). d In contrast, CD81 surface silencing in OCI-AML3 cells (CD81 shRNA, blue, n = 7) reduced leukemia infiltration in PB and spleen of AML xenografts compared to controls (Control shRNA, red, n = 7) and e prolonged survival as shown by the Kaplan–Meier plot of xenografted mice (CD81 shRNA, blue, n = 6) compared to control (Control shRNA, red, n = 7). Tissue infiltration of xenografts injected with 1.0 ×10⁶ AML cells was evaluated on day 28 (U-937) and day 31 (OCI-AML3) of engraftment (a, b, d), while CDX survival after injection of 0.5 ×10⁶ cells was monitored over an extended period (Supplementary Fig. 9). f Heatmap of differentially expressed genes in primary AML with low (n = 22) vs. high (n = 23) CD81 surface protein expression, based on HG U133 Plus 2.0 array data. AML samples are shown in columns with the color code reflecting CD81 surface expression and genes (probe ID, gene symbol) in rows. CD81 and LAPTM4B transcripts, overexpressed in high CD81 protein AML, are highlighted in green. g Gene Set Enrichment Analysis (GSEA) plot showing top seven gene sets linked to the high vs. low CD81 AML phenotype; top panel: running ES based on rank-ordered list, middle portion: gene members color-coded by gene set and bottom graph: value of the ranking metric. CD81 and LAPTM4B are indicated. h Line graph of CD81 (200675_at) and LAPTM4B (214039_s_at) transcript expression in AML cell models with low vs. high CD81 expression, based on HG U133 Plus 2.0 array data. Low CD81: U-937 WT (solid line) and OCI-AML3 CD81 shRNA (dashed line). High CD81: U-937 CD81+ (solid line) and OCI-AML3 control shRNA (dashed line). i Western blot analysis of activated STAT3 signaling in U-937 and OCI-AML3 high CD81 AML models. Representative blot (top) and summary histogram (bottom) showing phosphorylated STAT3 (pSTAT3) normalized to total STAT3 expression (n = 3). *P < 0.05, **P < 0.005

To gain mechanistic insights, we performed gene expression analysis on a subset of 45 diagnostic non CBF-AML samples. When comparing AML cases with high vs. low CD81 surface protein levels, we confirmed a strong association with CD81 mRNA expression, as expected. Interestingly, this analysis also revealed a positive association with lysosomal protein transmembrane 4 beta (LAPTM4B, Fig. 4f).

Gene Set Enrichment Analysis⁴¹ (GSEA) identified curated gene sets linked to AML and chronic myeloid leukemia. The corresponding network analysis indicated upregulation of genes involved in cancer stem cell programs, motility/adhesion/metastasis, and cell–microenvironment communication (Fig. 4g, Supplementary Fig. 10a). In addition, several Gene Ontology (GO) gene sets were enriched, implicating processes related to tissue and organ formation, vascular biology, cell-surface signaling, cell motility and migration, gene regulation, and intracellular protein organization (Supplementary Fig. 10b).

Notably, in accordance with CD81 surface protein levels, mRNA expression of CD81 and LAPTM4B were also correlated in our in vitro models (Fig. 4h). LAPTM4B is a component of the LSC17 gene signature⁴² and is known to promote poor prognosis in AML through RPS9/STAT3 signaling.⁴³ Consistently, we observed STAT3 activation in the U-937 CD81 gain-of-function model, whereas the OCI-AML3 CD81 loss-of-function model exhibited inhibition of STAT3 activity (Fig. 4i).

Altogether, these results demonstrate that CD81 is a critical determinant of leukemia chemoresistance and aggressiveness, exerts these effects at least in part through the LAPTM4B/STAT3 axis.

