Cell‐type‐specific tumour sensitivity identified with a bromodomain targeting PROTAC in adenoid cystic carcinoma

Salivary gland adenoid cystic carcinoma (ACC) is a rare malignancy with limited treatment options. The development of novel therapies is hindered by a lack of preclinical models. We have generated ACC patient‐derived xenograft (PDX) lines that retain the physical and genetic properties of the original tumours, including the presence of the common MYB::NFIB or MYBL1::NFIB translocations. We have developed the conditions for the generation of both 2D and 3D tumour organoid patient‐derived ACC models that retain MYB expression and can be used for drug studies. Using these models, we show in vitro and in vivo sensitivity of ACC cells to the bromodomain degrader, dBET6. Molecular studies show a decrease in BRD4 and MYB protein levels and target gene expression with treatment. The most prominent effect of dBET6 on tumours in vivo was a change in the relative composition of ACC cell types expressing either myoepithelial or ductal markers. We show that dBET6 inhibits the progenitor function of ACC cells, particularly in the myoepithelial marker‐expressing population, revealing a cell‐type‐specific sensitivity. These studies uncover a novel mechanistic effect of bromodomain inhibitors on tumours and highlight the need to impact both cell‐type populations for more effective treatments in ACC patients. © 2023 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of The Pathological Society of Great Britain and Ireland.


Introduction
Adenoid cystic carcinoma (ACC) is a rare malignancy that predominantly arises in the salivary gland.It is characterised by slow growth, extensive perineural invasion, frequent metastasis, and low survival rates [1,2].Standard-of-care treatment is limited to surgery at the primary site, which is often followed by post-operative adjuvant radiotherapy [3,4].Metastatic recurrence is common, and no systemic therapy has been found to be effective [5].This disease, therefore, represents an area of high unmet medical need and the identification of novel therapies is a priority.
Molecular profiling studies of ACC have shown a low and diverse mutation rate in ACC, with the only common genetic aberration being MYB or MYBL1 translocations, which are present in more than 50% of cases [6][7][8][9].Recent analysis of recurrent and/or metastatic samples has shown common genetic changes in members of the Notch signalling pathway and chromatin remodelling genes [10].High MYB or MYBL1 protein levels are a characteristic of ACC tumours [11], with translocations involving these genes being proposed to promote overexpression either by stabilising the protein or/and by providing novel enhancers for expression [12].Therapies targeting MYB expression and/or function have included bromodomain inhibitors and retinoic acid [12,13].However, these have not been particularly successful in clinical trials in ACC patients [5].
ACC tumours are typically biphasic, with co-existence of cells that express either myoepithelial or ductal/epithelial markers [14].Histopathological analysis of ACC shows three types of growth pattern: cribriform, tubular, and solid, with tumours usually having a mixed phenotype [15].The solid subtype is associated with high ductal marker expression and a more aggressive behaviour [16].The molecular mechanisms and underlying processes associated with the heterogeneous nature and dual cellular phenotype of ACC tumours remain unclear.
There is a paucity of clinically relevant models of ACC.There are few bona fide ACC cell lines, and most do not retain consistent MYB expression if a translocation is present [17,18].Patient-derived xenografts (PDXs) have been shown to be relevant in vivo preclinical models for many types of cancers, including ACC, as they retain the molecular characteristics, tumour histology, and heterogeneity of the original sample [19,20].Patient-derived tumour organoid models are emerging as important in vitro preclinical models, particularly as a reliable platform for therapeutic studies [21].These are characterised by growing from single stem/progenitor cells that self-organise into 3D structures that histologically resemble the original tumour and retain its molecular properties [22].
In this study, we have generated robust preclinical ACC models, including PDXs and patient-derived 2D and 3D cultures.We have used these models to identify bromodomain protein-degrader compounds as effective therapeutic compounds in vitro and in vivo, targeting the cell population within the tumour that expresses myoepithelial markers.

Study approval
All animal studies were conducted in accordance with protocols approved by the Institute of Cancer Research, London, UK and with the UK Animals (Scientific Procedures) Act 1986.Fresh tumour biopsies were obtained from 12 patients at The Royal Marsden under ethical approval and with patient consent.Matched peripheral blood samples were also acquired from all patients.All stored samples were recorded in accordance with the Human Tissue Act.
Detailed descriptions of methods used for cell culture, RNA, and protein analysis are provided in Supplementary materials and methods.

