Breast implant associated anaplastic large cell lymphoma is a heterogeneous T cell disease with active pro-tumour cross-talk and immune suppression

Abstract

Breast implant-associated anaplastic large cell lymphoma (BIA-ALCL) is a rare T cell lymphoma in women with textured implants. Little is known about the cell of origin (COO) and whether the tumour immune microenvironment (TIME) is critical for BIA-ALCL survival. Single-cell RNA sequencing revealed BIA-ALCL cells were heterogeneous with unique gene profiles per patient and signature genes BATF3, SERPINS, TNFSFR8, and IL2RA. The COO is a CD4⁺ or CD8⁺ memory T cell identified through rearranged TCR and gene expression. The BIA-ALCL TIME included distinct myeloid clusters, dendritic cells (DCs), and monocytes, which may support lymphoma cells. Cytokines IL-13, IL-10, TNF and secretory PDL1 were increased in BIA-ALCL seroma, indicating an immunosuppressive TIME. Interactome analysis revealed a complex network among lymphoma cells, dominated by IL-13, IL-10, and TNF members. Endogenous CD8⁺ T cells exhibit higher checkpoint expression than benign seromas. No clonal relationship exists between BIA-ALCL and endogenous T cells. Tissue-archetype and gene-expression analysis revealed T-cell exclusion and immunosuppressive TIME in invasive disease. In summary, BIA-ALCL cells are diverse with markedly different TIME across stages. The TIME provides immunosuppressive signals to activated/exhausted T cells and homeostatic signals that promote BIA-ALCL proliferation. These new findings may offer novel therapeutic targets for advanced disease patients.

Introduction

Breast implant associated-anaplastic large cell lymphoma (BIA-ALCL) is a rare complication of breast reconstructive or cosmetic surgery [1–3]. BIA-ALCL are of T cell origin and characteristically CD30⁺CD4⁺CD3^-CD25⁺ and anaplastic lymphoma kinase (ALK) negative [4]. BIA-ALCL often presents with a seroma around the implant within a fibrous capsule. The condition is commonly localised to the seroma (so-called in situ disease) or confined to the capsule lining. These patients with localised disease, which includes disease that penetrates the capsule but does not extend beyond the capsule are treated with surgery alone i.e. total or partial capsulectomy [5–7]. In contrast, patients who present with invasive disease (Stage IIb and above) that extends beyond the capsule, often associated with a mass, are at high risk of disease spread and are treated with more extensive surgery, systemic chemo-immunotherapy and/or radiotherapy [5–7].

Numerous mutations have been recognised in BIA-ALCL with the most common being activating mutations of JAK1 and STAT3 [8] [9,10]. Despite this well described mutational landscape, the precise aetiology of BIA-ALCL remains unknown [11]. What is known is that the presence of highly textured implant plays a critical role; as all patients with BIA-ALCL to date have or previously had, a textured implant [3]. There is evidence that such highly textured implants can develop a surface biofilm, and it has been postulated that this leads to the subsequent formation of a persistent microbiome which could provide a source of antigens to stimulate T cell proliferation [12–14]. However, this hypothesis has not been definitively proven, and indeed, when specifically studying BIA-ALCLs, no distinct microbiome signatures have been identified [15].

Breast tissue, including the ductal system, is subject to T cells trafficking; into both normal breast tissue and breast cancer to form transient effector or tissue resident T cell populations [16,17]. This immune surveillance of normal breast tissue indicates a T cell response to antigens generated by commensals, pathogens or breast ductal tissue [16]. Although breast implants are inserted adjacent to breast tissue in the periprosthetic space, which is considered an immune privileged site, T cells may still traffic to the capsule and respond to chronic inflammation or environmental cues. However, like systemic and primary cutaneous ALCL, at the time of diagnosis of BIA-ALCL (typically years after implant insertion) there is no T cell receptor (TCR) detectable on the cell surface [18]. We know the malignant T cells have either rearranged clonal TCR genes or non-productive clonal TCR genes, indicating that some of these clones were once able to recognize antigen via peptide-MHC or other non-peptide antigens through non-classical MHC I like molecules [9]. This suggests that once BIA-ALCL is established, the cells undergo antigen-independent proliferation in the seroma, which in turn may be driven by tumour intrinsic mechanisms or require cross-talk with the endogenous (endogenous cells refers to cells present within the patient tissues that are not lymphoma cells) immune cells within the seroma. However, little is known about the invasive form of the disease, in which BIA-ALCL tumour cells invade beyond the capsule into the breast tissue and lymph nodes, transforming into tumour mass-associated BIA-ALCL, similar to systemic ALCL or other forms of peripheral T cell lymphoma.

In this study, we examined BIA-ALCL patient specimens to understand the mechanisms driving lymphoma pathogenesis. Our working hypothesis is that tumour intrinsic (lymphoma-lymphoma) and extrinsic (lymphoma-TME) signals play different roles in seroma only and invasive stages of BIA-ALCL. To investigate this working hypothesis, we took an orthogonal approach to explore breast implant lymphoma intrinsic cues (gene expression, mutations, and clonality) as well as TME (extrinsic) signals through integrated single-cell analyses, gene expression and spatial analysis. We explored clinically annotated samples across the spectrum of disease (seroma only, tissue local and invasive) to address this working hypothesis.

Materials and methods

Patient specimens

Patients presenting with a breast effusion (seroma), either involving lymphoma (localised BIA-ALCL) or benign (no lymphoma cells identified), were included in this study (Table S1). Seroma fluid was collected at the time of capsulectomy surgery or by aspiration under sterile conditions. Specimens in excess of the patient’s diagnostic requirements were collected for research. FFPE blocks from patient tissues, including those from cutaneous ALCL, were also obtained and are detailed in Table S1. The number of samples used in each experiment is listed in every figure legend.

Cell preparation

Seroma fluid was filtered through a 100 μm strainer. The flow through was then centrifuged at 400 g for 5 min at 4^oC. The supernatant was frozen down at -80⁰C and the seroma cell pellet was cryopreserved in freezing media (90% v/v FBS + 10% v/v DMSO) at 1-5 million cells/cryovial for further studies.