CD81 supports leukemia stem cell function in primary AML

As LSC are considered the primary source of relapse,^44–46 and given the increased CD81 expression in blasts at the time of relapse, we investigated whether CD81 also affects LSC fitness. First, we confirmed that HSC (CD34⁺CD38^-CD90⁺CD123^-) are rare in AML and that LSC (CD34⁺CD38^-CD90^-CD123⁺) are absent in normal BM (Supplementary Figs. 2, 11). We then assessed CD81 membrane expression within the HSC, LSC and MPP-like (CD34⁺CD38^-CD90^-CD123^-)¹⁹ compartments. Normal BM progenitors (HSC, MPP-like) display low CD81 expression, in contrast all AML progenitors diagnostic and at relapse express CD81 at high levels (HSC, MPP-like and LSC). Notably, the proportion of CD81⁺ LSCs significantly increased from diagnosis to relapse (63% vs. 46%, Fig. 5a), both across individual samples and in paired diagnostic–relapsed cases (55% vs.11%, Fig. 5b). Furthermore, analysis of single-cell RNA-seq data from the Leucegene project identified CD81 as one of the top genes with higher expression in primitive AML or HSC-like cells compared to non-malignant HSCs.³³ Notably, CD123 and FLT3 are also characteristic of HSC-like AML cells (Fig. 5c, d). These known AML markers are currently of great interest for targeted therapeutic development.³³ Then we investigated whether a high proportion of CD81⁺LSC negatively affects AML outcome. Our results show that AML patients with a higher proportion of CD81⁺LSC at diagnosis are at increased risk of relapse and worse OS (RFS: HR [CI] = 2.44 [1.38–4.32]; OS: HR [CI] = 2.31 [1.32–4.04]; Fig. 5e). Particularly, the proportion of CD81⁺LSC is correlated with bulk CD81 protein expression (Fig. 5f), suggesting that bulk CD81 surface expression reflects and may serve as a surrogate for the proportion of CD81⁺LSC.

Fig. 5.

CD81 overexpression associates with enhanced LSC function, chemoresistance, and poor clinical outcome. a Box plots showing membrane expression of CD81 within hematopoietic progenitor populations defined by flow cytometry (FCM): HSC (CD34⁺CD38⁻CD90⁺CD123⁻), MPP-like (CD34⁺CD38⁻CD90⁻CD123⁻), and LSC (CD34⁺CD38⁻CD90⁻CD123⁺) in normal BM (n = 16, blue), non-CBF AML at diagnosis (n = 150, red open bars), and at relapse (n = 38, red pointed). While an increase in CD81⁺ HSC- and MPP-like progenitor fractions was observed from diagnosis to relapse, only the increase within the LSC fraction reached statistical significance in the global analysis. b Notably, statistical significance was further enhanced when assessed in paired diagnostic and relapsed samples (red: increase, blue: no change, green: decrease; n = 19, paired t-test). c Scatterplot of the Leucegene single-cell RNA-seq data (AML n = 20, normal BM n = 8) showing preferential expression of 16 genes in AML-HSC-like cells compared to normal HSC outside of the 90% prediction interval (dotted line) of the nonlinear regression model. Notably, these genes included CD123, CD81 and FLT3 (highlighted in red). d Boxplot of the Leucegene single-cell RNA-seq data showing higher proportion of CD81, FLT3 and CD123 positive cells in AML compared to normal BM.³³ e Kaplan–Meier plot showing worse RFS and OS in AML patients with high CD81 (red) compared to low CD81 (blue) median surface expression within the LSC fraction at AML diagnosis (n = 150). f Scatterplot correlating CD81 surface expression within the LSC compartment with CD81 protein expression of the leukemia bulk (n = 290; r² = 0.44, P < 0.0001). g Kaplan–Meier curves of RFS and OS for patients stratified by xenoengraftment positivity of primary AML (n = 36), where patients with xenoengrafting AML (red) had worse outcome compared to those whose AML cells did not xenoengraft (blue). h Cytarabine- and i daunorubicin drug-response curves and bar plots of IC₅₀ tested at AML diagnosis indicating drug resistance in xenoengrafting (red, n = 12) vs. not xenoengrafting (blue, n = 16) AML. j High CD81 surface expression is not only associated with successful xenoengraftment (n = 20), but also enhances engraftment efficiency as demonstrated by k time-dependent cumulative incidence of engraftment stratified by CD81 surface expression (n = 40). l Scatter plot of spleen weights reflecting splenic leukemia infiltration in xenografts derived from AML with high vs. low CD81 surface expression (n = 21), illustrated by two representative images of each group. m Scatterplot correlating LSC17 gene score and CD81 membrane expression in non-CBF AML (n = 28, r² = 0.37, P = 0.019). *P < 0.05, **P < 0.005, ***P < 0.0005