2D drug assays
Cultured 2D cells were trypsinised and replated in 96-well plates at 2,000 cells per well.One day after plating, JQ1 (Sigma, Gillingham, Dorset, UK) and dBET6 (Selleck Chem, Houston, TX, USA) were added to the 2D ACC medium (see Supplementary materials and methods).Cells were re-drugged on days 3 and 6 post-plating.Cell viability was measured on day 10 using the CellTiter-Glo Luminescent Cell Viability Assay (Promega, Madison, WI, USA) following the manufacturer's protocol and normalised to DMSOtreated controls.To establish the molecular effects of the drugs, ACC cells were grown in 2D for 1 week, following which they were drugged with IC50 concentrations of the drugs investigated for either 12 h (RNA) or 24 h (protein).

3D drug assays
Cultured organoids were dissociated into single cells following our passaging protocol as detailed in Supplementary materials and methods.They were plated (500 cells per well) in suspension in 96-well round bottom plates in 25 μl of Matrigel:medium (3:1).Three days after plating, medium was replaced with that containing dBET6.This was repeated on day 6, and cell viability was measured using the CellTiter-Glo 3D (Promega) cell luminescence assay on day 10.Drugged cell values were normalised to DMSO-treated control values.

In vivo drug assay
SG28 or SG32 PDX lines were used for in vivo drug studies.Mice were allocated to receive either vehicle (10% DMSO in 40% PEG 400, 5% Tween 80, 45% saline) or dBET6 (5 or 7.5 mg/kg) twice daily by i.p. injection.Tumour volume was measured twice weekly, and mouse body weight was recorded daily.Further details are provided in Supplementary materials and methods.

Progenitor/proliferation organoid assay
To establish the effect of dBET6 on progenitor capabilities of SG32 cells compared with proliferation, single SG32 cells were plated in Matrigel as described above.dBET6 was then added to the medium on either day 0 or day 3.After 10 days, organoids were imaged and the number of organoids per condition was counted and compared with a DMSO-treated control.
using SureSelect Human All Exome V4 reagents (Agilent, Santa Clara, CA, USA).Illumina paired-end libraries were prepared from the captured target regions and sequenced on a HiSeq 2500 (Illumina, San Diego, CA, USA), acquiring 2 Â 75 bp reads.DNA extracted from matched peripheral blood samples was also sequenced to identify germline mutations.Casava software (v1.8,Illumina) [23] was used for basecalling and to demultiplex the sequencing data.Sequences were output in FASTAQ format.Reads failing the Illumina chastity filter were removed before further analysis.BWA (v0.7.5a) [24] was used to align reads to the human reference genome (GRCh37 release 61).PCR duplicates were removed prior to further processing and variant detection.Variant calling was performed using GATK Best Practice Pipeline (v2.7.2) [25] from the Broad Institute with standard settings.Variants called in regions not covered by the capture probes were excluded, as were those with genotype qualities below 20 and those covered by fewer than 10 reads in either sample.
Short insertion/deletion (indel) mutations were selected from the complete set of variants called using the GATK unified genotyper based on differences in the variant allele fractions observed in the tumour and normal exome sequence data.Somatic single nucleotide variants (SNVs) were identified using MuTect (v1.1.4)[26] as detailed in Supplementary materials and methods.
ASCAT software (v2.1) was used to determine tumour purity, allele-specific ploidy, and to identify copy number aberrations [27].Further details are provided in Supplementary materials and methods.

RNA sequencing (RNA-seq)
RNA extraction, library preparation, and sequencing RNA was extracted from 2D cells and PDX samples as detailed in Supplementary materials and methods.Isolation of mRNA was carried out using the NEBNext Poly(A) mRNA Magnetic Isolation Module (New England Biolabs, Ipswich, MA, USA).This was followed by library preparation using the Ultra II Directional RNA Library Prep Kit for Illumina (New England Biolabs).

Sequencing of patient sample RNA
Sequencing was carried out using a HiSeq 2500 (v4 chemistry) (Illumina), acquiring a read length of 2 Â 75 bp at a 20 million (M) read depth per sample.TopHat2 (v2.0.7) [30] spliced alignment software was used to align reads to the reference genome (GRCH37) in combination with Bowtie2 (v2.0.6) [31] as detailed in Supplementary materials and methods.