Detailed methods on the assays and analysis are provided in the supplementary methods.

Results

Single cell RNA sequencing revealed diversity in ALCL tumour cells between patients and increased myeloid cell clusters

We used an orthogonal approach to explore BIA-ALCL cell development, their survival and interactions between BIA-ALCL cells and the endogenous immune infiltrate in the breast implant seroma. We obtained seroma samples from patients undergoing capsulectomy surgery in both benign disease (n = 10) and BIA-ALCL (n = 21). (Table S1). The entire specimen comprised the implant, capsule tissue, and seroma fluid. To better understand the BIA-ALCL microenvironment, we comprehensively characterised the cells present within the seroma using flow cytometry, single cell (sc) RNA citeseq, T cell receptor (TCR) sequencing and targeted DNA sequencing to define the signature mutations of BIA-ALCL (Fig. 1). We performed TCRVβ deep sequencing on the lymphoma cells and analyzed their clonal relationship with endogenous CD4⁺ T cells. Finally, to better understand the BIA-ALCL interface with immune and stromal components in breast tissue we analyzed the TIME (gene expression and spatial immune context) of Stage I and invasive (Stage II) disease and compared this to cutaneous ALCL.

Fig. 1. Study schema.

Specimens were obtained from patients who underwent capsulectomy surgery; the specimens included seroma cells and seroma fluid supernatant. The cohort included patients with BIA-ALCL or benign seroma. The seroma cells were analysed by flow and mass cytometry, single-cell RNA sequencing, TCR sequencing and targeted DNA sequencing for gene mutations. To study the tumour microenvironment, seroma fluid was analysed for cytokine/chemokine levels. In addition, using the single-cell RNAseq data, we performed an interactome analysis to reveal cell-cell interactions that were specific to the BIA-ALCL TIME. To study the tumour microenvironment and spatial relationships, bulk RNA sequencing and multiplex IHC (7-plex) were performed on formalin-fixed paraffin-embedded (FFPE) tissue sections from BIA-ALCL and pcALCL. n indicates the number of samples analysed per method.

Initial analysis was performed on seroma T cells and CD30+ cells (Fig. S1, S2). Collated data for the cohort showed no significant difference between BIA-ALCL and benign seroma for the myeloid subsets by total CD11c⁺, CD11c⁺HLA-DR⁺, total CD11b⁺ and CD11b⁺CD14⁺HLA-DR⁺ (Fig. S1).

Mass cytometry analysis of lymphoma seroma cells (as demonstrated in BIA-032) showed three distinct clusters of lymphoma cells, including CD30⁺, CD30⁺IL10⁺ and CD30⁺IFN-γ⁺ clusters (Fig. S2A,C). Endogenous T cells were present in both the malignant seroma (BIA-032) and the benign seroma (BIA-025). Whilst endogenous T cell clusters were similar between these two samples, the malignant seroma included CD8⁺TIGIT⁺PD1⁺ and CD45RA⁺ T effector memory cells consistent with activated/exhausted T cells suggesting an antigen specific response (Fig. S2C-D).

We next analyzed BIA-ALCL seroma cells using CITE-seq for surface proteins, scRNA-seq for immune cell clusters based on gene expression profile (GEP), and TCR sequencing to identify T cell clones, clonal expansion, and sharing across T cell differentiation clusters. We analyzed live sorted seroma cells from BIA-ALCL (n = 7) and benign seroma patient (n = 4) specimens. This is the first single-cell dataset published in this disease; 76,894 total cells were sequenced out of which 37,476 were tumour cells and 39,418 were endogenous immune cells. When all single cell events were collated, unsupervised clustering of the cells revealed 16 distinct cell populations (Fig. 2A). These were annotated using marker expression both via protein (CITEseq) and GEP (Table S2, Fig. S3A). This strategy identified five main tumour clusters (distinct for each patient), six myeloid cell clusters, eight T cell clusters and two B cell clusters. Analysis of individual patient samples revealed marked heterogeneity in immune cell lineage and differentiation within both the benign and malignant seroma cohorts (FigS. S3A–C). This could be attributed to the interactions between the cells and the cytokine/chemokine milieu. Analysis of the CD8+ tumour specimen BIA37 did not demonstrate dominance of any specific immune cell type, most of the seroma was dominated by the tumour (Fig. S3C). When the endogenous immune cells were compared between benign and malignant seroma (Fig. 2B, C) myeloid cell clusters were more abundant in the BIA-ALCL specimens. In addition, the BIA-ALCL patient seroma included more CD8⁺ T cells with effector memory or exhausted like phenotype, whilst the benign seroma patient had Th1/Th17 cells (Fig. 2D, S3B, C).

Fig. 2. Single-cell RNA sequencing reveals diversity among lymphoma patient cells, with an increased proportion of myeloid cells in malignant samples.

A Collated single-cell sequencing data is depicted in a UMAP for ALCL (n = 7) and benign (n = 4) seroma cells, where lymphoma cells (red) and endogenous immune cell clusters are annotated using gene expression of lineage and differentiation markers. This data is shown in Table S2. Live sorted cells were sequenced. Relative distribution of endogenous immune cells from benign B and ALCL C specimens are highlighted in black D Bar chart showing the relative distribution of cell clusters between samples from benign and malignant seroma.

BIA-ALCL cells gene expression differs between patients

BIA-ALCL cells from each patient clustered separately when visualised in a UMAP indicating a distinct gene expression profile (GEP) for each lymphoma (Fig. 3A). We identified the difference in tumour cell GEP between patients. This analysis revealed the top most differentially expressed genes (DEGs) in tumour cells per patient (Fig. 3Band Table S3). Unique lymphoma-related gene expression identified in patient samples included ribosomal protein related genes (RPL17, RPL23A), cytokine/chemokine or NK/T cell effector protein related genes (CCL5, IL10, CCL17, GzmK, GNLY), genes related to expression of a T cell receptor sub-unit, MHC related and a T cell homeostatic cytokine receptor (CD3E, B2M, IL7R) (Fig. 3Band Table S3). The differences in these lymphoma DEGs also related to both the state of differentiation and the cell of origin, which was either CD4⁺ or CD8⁺ T cells as the lymphoma cells expressed one of these co-receptors.