To further characterize the relationship between CD81 overexpression and LSC function, we performed serial lymphocyte depleted total AML blast xenoengraftment experiments using immunocompromised NSG mice (Supplementary Figs. 12, 13).⁴⁷ We adopted this strategy to capture both (i) CD34^– LSCs and (ii) more differentiated AML cells with de-differentiation potential both suspected to be able to initiate disease. Notably, CD81 surface expression remains similar when comparing primary AML and respective serial PDX (Supplementary Fig. 14). As expected, patients whose AML xenoengrafted, which is indicative of functional LSC, have worse relapse-free and overall survival than those whose AML failed to engraft (RFS: HR [CI] = 4.45 [1.31–15.16]; OS: HR [CI] = 2.74 [1.15–6.51]; Fig. 5g) and this effect is not due to preferred engraftment of known poor prognosis AML subtypes (Supplementary Fig. 15). Ex vivo drug testing of diagnostic AML blasts⁴⁸ demonstrated higher drug resistance to cytarabine and daunorubicin for AML with positive engrafting capacity (Fig. 5h, i, Supplementary Fig. 16). Moreover, our results indicate that AML with high CD81 membrane expression are not only associated with positive xenoengraftment as a binary variable (Fig. 5j) but also more efficiently engrafted in mice when analyzed in a time-dependent manner (HR [CI] = 4.53 [1.64–12.56]; Fig. 5k Supplementary Fig. 13). Interestingly, patient-derived xenografts (PDX) from AML with high CD81 surface expression have larger spleens at sacrifice than those with low CD81 expression (Fig. 5l), indicative of higher extra-medullary infiltrative potential. Finally, CD81 surface expression is correlated with the LSC17 gene signature score (Fig. 5m), a well-known and robust marker of functional LSC and clinical outcome of AML patients.^42,49

Taken together, these results unambiguously show that high CD81 cell surface expression detected in AML from patients is associated with higher amounts of functional LSC.

Targeting CD81 exposes a therapeutic vulnerability in AML

Having established the deleterious role of CD81 in AML and its translational relevance, we then investigated the therapeutic potential of an anti-hCD81 antibody. Firstly, we evaluated whether in vitro exposure of AML cells with this immunotherapy is able to inhibit their engraftment in NSG mice. Consistent with the data obtained using CD81-depleted cell lines, CD81 antibody-treatment of OCI-AML3 cells effectively reduced their subsequent xenoengraftment potential in BM, PB and their splenic infiltration, the preferential accumulation site of OCI-AML3 cells (Fig. 6a). Additionally, CD81 targeting using this immune-therapeutic approach decreased in vitro resistance of OCI-AML3 to daunorubicin and cytarabine compared to isotype control (Fig. 6b). To evaluate potential hematotoxicity, we exposed healthy donor BMNC to therapeutic concentrations of anti-hCD81 antibody. Notably, antibody-treatment neither affected BMNC viability at 4 and 24 h nor altered BMNC cell cycle distribution when compared to control (Fig. 6c, d). Importantly, no evidence of BMNC colony-forming toxicity was observed in any hematopoietic progenitor population after antibody treatment (Fig. 6e).

Fig. 6.

Efficacy and toxicity of therapeutic CD81 immunotargeting in AML. a Anti-hCD81 antibody-treatment (IgG CD81) reduced OCI-AML3 engraftment capacity in BM, PB and spleen (n > 10), and b sensitized OCI-AML3 cells to chemotherapies: daunorubicin (DNR, n = 6) and cytarabine (AraC, n = 4) compared to isotype control (IgG CTL; n > 3). Anti-hCD81 antibody-treatment of normal BM cells affected neither c viability, d cell cycle nor e hematopoietic progenitors evaluated by colony forming unit (CFU): CFU-granulocyte/erythroid/macrophage/megakaryocyte (CFU-GEMM), CFU-granulocyte/macrophage (CFU-GM), burst forming unit-erythroid (BFU-E), CFU-erythroid (CFU-E). f Curative treatment schema of OCI-AML3 CDX and two PDX models assessing anti-hCD81 antibody-treatment efficacy. When AML was detectable in PB (>1% hCD45), treatment was initiated with anti-hCD81 or isotype control (black arrows) in combination with chemotherapy consisting of cytarabine (5 days, orange), doxorubicin (3 days, yellow) and gemtuzumab ozogamicin for CDX (day 3 and 5, green). Treatment induced complete remission in CDX and PDX. g Line graph illustrating therapeutic efficacy of anti-hCD81 antibody combination (blue, n = 11) in reducing leukemia burden in PB over time compared to control (red, n = 10) using OCI-AML3 CDX (left panel) with representative corresponding bioluminescent images of live mice (right panel). h Line graph illustrating decreased AML burden in PB (left panels) and lower cumulative incidence of relapse (right panels, defined as PB hCD45 > 1% after achieving complete remission defined as hCD45 < 1%) of two different PDX models treated with anti-hCD81 antibody combination (blue, n = 5 and 3) compared to control (red, n = 3 each). Supplementary Fig. 17 for the more detailed treatment period. Kaplan–Meier curves of i CDX- and j PDX mice stratified by treatment arm showing prolonged OS with the anti-hCD81 antibody chemotherapy combination vs. IgG control chemotherapy combination. CTL control, CFU colony-forming unit, *P < 0.05, **P < 0.005, ***P < 0.0005