Sequencing of RNA from 2D cells, PDX, and drug study samples
Sequencing was carried out using a NextSeq 2000 P2 acquiring either 2 Â 50 bp reads or 2 Â 100 bp (Illumina) at a read depth of 20 M per sample.Sequences were output in FASTAQ format.To compare gene expression between vehicle and dBET6-treated samples, read mapping was performed using the STAR alignment software (v.2.7.6a) [32] to Ensembl Human and Mouse reference genomes (Ensembl Human reference genome version GRCh38.92 and Ensembl Mouse reference genome version GRCm38.83).Once the reads were mapped, the XenofilteR R package [33] was used to remove mouse reads from human alignments.After the filtering, HTSeq-count (HTSeq v2.0.2) [34] was used to count the number of reads mapping unambiguously to genomic features in each sample.

Differential expression between PDX samples
Differential gene expression of the ACC samples was normalised in two different ways.Firstly, all ACC samples were normalised to published raw sequencing data from five normal salivary gland samples [35].Secondly, each ACC sample was normalised to the sample SG0043, acinic cell carcinoma of the salivary glands.Enrichr was used for Kyoto Encyclopaedia of Genes and Genomes (KEGG) pathway analysis of differentially expressed genes (DEGs) [36,37].

Differential expression and pathway analysis of drug study samples
Differential expression analysis of the count data was performed in R using the Bioconductor package DESeq2 (v1.34.0) [38] as detailed in Supplementary materials and methods.

Statistics
Data are presented as mean ± SD unless otherwise stated.Statistical significance of differences between two groups was evaluated using two-tailed Student's t-tests, while statistical significance of differences among multiple groups was analysed by one-way ANOVA or two-way ANOVA as stated.In vitro drug-response curves were generated using non-linear regression.All statistical analyses were carried out using GraphPad Prism v9 software (GraphPad Software Inc., San Diego, CA, USA).p values less than 0.05 were considered statistically significant.

Generation of ACC PDX tumours
To generate ACC PDX tumours, we implanted patient salivary gland tumour samples subcutaneously into immunodeficient mice.Most samples were taken from metastatic sites such as the lung (for sample details Tumour ACC models predict cell-type-specific drug sensitivity see supplementary material, Figure S1A).Six implanted samples out of ten developed into tumours, which we were able to passage and use to generate established PDX lines (see supplementary material, Figure S1B for tumour growth curves).Consistent with published ACC datasets [7], whole exome sequencing on patient tumour samples showed a low mutation rate with few common genetic changes.Copy number loss at 6q24 was identified in three samples and loss at 12q13-14 in four tumours (Figure 1A and supplementary material, Tables S1 and S2).RNA-seq analysis of patient tumour samples showed transcripts with MYB and NFIB sequences in three of the samples and with MYBL1 and NFIB sequences in one sample, highlighting the presence of classical ACC translocations (Figure 1B).Cluster analysis showed that the ACC samples were more similar to each other than to an acinic cell carcinoma PDX (SG43), and tumours with MYB-NFIB translocations clustered together (Figure 1C).DNA sequencing and RT-PCR were performed to confirm that the genomic changes, including fusion transcripts, present in the tumour of origin were maintained in the PDX tissue (supplementary material, Figure S1C,D).Histological analysis of the PDX lines showed these had the same phenotypes as the tumour of origin, with evidence of cribriform and solid growth in all samples (Figure 1D; PDX tumour grading shown in supplementary material, Figure S1A) [39,40].MYB immunohistochemistry showed high expression in all samples except SG69, which harboured a MYBL1::NFIB translocation (Figure 1D).Staining for p63, a myoepithelial marker, and KIT, a ductal/epithelial marker, showed that all PDX lines, with the exception of SG69, showed a biphasic mode of growth (Figure 1D and supplementary material, Figure S2).These properties were maintained after passaging of the PDX lines, which was done at least four times for each line.