Fig. 3. Tumour cell gene expression profile reveals diversity between patient tumours.

A UMAP depicting distinct cell clusters of lymphoma cells from BIA-ALCL patients. Two patient samples were excluded from this analysis due to low tumour cell numbers. B Bubble plots showing the top 10 highly differentially expressed genes (red) for each patient’s tumour cells when compared to the other patients’ tumour cells. Shown is the normalised expression of the corresponding gene in each patient. The size of the bubble correlates to the percentage of cells expressing that gene. Gene expressions were normalised using Seurat [51]. The LogNormalize method was used with a scale factor of 10,000. Effectively, each cell’s gene normalised expression is calculated as the log value of the scale factor times gene count divided by the cell’s total gene count. C Selective commonly expressed genes that are shared in the lymphoma cells across all patients and distinct from endogenous T cells within the cohort (shown are 13 out of 1233 DEG). The log fold change (log FC) and adjusted p value (p_val_adj) is depicted. Pct corresponds to the proportion of cells expressing a gene (pct 1 is BIA tumour cells and pct 2 is endogenous T cells). D Ontogeny analysis for BIA-ALCL CD4⁺ tumour cells in comparison to mycoses fungoides and Sezary syndrome is shown as a UMAP with individual clusters for BIA-ALCL patient lymphoma cells, mycoses fungoides (MF early stage (n = 5) and advanced stage (n = 9)), Sezary syndrome (n = 10) and control normal CD4⁺ T cells (PBMC CD4 T, skin polyclonal CD4 T, seroma CD4 T from all samples). E Gene set enrichment analysis comparing BIA-ALCL tumour cells to MF cells or Sezary syndrome tumour cells respectively. Bubble plots represent Normalised Enrichment Score of gene sets from the Gene Ontology (GO) pathway datasets, with each bubble representing a gene set and its size reflecting the number of genes assigned to that set.

To identify common gene signatures between BIA-ALCL cells from different patients, we combined the GEP of lymphoma cells from all patients and compared these to the GEP of all the endogenous CD4⁺ and CD8⁺ T cells from the seroma (Fig. 3C). This differential expression analysis revealed that the BIA-ALCL cells expressed genes from the serpin family (SERPINE2, SERPINA1, SERPINB1), cancer associated genes (PRDX2, MIR155HG, SLC16A3) and TNF signalling pathway-related genes (TXNDC17, TNFRSF8). TNFRSF8 encodes for CD30, a protein commonly expressed by ALCL cells. Other expressed genes common to BIA-ALCL cells included the transcription factor BATF3, the T cell effector protein GZMB, Thioredoxin TXN, the high-affinity IL-2 receptor (IL2RA), and the Th2 cytokines IL13, IL10 and the hypoxia-related gene CA9 (Fig. 3C). Some of these genes, highly expressed in BIA-ALCL, such as BATF3, TNFRSF8, and SERPINA1, have been identified before for systemic ALCL (ALK+ or ALK-) (Fig. 3Cand Table S4) [19]. The hypoxia-related gene CA9 was also identified through bulk RNA sequencing in BIA-ALCL cells from capsule tissue in several other studies [20–22].

We then performed an ontogeny analysis and compared the BIA-ALCL (excluding BIA-37, a CD8 ⁺ ALCL tumour) to other T cell lymphoma cells using published datasets, including mycosis fungoides (MF), Sezary syndrome (SS) and normal CD4⁺ T cells (Fig. 3D) [23, 24]. UMAP display showed the BIA-ALCL cell clusters were distinct from each other and from MF (n = 14) and SS (n = 10) cells [23,24]. In contrast, the endogenous normal CD4⁺ T cells from all the samples analysed clustered closely together as well as to the SS. Because the UMAP display clusters cells based on similar GEP, our data indicate that BIA-ALCL cells have a GEP that is less related to normal CD4⁺ T cells, MF tumour, and SS tumour cells. Further support for this was provided by directly comparing BIA-ALCL tumour cells with MF and SS tumour cells, and by examining signalling pathways in DEGs unique to BIA-ALCL tumour cells, MF cells, and SS tumour cells, when compared with their respective endogenous CD4 ⁺ T cells (Fig. 3Eand S4). Gene set enrichment analysis (GSEA) and GO pathway analysis using the BIA-ALCL DEGs revealed that pathways relating to hypoxia, cellular metabolism (glycolysis, fatty acid metabolism), were all upregulated compared to SS and MF (Fig. 3Eand S4), whilst cancer-related pathways were only upregulated when compared to SS (Fig. 3E). In contrast, pathways related to T cell differentiation and receptor signalling were downregulated, suggesting overall dysregulation of normal T cell function processes (Fig. S4).