Finally, we evaluated a curative protocol⁵⁰ (Fig. 6f) with the addition of intraperitoneal (IP) administration of anti-hCD81 antibody before, during and after standard chemotherapy²⁹ in leukemia bearing NSG mice using not only OCI-AML3 (CDX, Fig. 6g, i), but also two different de novo primary AML (PDX, Fig. 6h, j) and assessed engraftment and treatment response by in vivo imaging and FCM, respectively, in comparison to CD81-IgG control. Our results demonstrate that addition of anti-hCD81 significantly reduced tumor burden as early as one week after the last injection of the antileukemic drug regimen in the OCI-AML3-CDX model (Fig. 6g, Supplementary Fig. 17a). Importantly, in two distinct PDX models, antibody treatment led to a substantial and sustained reduction in tumor burden ten weeks after chemotherapy cessation (Fig. 6h [left panels], Supplementary Fig. 17b). Moreover, all leukemia-bearing mice treated with the anti-hCD81 antibody in combination with chemotherapy exhibited a less aggressive disease, delayed relapse and improved survival (PDX#1: HR [CI] = 5.23 [0.87–31.3], PDX#2: HR [CI] = 6.6 [0.84–54.7], Fig. 6h; CDX: HR [CI] = 2.60 [1.06–6.34], Fig. 6i; PDX#1: HR [CI] = 5.20 [0.53–50.6], PDX#2: HR [CI] = 4.5 [0.64–34.1], Fig. 6j).

In summary, immunotargeting of CD81 in combination with standard chemotherapy may represent a novel safe and effective therapeutic strategy for AML.

Discussion

AML is a hematological stem cell malignancy originating from a rare subpopulation of LSC that is responsible for the accumulation of undifferentiated myeloid blast cells at the expense of the hematopoietic system.⁵¹ Recent sequencing studies have revealed that AML is genetically complex and heterogeneous, with patients often carrying multiple distinct driver mutations, yet no universally dysregulated oncogenic pathway has been identified.⁵² This implies that specific targeting of any particular leukemogenic molecular alteration may not only be effective in just a small subset of patients but may also only eliminate a fraction of neoplastic cells, as multiple mutations frequently coexist within a leukemia. Therefore, it is increasingly evident that achieving long-term remissions or even curing AML would unlikely rely on a single targeted-therapy^53–55 but instead should ideally require the development and inclusion of treatment strategies effective against LSC and across a wide range of AML subtypes, irrespective of their mutational status. In this regard, AML-specific cell surface proteins detectable in most patients have raised particular theranostic interest such as CD33, CD123, CLL-1, CD44, CD47 or TIM-3.^9,51 Nevertheless, targeting these molecules has lagged behind expectations, as severe toxicities due to limited expression specificity and resistance often occur. For example, targeting CD33 (Siglec-3), a myeloid cell surface antigen highly expressed in most AML subtypes and LSC-enriched populations, using a recombinant humanized anti-hCD33 antibody conjugated to the cytotoxic calicheamicin (gemtuzumab ozogamicin)⁵⁶ has shown efficacy especially in non-adverse AML while also inducing severe toxicity given its expression on normal hematopoietic cells.