Development of a 2D patient-derived ACC model
To develop preclinical in vitro models that could be used for drug studies, we developed 2D growth conditions for cells derived from PDX tissue.As has been seen for other PDX tumours, the stromal tissue within the tumour was of mouse origin, which allowed us to separate this population and only grow tumour cells of human origin (Figure 2A).Although good 2D growth was observed in the cultures, low levels of MYB expression, including from the translocated allele, and of EN1, a potential MYB target in ACC [41,42], were found (Figure 2B,C and supplementary material, Figure S3).RNA-seq analysis of 2D-grown cells from one of the PDX lines, SG32, showed differing expression of several signalling pathways from the corresponding in vivo PDX tumour, including members of the TGFβ, BMPs, and WNT pathway (Figure 2D).Taking into account these differences, PDX-derived cells were grown in 2D with the addition of WNT3A or BMP7 or the TGFβ inhibitor A83-01.Of these, only A83-01 had an impact on MYB levels, with an increase in MYB and EN1 observed in treated samples (Figure 2E,F).This difference was also observed in other PDX lines (Figure 2G,H and supplementary material, Figure S3B), showing it to be a general effect.For PDX lines without a MYB fusion, we saw an effect of A83-01 on MYB expression in SG28 but not in SG27 (supplementary material, Figure S3).This suggests that the action of A83-01 on MYB expression is not specific to the translocated locus.Therefore, A83-01 was added to all growth media for ACC-derived cells.Staining of cells grown in 2D with p63 and KIT showed a bimodal pattern of growth, with cell clumps containing cells expressing either myoepithelial or ductal/epithelial markers (supplementary material, Figure S4A).

Patient-derived 3D models recapitulate ACC tumour phenotypes
Patient-derived 3D tumour organoid models have been shown to be faithful preclinical models for many different cancer types.We adapted several protocols from published work for growing organoids from normal salivary gland and other tumour-derived tissue [43,44] to establish growth conditions for PDX-derived ACC cells (Figure 3A).As seen for cells grown in 2D, the addition of A83-01 to the media did increase the levels of MYB and EN1 expression somewhat when ACC cells were grown in 3D (Figure 3B,C).In addition, the number and size of organoids were increased when cells from four PDX lines were grown in low oxygen conditions (Figure 3D,E).Using our established conditions, we were able to grow organoids from all six of our PDX lines, which showed different phenotypes, reflecting the original ACC tumour, and levels of MYB expression (Figure 3F,G).Staining of organoid sections from two PDX lines, SG32 and SG28, for the myoepithelial marker, p63, and the ductal/epithelial marker, KIT, showed the presence of both types of cells in proportions relative to their presence in the original PDX as determined by cell sorting experiments for KIT + and CD49f + cells (supplementary material, Figure S4B,C).

Increased sensitivity of ACC cells to proteolysistargeting chimera (PROTAC) bromodomain inhibitors
Using our 2D and 3D models, we performed drug studies to identify novel therapies for ACC patients.BET bromodomain inhibitors have been proposed to be good agents for ACC as they have been shown to suppress MYB function in acute myeloid leukaemia [45] and BRD4 has been suggested to bind to the superenhancers found in ACC tumours [12].We found some sensitivity in 2D cultures to the bromodomain inhibitor JQ1, with one line, SG32, being the most sensitive (Figure 4A).To potentially increase the effective bromodomain inhibition, we used the degrader compound dBET6 [46].Drug sensitivity was greatly increased for all lines when dBET6 was used (Figure 4B) (JQ1 IC 50 range = 96 nM-5 μM; dBET6 IC 50 range = 25-135 nM).A similar effect for dBET6 was seen when the drug assay was performed in ACC samples grown in 3D (Figure 4C).As expected, treatment with dBET6, but not JQ1, led to the 40 AJ Rose, MM Fleming et al Tumour ACC models predict cell-type-specific drug sensitivity downregulation of BRD4 protein levels (Figure 4D,E).RT-qPCR analysis showed reduced levels of MYB, EN1, and the BRD4 target MYC [47] in treated samples from two PDX lines, SG28 and SG32 (Figure 4F,G).These studies confirm that bromodomain inhibitors can inhibit growth and target MYB expression and function in ACC tumours with and without MYB translocations and suggest that the degrader compounds are more effective.