Exhausted endogenous T cells are present in BIA-ALCL seroma

To further explore the BIA-ALCL microenvironment we characterised the endogenous immune infiltrating cells from benign and BIA-ALCL seroma. Unsupervised clustering of the endogenous immune cells revealed 16 distinct clusters (Fig. 4A), these were annotated based on marker expression (Table S2). Myeloid cells were largely present in the BIA-ALCL patient samples and comprised four main dendritic cell clusters (cDC1, cDC2, myeloid DC, pDC), two monocyte clusters (classical and intermediate) and B cells (Fig. 4A). Figure S5 depicts the distribution of individual patient samples across these endogenous immune clusters. To better visualise the T cell clusters, we performed unsupervised clustering and UMAP visualisation on the T cells identified in Fig. 4A; using this approach we identified eight T cell clusters (Fig. 4B). CD8⁺ TEM were the major T cell subset in the BIA-ALCL patient specimens with exhausted-like T cells identified in one ALCL patient sample BIA015, which could be an indicator of an active anti-tumour response (Fig. 4Band Fig. S3). In contrast, CD4⁺ T cells were dominant in the benign patient samples and were either Th1 or Th1/Th17. Other T cell subsets in both cohorts included MAIT cells, γδ-T cells, CD8⁺ TCM and CD4⁺ Tregs (Fig. 4B). The CD8⁺ TEM cells from the BIA-ALCL patients expressed higher levels of cytokine IFNG, immune checkpoint genes (LAG3 and TIGIT) and were enriched for a T cell exhaustion signature (Fig. 4C). These findings indicate that the endogenous T cells in BIA-ALCL were chronically activated and capable of producing inflammatory cytokines. However, the lack of available experimental models affect our ability to test this hypothesis. In contrast, we observed a higher proportion of TCF7 expressing T cells in the benign samples, suggesting more progenitor memory T cells (Fig. 4C). Unsupervised clustering of the myeloid cells revealed multiple subsets of myeloid cells including myeloid DC, and monocyte clusters present within the BIA-ALCL seroma (Fig. 4D). The top 10 genes expressed by monocyte clusters related to monocyte to macrophage differentiation (APOC1), neutrophil chemotaxis (S110A8, S100A9), and inflammatory monocyte signatures (IFITM3, APOBEC3A) (Fig. S6and Table S5).

Fig. 4. ALCL patient seroma endogenous immune cells comprise an activated exhausted T cell profile, inflammatory monocytes and dendritic cells.

UMAP showing A all endogenous immune cells with each colour representing a cell type (left panel) and divided by group (right panel) B T cells only with different subsets highlighted by colour C multiple panels showing distribution of T cell immune checkpoint, effector molecules and T cell exhaustion signature [52] score calculated using UCell method [53] across T cell clusters from benign and malignant seroma. The scale on the right of each panel shows expression levels for each cell, displayed on a grey to purple scale, with increased expression levels highlighted in purple and zero indicating no expression. D UMAP showing seroma myeloid cells from benign and ALCL patients, with each colour representing a myeloid cell subset (left panel) and divided by group (right panel).

Whilst specific endogenous immune cell clusters were unique to BIA-ALCL samples, others were shared between BIA-ALCL and benign seroma (Fig. 2). Despite this, we hypothesized that these ‘shared’ immune cell clusters had distinct signalling pathways between the two cohorts. To address this, we compared the GEP for each shared cluster using pseudobulk analysis and identified the DEGs. For the CD4⁺ Th1 cell cluster, gene set enrichment analysis (GSEA) and GO pathway analysis identified genes upregulated in the BIA-ALCL samples were enriched in pathways related to leukocyte chemotaxis, activation, differentiation, glycolysis, TNF-NFκb signalling, and hypoxia (Fig. S7A). For the CD4⁺ Treg cluster pathways related to hypoxia, glycolysis, IL2-STAT5 signalling, and proliferation were upregulated in the BIA-ALCL samples (Fig. S7B). Finally, for the CD8⁺ TCM cluster and CD8⁺ TEM cluster, upregulated pathways in BIA-ALCL seroma included antigen processing and presentation, activation, differentiation, cytotoxicity, chemotaxis, TNF signalling via NFκb and hypoxia (Figs. S7C and S7D). In summary, our findings indicate BIA-ALCL seroma included activated and exhausted endogenous T cells, indicating chronic stimulation to antigens specifically in the BIA-ALCL TIME.

Interactome analysis reveals cell-cell interactions specific to the BIA-ALCL tumour microenvironment

To understand the interactions between the BIA-ALCL cells and endogenous immune cells, and their influence on the tumour microenvironment, we performed a cell-based interactome analysis. This interactome analysis using Cellphone DB and Cellchat revealed the potential ligand-receptor pair interactions between cells at a cluster level. These cluster interactions were different for cells from benign vs. BIA-ALCL seroma, more interactions were seen between tumour cells and other immune cells in the BIA-ALCL seroma (Fig. 5A). We examined the potential cell-cell interactions and grouped them into three categories [1] dominant in BIA-ALCL seroma [2] common to BIA-ALCL and benign seroma [3] dominant in benign seroma only (Fig. 5B). Unique interactions for BIA-ALCL included those between tumour and DCs (plasmacytoid and myeloid DCs), plus tumour to CD8⁺ TEM (Fig. 5B). Potential receptor-ligand pairs between tumour and myeloid DCs include CD40-TNFSF13B (BAFF), ICAM1-SPN (CD43), CD74-COPA (Coatomer subunit alpha), CD74-MIF (Macrophage migration inhibitory factor), LGALS9 (Galectin 9)-CD47 and SIRPA (Signal regulatory protein α)-CD47. Receptor-ligand pairs between pDCs and tumour cells include CD74-COPA and CD74-MIF. Finally, potential CD8⁺ TEM-tumour interactions occur via HLA-C-FAM3C (family with sequence similarity 3), suggesting immune recognition but may also play a role in T cell exhaustion [25]. In addition, potential common tumour-tumour interactions (present in >6 samples in the cohort) included CXADR-FAM3C (coxsackie and adenovirus receptor- family with sequence similarity 3, member C) and TNF-Fas (tumour necrosis factor-Fas) (Figs. 5Cand S8). Less common tumour-tumour interactions (present in > 3 BIA-ALCL samples) include IGF1-IGF1R (insulin like growth factor 1- insulin like growth factor 1 receptor), IL13-TMEM219 (transmembrane protein 219), IL15-IL15RA, LTA-TNFRSF14 (Lymphotoxin alpha-CD258), LTA-TNFRSF1A (tumour necrosis factor receptor superfamily member 1A), TNF-FAS, TNF-RIPK1 (receptor interacting protein kinase 1), TNF-TNFRSF1A, TNFRSF11B-TNFSF10 (osteoproterogerin-tumour necrosis factor ligand superfamily member 10) (Fig. 5C). Taken together, this interactome analysis indicates cell-cell interactions within the tumour cell cluster and also between tumour and endogenous immune cells, which support BIA-ALCL cell survival and proliferation, plus indicate T cell exhaustion.