In this study, we aimed to further assess the therapeutic potential of CD81 in AML, a cell surface glycoprotein expressed on immune cells and known to be essential for B cell maturation and differentiation.^57,58 Our results notably now show that CD81 expressed on blasts, is an independent prognostic marker of treatment response especially in non-CBF AML. Interestingly, we reinforced the prognostic impact of CD81 on RFS and OS, but not on RFS within NPM1-mutant AML³² suggesting that CD81 may not predict relapse risk directly but could inform decisions regarding treatment intensity or post-relapse strategies such as HSCT. Consistent with this, we functionally validated its deleterious role using gain and loss of function in vitro and in vivo approaches by demonstrating that CD81 surface expression affects leukemia aggressiveness and tumor burden by enhancing cell adhesion, migration and invasion processes.^59–61 Indeed, tetraspanins such as CD81 are major components of specialized membrane microdomains known to interact with the extracellular matrix to mediate cell adhesion and motility,^24,62 but a functional role of CD81 in myeloid cell adhesion has not yet been reported.

Since LSCs are thought to play a pivotal role in AML relapse, targeting these cells with a highly expressed, if not entirely specific, therapeutic marker remains crucial for ultimately curing AML while preserving essential cell populations, such as normal HSC clones.⁶³ As cell adhesion and motility are fundamental for LSC interaction with the BM microenvironment and given our results showing the importance of CD81 in these processes, we reasoned that this cell surface protein may also influence LSC function. To this end, we first show that membrane CD81 is detected within the LSC subpopulation. Indeed, previous studies combining immunophenotyping of AML samples and xenotransplantations have established that LSC are characterized by the CD34⁺CD38^-CD90^- immunophenotype in more than 90% of AML cases.^19,46 Interestingly, surface CD81 was also expressed in the CD34⁺CD38^-CD90^-CD123⁺ subset which included CD123 (IL-3Rα), a well-established LSC specific cell surface marker.^11,12,64 As targeting several antigens may be necessary to eliminate leukemic cells that persist at remission, combinatorial immunotargeting⁶⁵ of both CD81 and CD123 may represent a novel therapeutic approach to circumvent antigen escape, a major cause of resistance to immunotherapy. Nevertheless, whether CD81 surface expression persists in LSC and progenitor populations of AML patients during clinical remission remains to be investigated. Consistent with previous studies related to CD33⁶⁶ and CD123,⁶⁷ we show that surface CD81 was overexpressed on bulk tumor and LSC in most AML patients across different subtypes of AML, and more importantly, that its expression increased from diagnosis to relapse, suggesting a role for CD81 in AML treatment escape and disease progression. Therefore, targeting CD81 may not only be effective against persistent leukemic cells but also as second-line treatment when relapse occurs. Importantly, we demonstrate that CD81 expression is highly associated with functional LSC¹⁰ using serial xenoengraftment experiments as well as the LSC17 score,⁴² a known surrogate marker of LSC activity.

The identification of LAPTM4B in association with high CD81 expression is particularly intriguing, given its inclusion in the LSC17⁴² gene signature and its reported involvement in STAT3–AKT–mTOR signaling⁴³ and drug efflux mechanisms.⁶⁸ These features suggest a potential link between CD81 expression and leukemia stem cell–associated programs.⁶⁹ Consistent with this notion, alterations in CD81 levels were accompanied by corresponding changes in STAT3 signaling in our experimental models, and STAT3 activation was also observed in primary AML samples with elevated CD81 expression.

However, changes in gene expression alone may not fully explain the mechanisms by which CD81 promotes leukemia aggressiveness. As a tetraspanin, CD81 is known to organize membrane microdomains and mediate protein–protein interactions that regulate intracellular signaling. It is therefore plausible that CD81 contributes to leukemic phenotypes not only through transcriptional programs, but also through physical interactions and signaling events mediated at the protein level.

The association between high CD81 and chemotherapy resistance is surprising given that CD81 did not alter proliferation rates; however, adhesion to fibronectin induced a G2 arrest, which could partly account for the reduced chemosensitivity⁷⁰ and is consistent with a shift toward a stem-like state.