Fig. 5. ALCL seroma displays a mixed Th1/Th2 immune phenotype.

A Representative potential cell-cell interactions visualised using CellChat [54] showing ligand-receptor pairs between cells within the seroma for each patient, where a minimum of 10% of the cells within the cluster expressed the ligand-receptor gene interaction, which were identified using the CellphoneDB dataset [55]. B Bubble plot showing ligand-receptor interactions across clusters and subdivided into three categories: high-high (present in more than 75% of specimens within each group), high-low (more than 75% in ALCL and less than 25% in benign samples), low-high (less than 25% in ALCL and more than 75% benign samples). C Interactome map for BIA-ALCL samples showing the dominant ligand-receptor interactions between tumour cells and immune cells within the seroma across all patients. The size of the dot corresponds to the number of samples that displayed a specific interaction. D Dot plots depicting cytokine levels (IL-13, sPD-L1, TNF, IL-10, IL-8, IFNγ, IL-18) quantified in the seroma fluid from ALCL and benign seroma patients. (n = 10 benign and n = 21 ALCL specimens). Data were analysed using the Wilcoxon signed-rank test, *P < 0.05, ****P < 0.001.

To further evaluate the BIA-ALCL TIME we performed cytokine and chemokine analysis on the seroma fluid. This showed BIA-ALCL seroma was significantly enriched for the cytokines TNF, IL-13, IL-10 and soluble PDL1 (sPDL1), but not IL-18 or IL-23. In contrast, benign seroma was significantly enriched for IL-8 (Figs. 5Dand S9). The significantly increased IL-13, IL-10 and TNF levels in our BIA-ALCL cohort confirm a prior observation [20, 26,27]. In addition, the IL13-TMEM219 interaction was predicted between lymphoma cells, or lymphoma cells and myeloid DCs (Fig. 5C), suggesting this signature cytokine has pleiotropic effects within the BIA-ALCL seroma.

Whilst the T cell effector cytokines IFN-γ and TNF levels were increased in BIA-ALCL patients (Figs. 4C and 5D), the CD8⁺ T cells also expressed an exhausted phenotype with increased immune checkpoint genes (TIGIT and LAG3) (Fig. 4C). Mutational analysis using targeted DNA sequencing identified activating STAT3 mutations along with novel JAK1 mutations, which may lead to constitutive activation of the NFκb pathway and secretion of TNF (Table 1). Taken together, these findings indicate the BIA-ALCL tumour microenvironment includes a complex network of cell-cell interactions including those between the lymphoma cells, and the lymphoma cells and endogenous immune cells. Collectively, these interactions support the immunosuppressive nature of BIA-ALCL.

Table 1.

List of gene mutations on BIA-ALCL seroma samples.

Targeted sequencing of ALCL patient tumours
Samples	20q del (incl 20q13.13)	RPL5 del (focal 1p del, incl 1p22.1)	Variants	Chromosome location	Amino acid changes
BIA-008	Yes	Yes	SOCS1	NM_003745.1:c.106_122dup	p.(Cys42Profs*49)
BIA-012	No	No	PRDM1	NM_001198.3:c.665-1 G > C	p.?
			JAK1	NM_002227.2:c.3440 A > T	p.(Glu1147Val)
			ATM	NM_000051.3:c.974 A > T	p.(His325Leu)
			ATM	NM_000051.3:c.1370 G > A	p.(Arg457Gln)
BIA-015	Yes	Yes	JAK1	NM_002227.2:c.3299 C > T	NP_002218.2:p.(Thr1100Ile)
			JAK1	NM_002227.2:c.3301 G > T	NP_002218.2:p.(Val1101Phe)
BIA-032	Yes	No	STAT3	NM_139276.2:c.1940A>T	p.(Asn647Ile)
BIA-037	Yes	Yes	STAT3	NM_139276.2:c.1840A>C	p.(Ser614Arg)

BIA-ALCL seroma include clonally expanded endogenous T cells with no clonal relationship to the tumour cells

Using single cell TCRseq (αβ TCR) and TCRVβ deep sequencing we explored the relative abundance of individual T cell clonotypes within the benign and BIA-ALCL seroma. The endogenous T cells from BIA-ALCL were more clonally expanded than in the benign seroma (Fig. S10 A,B). We then explored whether individual endogenous T cell clones were shared across different T cell clusters using the combined single cell TCR and RNAseq data. This analysis showed TCR sharing between endogenous T cell clusters was limited for both benign and BIA-ALCL seroma (Fig. S10 C,D). Furthermore, we compared the clonal relationship between the endogenous seroma T cells and the BIA-ALCL cells, which revealed that there was no TCR sharing between the ALCL cells and endogenous T cells in all BIA-ALCL samples (Fig. S10D), indicating there is no clonal relationship between BIA-ALCL cells and the endogenous T cells (neither CD4⁺ nor CD8⁺).

BIA-ALCL invasive disease displays a unique signature when compared to cutaneous ALCL and non-invasive BIA-ALCL

To this point, we have focused the study on breast seroma cells; here, we extend our investigation to further explore the BIA-ALCL TIME in tissue. To achieve this, we first performed RNA sequencing on tumour tissue from 11 BIA-ALCL and 6 cutaneous ALCL patients (Fig. 6) and explored the differences in gene expression between invasive BIA-ALCL, non-invasive ALCL, and cutaneous ALCL (pcALCL). This approach revealed 27 upregulated genes and 106 downregulated genes when BIA-ALCL was compared to primary cutaneous ALCL (pcALCL). Some of the genes that were upregulated included those related to cytokine and chemokine signalling (CCL26, SERPINE1, SOCS3) and macrophages (CD163), as well as cell cycle and progression (IGF2BP1) (Fig. 6Aand Table S8). Genes that were downregulated were previously implicated in pcALCL (FGFR3, TRIM29, IL22RA1), as well as in cell cycle and DNA repair (Fig. 6A). Gene set enrichment analysis (GSEA) using hallmark and GO pathway analysis using the BIA-ALCL DEGs revealed that pathways relating to hypoxia, glycolysis, JAK STAT signalling, inflammatory response, EMT, TNF signalling, leukocyte migration and chemotaxis, and immune response genes were all upregulated especially when compared to pcALCL (Fig. 6B, C). These findings indicate that the BIA-ALCL TIME is more inflamed compared to pcALCL, involving multiple signalling pathways of innate and adaptive immunity. Similarly, we observed increased inflammatory response, hypoxia, IL6 JAK STAT3 signalling, and TNFα signalling gene signatures when we compared invasive BIA-ALCL with pcALCL (Supplementary Fig. S11), suggesting inherent differences between any stage of BIA-ALCL and pcALCL.