Finally, as targeting of LSC cell surface proteins is often associated with severe toxicity related to their expression in hematopoietic tissues,^11,46 we evaluated whether anti-hCD81 antibody-treatment may be a safe therapeutic strategy. When exposed to this antibody-treatment, normal BM revealed no hematopoietic toxicity either alone or in combination with intensive chemotherapy and was well tolerated by AML-bearing mice during the course of therapy. This is consistent with previous reports showing that (i) CD81 expression is low in normal BM cells,³² (ii) anti-mCD81 antibody-treatment is well-tolerated in inflammatory bowel diseased mice⁷¹ and more importantly, (iii) an anti-hCD81 antibody developed as an inhibitor of HCV infection⁷² and recognizing both human and monkey CD81 is well-tolerated in primates using several therapeutic doses over a prolonged time.⁷³ Lastly, murine knockout models of CD81 have been established and no life threatening hematopoietic defects were observed except reduced B cell activity,^57,73,74 mirroring the phenotype characterizing human autosomal recessive CD81 deficiency.⁷⁵

In conclusion, we provide proof of concept for CD81 as a new relevant prognostic and therapeutic target in AML independent of its expression in lymphocytes. Importantly, our work demonstrates the functional role of CD81 as a regulator of LSC biology, thereby supporting immunotargeting of CD81 in combination with standard chemotherapy as a promising strategy to improve outcomes, particularly during post-remission therapy and at relapse. While the precise molecular mechanisms remain to be fully delineated, our findings establish that CD81 functions at least in part via the LAPTM4B/STAT3 axis. Future studies will be essential to further decipher how CD81 mediates its effects, whether through ligand binding, adhesion, membrane scaffolding, regulation of cellular trafficking, or modulation of downstream signaling pathways.

Materials and methods

Primary human samples

This study included 16 bone marrow (BM) samples from healthy donors and 290 BM samples from AML patients (i.e., 252 diagnostic and 38 relapsed samples) recruited from 2009 to 2016 at Lille University Hospital. Patient inclusion details and diagnostic characteristics are given in the CONSORT diagram Supplementary Fig. 1 and Table 1. Patients and donors have provided written informed consent in accordance with the Declaration of Helsinki and the Institutional Review Board approved the study (ANA-FE-BMH 20221007).

Cell lines and CD81 gain-of function and loss-of-function models

All cell lines (U-937, OCI-AML3, HNT-34; [DSMZ], MS-5 and HS-5 [ATCC]) were cultured as recommended, authenticated via short tandem repeat profiling (FTA Kit, LGC, ATCC-135-XV-20) and routinely tested for Mycoplasma contamination (Universal Mycoplasma Detection Kit, ATCC). AML cells were transduced with the luciferase containing lentiviral vector (pLenti PGK V5-LUC Neo (623-2), Addgene) and selected for 15 days using G418 (Geneticin antibiotic, InvivoGen, 0.5 mg/ml).

U-937 cells were transfected with pCDM8 hCD81 (Addgene, plasmid #11588) using the Amaxa Nucleofactor II, (kit C, W001; Lonza) and repeatedly sorted to select stable CD81⁺ and WT CD81^- cells (ARIA III, BD Biosciences). HNT-34 and OCI-AML3 cells were transduced with CD81 shRNA or control shRNA (non-targeting) expressing lentiviral vectors (TRCN0000300291[sh291], TRCN0000300293[sh293], TRCN0000300433[sh433] or TRC2 pLKO.5-puro; Sigma-Aldrich) and then kept under puromycin selection pressure (1 µg/ml).

AML xenografts

We generated patient-derived (PDX) and cell-line-derived xenografts (CDX) as we have previously reported.⁴⁷ Immunodeficient mice (NSG, NOD.Cg-Prkdcscid Il2rgtm1Wjl/SzJ, Charles River) are not irradiated or conditioned prior to AML inoculation via tail vein injection of 0.5 ×10⁶ (CDX survival studies) or 1 ×10⁶ cells (CDX tissue studies) and 5 ×10⁶ BMNC cells from AML patient samples containing <15% lymphocytes. We monitor PDX engraftment by flow cytometry (FCM) analysis of murine peripheral blood sampling performed every other week for up to 12 months. Positive PDX engraftment is defined as >1% hCD45⁺ cells relative to mCD45⁺ cells (Supplementary Fig. 13). CDX experiments were monitored by in vivo imaging (luciferin 0.1 mg/kg IP, Xenogen IVIS 50, Perkin Elmer). We collected femurs, sternums, peripheral blood and spleens from CDX mice on day 28 or 31 (U-937 and OCI-AML3 respectively) to determine tissue infiltration of hematopoietic tissues. Bones were fixed with 4% paraformaldehyde-PBS overnight at 4 °C, decalcified (DC1, Q Path) for 18 h at 4 °C, dehydrated and embedded in paraffin. Immunohistochemistry was performed on 4 μm sections (BenchmarkUltra [Roche] using anti-hCD45 [M0701, Dako] and UltraView DAB Detection Kit [Ventana]). We evaluated AML cell homing by FCM analysis of BM as described above using a minimum of 1.0 ×10⁶ acquired events, 72 h after IV injection of 12 ×10⁶ (OCI-AML3) or 20 ×10⁶ (U-937) AML cells. The national ethics committee approved all animal experiments (APAFIS#33794-2021100815558055).

CD81 surface protein expression

We determined CD81 cell surface protein expression on AML BMNC and its progenitor populations by FCM (Navios, Beckman Coulter). 5 ×10⁵ cells after red blood cell lysis, were washed twice with PBS and stained for 30 min at room temperature with an antibody panel as described in the Supplementary Material and Methods and Fig. 2 with details on the gating strategy.⁴⁷

In vitro drug resistance

Leukemic cells were cultured in 96-well plates in the presence of a serial eight-point drug dilution concentration range of cytarabine ([AraC] 0.0025-40 µg/mL) or daunorubicin ([DNR] 0.000042-3.33 µg/mL) for 72 h (AML cell lines) or 96 h (primary AML cells). For each drug concentration, we determined leukemic cell survival and then estimated the half-maximum inhibitory concentration (IC₅₀) for each drug.⁴⁸

Cell adhesion, migration and invasion

96-well plates were coated with fibronectin (1 μg per 0.3 cm², S5171, Sigma-Aldrich) or seeded with 4 ×10⁴ MS-5 or HS-5 for 48 h (70-80% confluence), AML cells (U-937, OCI-AML3 [1.0 ×10⁶]; HNT-34 [0.5 ×10⁶]) were then seeded in triplicates in 100 µL and allowed to adhere for 90 min (U-937, OCI-AML3) or 30 min (HNT-34). We removed non-adherent cells by washing three times with PBS and quantified adherent cells by bioluminescence imaging (luciferin 0.3 mg/mL, Xenogen IVIS 50). For adhesion prior to cell cycle analyses, AML cells were seeded in 12-well plates pre-coated or not with fibronectin. After 16 h of incubation, wells were washed three times with PBS, and adherent cells were collected for downstream analyses. To evaluate migration and invasion, we used 24-transwell plates (5 μm 29442-118, Corning). First, we seeded lower chambers with 5 ×10⁵ HS-5 cells, after 24 h of cell culture, we then loaded 3 ×10⁵ AML cells onto the upper chambers and cultured for another 24 h. Cells which migrated to the lower chambers were quantified by bioluminescence as described above. For invasion experiments, upper chambers were coated with 100 µL gel matrix 0.1X per insert (Basement Membrane Extract (BME) Cultrex 5X Solution, Trevigen) 4 h prior to AML cell loading. Experiments were performed in duplicates. Adhesion, migration and invasion rates were determined as the percentage of cells vs. input control (Supplementary Fig. 8).

Confocal microscopy

AML cells (6 ×10⁴ cells/cm²) were seeded for 24 h on poly-D-Lysine (Gibco) coated Lab-Tek II Chamber Slides (Nunc), slides were rinsed with PBS, and cells fixed with 4% paraformaldehyde-PBS for 20 min and permeabilized with PBS-Triton X-100 0.1% for 10 min at room temperature (RT). After blocking with PBS-BSA 2% for 45 min, cells were stained with Alexa Fluor 488 phalloidin (1:40, Invitrogen) for 30 min at RT in the dark. Nuclei were stained with DAPI (Thermo Fisher, 1:1000) for 10 min. Images were generated (LSM710, Zeiss) and cellular diameter and circularity estimated (ImageJ, https://imagej.nih.gov/ij/).⁷⁶

Anti-hCD81 antibody-treatment

BMNC and CD81 model cells (1.0 ×10⁶) were treated in vitro with anti-hCD81 monoclonal antibody or isotype control at 0.01 µg/µL for 4 h. AML-bearing mice were treated in vivo with chemotherapy in combination with either anti-hCD81 antibody or isotype control (JS-81, BD Biosciences) 100 µg IP using three (OCI-AML3-CDX) or five doses (PDX1 and PDX2, Supplementary Materials), respectively.⁵⁰ Due to the aggressive nature of CDX models, gemtuzumab ozogamicin (GO) was added to chemotherapy, justified by the clinical use in CD33⁺ high-risk AML.⁷⁷ Leukemia progression was determined as described above.