Fig. 6. Gene expression analysis comparing BIA-ALCL tumour tissue to cutaneous ALCL (pcALCL) revealed distinct gene signatures.

A Volcano plot showing differentially expressed genes comparing BIA-ALCL (n = 11) versus pcALCL (n = 6). Upregulated and downregulated genes are highlighted in colour. B GSEA pathway analysis comparing BIA-ALCL versus pcALCL. Bubble plots display normalised enrichment scores of gene sets using Hallmark or Gene Ontology (GO) Biological Processing pathway datasets, with each bubble representing a gene set and its size reflecting the number of genes assigned to that gene set. P-values are depicted on a colour scale. C GSEA immune pathway analysis comparing BIA-ALCL versus pcALCL. Bubble plots showing normalised enrichment scores of gene sets and p-values are depicted on a colour scale. D Volcano plot illustrating differentially expressed genes comparing invasive versus non-invasive BIA-ALCL patient samples. Upregulated and downregulated genes are highlighted in colour. E GSEA pathway analysis comparing invasive versus non-invasive BIA-ALCL tissue. Bubble plots display normalised enrichment scores for gene sets from immune and Hallmark pathway datasets, as shown above.

We then compared invasive (Stage Ib and above) to non-invasive ALCL (Stage IA only), we identified 159 upregulated genes and 64 downregulated genes (Fig. 6Dand Table S9). Some of the genes that were upregulated in the invasive BIA-ALCL tumour tissues include cell cycle/ cancer proliferation genes (RRM2, TPX2, PTTG1, TROAP), ECM / tumour microenvironment remodelling genes (COL10A1, ITGA10), metabolism related genes (SLC16A10, SLC29A2, SLC10A5, SLC04A1) and immune escape [HLA-G (NK inhibitor), RAET1E1 (an NKG2D ligand)] (Fig. 6D) [28–31]. GSEA analysis further revealed that the genes upregulated in invasive ALCL are associated with tumour-related pathways. These upregulated gene pathways were Myc targets, as well as E2F and cellular metabolism and cell cycle pathways (Fig. 6E). In contrast, immune signalling pathways were downregulated, such as TNF signalling via NFκB and inflammatory response (Fig. 6E). Overall, our analysis indicated that the anti-tumour immune response is diminished in invasive ALCL, alongside increased ECM remodelling and cell proliferation. This may either be a cause or a consequence of the infiltrative nature of the BIA-ALCL tumour, or it could explain how ALCL cells spread beyond the capsule and form tumours.

Spatial analysis revealed an inflamed TIME in non-invasive BIA-ALCL

Cellular density analyses revealed that immune cells such as macrophages, DCs, B cells and PDL1 expressing myeloid cells were significantly enriched in non-invasive early-stage BIA-ALCL tissue sections (Fig. 7A, B). We then analysed immune cell neighbourhoods (tissue archetypes) and observed that tumour enriched archetypes were mostly devoid of immune cells (macrophages, DCs and B cells), clustered separately (Figs. 7Cand S12). Of note, these tumour containing archetypes were enriched in invasive BIA-ALCL (Fig. 7D). We also analysed the T cell subsets within the infiltrating immune cells and did not observe significant differences in T cell subsets including CD8 ⁺ T cells, CD4⁺ Tregs, and CD8 ⁺ TRM. However, we observed an increase in CD4 ⁺ T cells in invasive BIA-ALCL, accompanied by a decrease in CD4 ⁺ TRM cells (Figs. 7E and F). The T cell neighbourhood analysis revealed that tumour-dominated areas in invasive BIA-ALCL were devoid of CD8 ⁺ T cells, dnT (CD4^-CD8^- T cells) and dpT (CD4 ⁺ CD8 ⁺ ) T cells, with CD4 + T cells being an exception. There were no differences in the mixed CD4 ⁺ T cell and tumour cell archetypes between invasive and non-invasive disease. Overall, our analysis revealed that although there are immune cells in the TME of BIA-ALCL in tissue, the immune response is more inflammatory or dominant in non-invasive disease compared to invasive disease.

Fig. 7. Spatial analysis revealed myeloid and CD4 T cell differences across different stages of disease.

A Representative spatial plot of a stage I non-invasive capsule tissue using SPIAT. B Density analysis of the different cell types identified using the pan-immune panel (Invasive samples (n = 4), non-invasive samples (n = 4)). Data were analysed using the Wilcoxon signed-rank test, *P < 0.05. C UMAPs depicting cell cluster archetypes analysed using SPIAT software. (Left panel) UMAP of different archetypes across all samples and (right panel) UMAP of archetype distribution based on the group highlighted in colour. D Heatmap of collated data displaying the proportion of different archetypes across the two groups, with colour key showing clusters from C and abundance across the two groups, as well as proportion of each cell type within an archetype. E Representative spatial plot of a stage II invasive tumour tissue using SPIAT. F Density analysis of the different cell types identified using the T cell panel (Invasive samples (n = 4), non-invasive samples (n = 5)). Data were analysed using the Wilcoxon signed-rank test; *P < 0.05. G UMAPs depicting T cell cluster archetypes analysed using SPIAT software. (Left panel) UMAP of different T cell archetypes across all samples and (right panel) UMAP of archetype distribution based on the group highlighted in colour. H Heatmap of collated data displaying the proportion of different T cell archetypes across the two groups, similar to the analysis above. Invasive samples (n = 4), non-invasive samples (n = 5).