Colony-Forming Unit (CFU) assays

CFU assays were performed according to the MethoCult™ H4034 Optimum protocol (Stemcell Technologies). Briefly, 5 ×10⁴ human BM mononuclear cells (BMNC, donor: 43975, 45189, Lonza) were plated in triplicate and cultured for 10 days. Colonies were counted and normalized to the number of cells originally seeded.

Cell cycle and viability analyses

BMNC (0.25 ×10⁶ cells) were washed with PBS and fixed with ethanol 70% at −20 °C for 2 h, washed twice with PBS-FBS 1% and incubated with RNase A 2 µg/mL for 15 min at 4 °C, stained with PI 35 µg/mL for 30 min and analyzed by FCM. Cell viability was determined using Annexin V and 7-AAD staining following the manufacturer’s recommendations (APC Annexin V Apoptosis Detection Kit 7-AAD, Biolegend).

Gene expression analysis

Total RNA was extracted from diagnostic AML samples (n = 45) and from in vitro models (U-937, OCI-AML3) and hybridized to Affymetrix HG-U133 Plus 2.0 arrays (Thermo Fisher) following the manufacturer’s protocol.⁴² Raw intensity data were background-corrected, low-intensity filtered, and normalized using the robust multiarray average (RMA) method and z-score transformed (affy R package v1.86.0).⁷⁸ Differential expression analysis was conducted with the limma (v3.64.1) R package,⁷⁹ implementing empirical Bayes moderation to identify genes significantly different in AML with high vs. low level of CD81 surface protein. The DE table obtained from limma, provides log2 fold-changes and t-statistics; significant genes were used for downstream gene set, functional and pathway enrichment analyses using the clusterProfiler (v4.16.0) R package,⁸⁰ and genekitr (v1.2.8) R package.⁸¹ The code used was version-controlled and executed in R (v4.5.1) to ensure reproducibility.

Western blot

For each sample, 10 µg of total protein was denatured (10 min, 70 °C), separated on pre-cast polyacrylamide gels (4–12% Bis-Tris Plus or 7% Tris-Acetate; Thermo Fisher), and transferred to nitrocellulose membranes (Bio-Rad Trans-Blot Turbo). Membranes were blocked in 5% milk/TBST for 1 h, incubated overnight at 4 °C with primary antibodies (Supplementary Materials), and then with HRP-conjugated secondary antibodies for 1 h at room temperature. Signals were detected by chemiluminescence (ECL™ Select™, Cytiva) and imaged using the ImageQuant LAS 4000 system.

Statistics

We used Student’s t test and Pearson correlation statistics unless otherwise specified. We evaluated time-to-xenoengraftment using Cox proportional hazard model with P-values (log-rank test). We censored time-at-transplantation date in the event of HSCT or at the date of last follow-up and defined overall survival (OS) as the time from date of inclusion to date of death from any cause. Relapse-free survival (RFS) was determined in patients achieving complete remission (CR) and is defined as the time from date of CR to date of first relapse, or in the case of no relapse, time was censored at the date of last follow-up. The alpha level was set to 5% and all statistical tests were two-sided. We used GraphPad Software version 10 (Prism) and R-v.4.5.1.⁸² Survival curves and analyses were determined using Kaplan–Meier or cumulative incidence methods and hazard ratio (HR) and 95% confidence intervals (CI) are reported. CD81 mRNA expression and patient survival of publicly available Beat-AML dataset³⁴ were obtained via the cBioPortal.^35,36 Similar to our cohort, the analyses included AML (non-FAB-M3, non-core-binding factor (CBF), age >50, treated with intensive chemotherapy).

Author contributions

C.B., C.P., N.D., C.R. and M.C. are responsible for the conception and design of the study; F.G., P.P., D.L., T.B., N.B., S.G., A.B., A.B., F.S., A.P., V.L., V.L., K.G. and MC conducted experiments and analyzed results; C.R. and M.C. assembled and interpreted results; C.B. is the referring physician. F.G., P.P., C.C., N.P. and M.C. wrote the manuscript; and all authors critically revised the article and approved the final version.

Data availability

The data that support the findings of this study are available upon reasonable request. Gene expression data is available at https://www.ebi.ac.uk/biostudies/arrayexpress ArrayExpress accession E-MTAB-16693.

Competing interests

The authors have no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41392-026-02697-2.