Discussion

BIA-ALCL occurs exclusively in patients who have a textured implant [3,4], used for both cosmetic and reconstructive reasons. The incidence of BIA-ALCL is variable depending on the study with multiple different studies reporting different risk estimates [2,3, 32]. Textured implant usage peaked in 2016 and consequently, it is expected BIA-ALCL incidence will peak around 2026 [33].

Indeed, this rare disease is unique in that, unlike any other lymphomas, there is a consistent underlying ‘smoking gun’ – in this context, it is the exposure to a textured breast implant. Thus, investigating the pathobiology of BIA-ALCL may provide insights into other ALK negative ALCL, particularly the systemic or primary cutaneous types, where there are no recognised developmental associations. A number of studies have identified the genomic mutations, particularly in the JAK-STAT pathway, that are relatively common. However, there are a number of issues that remained unresolved; these include the cell of origin, the role of the breast TIME, the association to endogenous T and immune cells, and what facilitates ongoing lymphoma proliferation. We know for in situ disease, surgical removal of the capsule (including implant and seroma) prevents redevelopment of BIA-ALCL [34]. This suggests that indeed the breast implant TIME is critical for establishing and maintaining BIA-ALCL, but exactly how this occurs has not been fully revealed to date. There are no syngeneic mouse models and because this is a rare disease, studies to date comprise case study reports and small cohort studies, and none of these have addressed the role of the TIME in BIA-ALCL. Nonetheless, Di Napoli et al. and Oishi et al. showed that they could create a xenograft mouse model in NSG mice and the cells produced IL-10 in the xenograft [20, 22].

This is the first study to reveal BIA-ALCL cell gene expression at a single cell level. In previous studies, the gene expression profile of BIA-ALCL by bulk RNAseq analysis was shown to be enriched for genes in cell motility programs, myeloid cell differentiation and viral gene transcription [20,21, 35]. To date there is little single cell data published on T cell lymphomas [36]. Moreover, with respect to BIA-ALCL, prior studies have used RNAseq or microarray analyses [20,21, 37]. Our single cell data not only identified a core set of genes common to BIA-ALCL cells within the cohort (Fig. 3C), but also a level of heterogeneity that was previously under-appreciated. This heterogeneity exists for BIA-ALCL cells between patients, we also showed that each patient has more than one clone present within their ALCL cluster. This marked tumour heterogeneity of BIA-ALCL is further emphasised when the GEP are compared to the relatively homogenous gene expression profile of MF and SS cells. In contrast, there is no other published study in systemic ALCL that has examined the gene expression for the systemic ALCL cells at a single-cell level. In a recent study by Fiore et al., bulk RNA sequencing on a PDX model identified a core set of genes; some of these genes (UBL, RPL18A, RPS10 and PPARG) were also identified in our single-cell RNA seq as highly expressed in individual patient tumour cells [38]. However, we did not identify all the genes reported in this study in our cohort, which also confirms our observation of GEP heterogeneity among patient tumours.

The BIA-ALCL lymphoma cells expressed common hallmark genes (Fig. 3C); however, they also formed individual gene expression ALCL clusters for each patient, which were distinct from endogenous immune cells in the seroma (Fig. 2A, B). Furthermore, the BIA-ALCL cells have a distinct profile from other cutaneous T cell lymphomas (MF and SS) and also normal CD4⁺ T cells from patients with T cell lymphomas (Fig. 3D, E). There was no evidence that the malignant T cells in BIA-ALCL were clonally related to normal T cells present in the same seroma. Taken together, our findings suggest that BIA-ALCL cells were generated from an adaptive T cell which infiltrated the breast implant associated tissue, may have engaged with antigen (although commonly there is no functioning TCR evident by the time BIA-ALCL is diagnosed), and subsequently underwent malignant transformation to ALCL. However, there is no evidence that antigen interaction was alone causative of the malignant transformation.

In a prior study, textured implants from BIA-ALCL patients showed biofilm development enriched for specific bacterial species [12–14]. Whilst this is interesting, it is not clear how the bacterial antigens would drive ALCL proliferation as most BIA-ALCL cells do not express a TCR on the cell surface and lack the ability to respond to antigen via peptide-MHC (Figs. 2–4). These observations suggest that by the time a patient presents with clinical symptoms, their disease has become independent of direct antigen stimulation. Rather the ALCL cells are either proliferating following intrinsic or autocrine signals. Alternatively, they may be proliferating following interaction with infiltrating immune or stromal cells, eg monocytes or DCs. In support of autocrine/paracrine signalling, we observed tumour-tumour interactions (Fig. 5C, D) involved in T cell co-stimulation, proliferation and cell survival. Our analysis also identified CXADR-FAM3C tumour-tumour interaction, which may contribute to activation of the PI3K-AKT signalling pathway. Indeed, aberrant activation of AKT signalling is a common feature across multiple subtypes of ALCL [22, 39]. In addition, we also derived tumour-pDC (CD40-TNFSFB) and myeloid DC-tumour (CD74-COPA/MIF, LGALS9-CD47/SLC1A5, SIRPA-CD47) and tumour-monocyte interactions (LTA-TNF family members TNFRSF14/ TNFRSF1A/ TNFRSF1B) which have the potential to drive tumour cell proliferation and pro-survival signals in the TME.

As previously reported, we demonstrated that the BIA-ALCL seroma was enriched for the cytokines IL-10 and IL-13 [20, 27]. Moreover, we demonstrated that these ALCL cells also expressed the receptors for these cytokines suggesting an autocrine pathway. Indeed, CD30 expression was critical for p38 MAPK signalling and activation, and IL-13-STAT6 signalling. However, rather than proliferation, these cytokines are likely to reinforce an immune suppressive phenotype in the ALCL cells. Supporting this concept, the BIA-ALCL cells express PDL1 and secrete soluble PDL1 into the seroma (Fig. 5D). Of note, previous targeted sequencing studies identified activating JAK1 and STAT3 mutations for BIA-ALCL patients[8] which could support a role of intrinsic drivers of proliferation. However, as seen in T large granulocytic lymphocyte leukemia, these mutations may be just increasing the sensitivity to exogenous cytokines, rather than intrinsically driving tumour proliferation per se [40] (Table 1).

Our data shows the BIA-ALCL cell of origin (COO) is a mature CD4⁺ (or CD8⁺) T cell because the ALCL cells have rearranged TCRVβ genes, and no innate-like T cell signatures were identified in our study. The only innate cell type with potential to be the COO is the innate lymphoid cell type 2 (ILC2) which characteristically express GATA3, IL4, IL13, CD117, CD127 and CD25; however, these cells do not have rearranged TCRVβ [41]. What remains unknown is the antigenic target. Our ontogeny studies showed the CD4⁺ BIA-ALCL cells have a completely distinct GEP from that of normal CD4⁺ T cells. Interestingly, the SS and normal CD4⁺ T cells shared a more similar GEP than BIA-ALCL cells. Furthermore, the BIA-ALCL clones’ GEP profiles are markedly heterogeneous between patients suggesting that their development is not uniform, bordering on individually unique.

BIA-ALCL has been classified as a Type I ALCL based on its genetic subtype [42]. Previous RNA sequencing studies showed that Type I ALCLs are enriched for JAK-STAT3 signalling genes and TNF-NFκB signalling pathway genes [43–48]. Our GSEA analysis on invasive and non-invasive BIA-ALCL in comparison to cutaneous ALCL also demonstrated that pathways related to hypoxia, glycolysis, JAK STAT signalling, inflammatory response, and TNF-NFκB signalling were upregulated, suggesting a unique microenvironment that is distinct from other Type 1 ALCLs, such as cutaneous ALCL (Fig. 6A–C) [42]. Indeed, studies by other groups also identified hypoxia signatures, genes related to cell trafficking, and myeloid differentiation as being unique to BIA-ALCL [20,21]. When we compared invasive to non-invasive ALCL, we observed that the tumour cell signatures related to Myc targets and wnt signalling were further enriched and inflammatory response gene signatures were downregulated (Fig. 6D, E). Indeed, in a previous study, MYC was also found to be upregulated in BIA-ALCL compared with pcALCL, and was associated with the PI3K/AKT pathway and WT1 upregulation [22], which is known to modulate Wnt signalling [49], similar to our findings. This suggests that a lack of an inflammatory immune response could be contributing to further establishment of the tumours and spread beyond the breast capsule. Indeed, our spatial analysis confirmed this finding, where we observed that tumour-rich regions in invasive disease were devoid of immune cells such as CD8⁺ T cells with a high expression of PD-L1 on tumours when compared to early-stage non-invasive disease (Fig. 7C, D). A previous study identified PDL1 in 5 out of 9 cases with increase in PDL1 TAMs and PD1 TILs observed in both PDL1 positive and negative BIA-ALCL [50]. However, PDL1 positivity was observed in advanced cases. In summary, the tumour cells in invasive stages are most likely creating a microenvironment that is immunosuppressive and inhibiting the CD8⁺ T cells from responding through the PD1-PDL1 inhibitory axis, along with inhibitory cytokines such as IL-10, further reducing the infiltration of other cells such as M1 macrophages, B cells and DCs (Fig. 7A–H).

Taken together, our single cell RNA sequencing data gave us high resolution for not only the lymphoma cells, but also the infiltrating immune cells which are clearly very different between BIA-ALCL and benign seroma. This includes myeloid cells (DC subsets and monocytes) which could be responding to danger associated molecular patterns (DAMPs), which develop in the context of a highly-textured implant. By collating the single cell and seroma cytokine/chemokine data, gene expression and spatial analysis we now provide information on a complex network of tumour-tumour and tumour-immune cell interactions which provide proliferative and pro-survival signals to the ALCL cells, whilst also suppressing effector function by infiltrating CD8⁺ T cells. It appears that at some point after TCR rearrangement, autonomous growth occurs, likely contributed by mutational changes in key proliferative and survival pathways, which are in turn highly influenced by the tumor microenvironment. However, further studies are required to identify the transformation ‘hits’ (such as what leads to the development of mutations in the JAK-STAT pathway) and key ‘suppressor changes’ (that change in the TME) which lead to the development of BIA-ALCL.

In conclusion, our study is the first comprehensive multimodal study on BIA-ALCL and provides new insights into the TME of BIA-ALCL. Our findings suggest why the TME is crucial for BIA-ALCL survival and also identify potential pathways/targets for interventional therapy. This could include anti-PD1 or anti-PDL1 therapy for patients with high PDL1 expression or JAK inhibitors such as ruxolitinib or golidocitinib, or STAT degraders.

Author contributions

CDS, HMP and PJN developed the study concept and project structure. CDS performed the study, and analyzed the data in conjunction with NT, RQ and HDS. MMP, LM, TS, ET, PB and SL provided key technical and academic input into the manuscript. MM, NM, SO, HT, CVD, and AD all provided biospecimens for the study cohort and read and edited the manuscript. CDS, HMP and PJN wrote and edited the manuscript.

Funding

The research was funded by AbbVie. AbbVie was not involved in performing the work, interpreting the data or editing or writing the manuscript. CDS is supported by a Victorian Cancer Agency Fellowship ECRF22015. HMP is supported by an NHMRC Investigator grant 2017646.

Data availability

The sequencing data have been deposited in GEO under accession number GSE326046.

Competing interests

PJN received research funding from Roche Genentech, BMS, MSD, Compugen, Beigene and CRISPR Therapeutics. HMP received research funding from AbbVie for this study. The authors have no other conflicts of interest to declare.

Ethics approval

This study was approved by Peter MacCallum Cancer Centre human ethics committee (Reference no. 44068) and by Macquarie University human ethics committee (Reference No. 5201600427). Informed consent was obtained from all participating patients in this cohort study. All methods were performed in accordance with the approved ethics guidelines and regulations.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41375-026-03049-3.