In vivo efficacy of dBET6 on ACC PDX tumours
In vivo studies with dBET6 were performed on two PDX lines, SG32 and SG28.Treatment with 5 or 7.5 mg/kg dBET6 inhibited SG32 PDX tumour growth while not affecting mouse weights (Figure 5A-C and supplementary material, Figure S5).Some in vivo growth inhibition was observed with SG28 tumours upon treatment; however, tumour size increased at later stages (supplementary material, Figure S6A,B).Short-term in vivo treatment of SG32 and SG28 tumours was performed to identify molecular mechanisms of tumour growth inhibition.Western blot data showed that treatment with dBET6 did not greatly affect BRD4 or MYB protein levels within tumours for either SG32 or SG28 (Figure 5D and supplementary material, Figure S5C).Staining of sections of treated tumours with an antibody to BRD4 did not reveal any major cell-type-specific effect of dBET6 on BRD4 levels (supplementary material, Figure S5D).Consistent with this, RT-qPCR analyses showed that the transcript levels of MYB, EN1, and MYC were also not significantly affected by dBET6 treatment of in vivo treated tumours (Figure 5E and supplementary material, Figure S5D).

dBET6 treatment induces cell-type-specific expression changes in ACC tumours in vivo
Although growth of SG32 tumours was inhibited by dBET6, a decrease in proliferating cells, as marked by Ki67 immunohistochemistry, was not observed in treated samples (Figure 5F).Cleaved caspase-3 staining did not show any differences between vehicle-and dBET6-treated samples, implying that treatment did not increase cell death (supplementary material, Figure S5D).To investigate the mechanism of tumour growth inhibition, RNA-seq was performed on shortterm dBET6-treated samples.Differential gene expression analysis showed a decrease in BRD4 targets such as HIF1A, TNS1, and SLIT3 (Figure 5G) [48,49].Tumour ACC models predict cell-type-specific drug sensitivity 43 Comparing the differential genes with 6,816 genes associated with MYB binding sites in ACC determined by ChIP studies [12] showed an overlap of 1,344, suggesting that the expression of some MYB targets had been affected by dBET6 treatment (Figure 5H).The most striking observation was the presence of cell-type markers within the differential gene list.
Comparison of the differential genes with markers identified in published single-cell analysis studies on ACC samples [50,51] showed a general increase in ductal/ epithelial-type markers such as KIT, and a decrease of myoepithelial markers such as ACTA2 and TP63 upon dBET6 treatment (Figure 5I,J and supplementary material, Table S3).Confirmation of this result was seen through KIT immunohistochemistry on treated samples that showed an increase in cells expressing this marker in areas between the pseudocysts found in cribriform ACC (Figure 5K).

Cell-type-specific precursor abilities are affected by dBET6 treatment
To understand the effect of dBET6 on ACC tumours that would lead to growth inhibition, we used our 3D organoid model system.Treatment with dBET6 from day 0, when single-cell progenitors give rise to organoids, showed higher growth inhibition than treatment at day 3, when multicellular organoid structures have formed at concentrations below the IC 50 (Figure 6A,B).This suggests that dBET6 is targeting progenitor abilities in ACC cells.To investigate whether this effect was cell-type-dependent, we isolated cells expressing CD49f, which has been proposed to be a myoepithelial marker [50], and cells expressing KIT, a ductal/epithelial marker [52], and generated cell-type-specific organoid cultures.Confirmation of the relative purity of our selected cells was performed using immunohistochemistry and RT-qPCR (supplementary material, Figure S7).Treatment with dBET6 on the cell-specific 3D organoid cultures revealed a higher impact on the growth of CD49f-positive-derived organoids than the KIT-positive-derived population.This effect was seen for cells derived from both SG32 and SG28 PDX samples (Figure 6C,D).Analysis of RNA-seq data from tumour samples showed that SG32 had a lower expression of ductal/epithelial markers than SG28, whereas myoepithelial markers were higher in SG32 samples (Figure 6E).Confirmation of these results by immunohistochemistry showed a larger population of ductal/epithelial markerexpressing cells in SG28 (Figure 6F).These data suggest that dBET6 preferentially targets progenitor cells expressing myoepithelial markers and may explain the increased sensitivity of SG32 compared with SG28, which has a higher number of relatively resistant cells expressing ductal/epithelial markers.

Discussion
Few robust preclinical models of ACC exist, and this has significantly held back research on this disease of high unmet clinical need.In this study, we describe the development of patient-derived xenografts from metastatic tumours that maintain characteristic properties of ACC, such as histological appearance and MYB expression, for in vivo growth and in vitro 2D and 3D models.Our complementary and integrative approach allows for the identification of novel therapies for this treatmentrefractory disease using in vitro systems and their subsequent validation in vivo.Growth efficiency did vary between lines, with SG69 being the most problematic to grow both in vivo and in vitro.Cell passaging for in vitro models was limited, up to 4-5 times for 2D models, and most assays on organoids were performed without passaging.This was done to generate robust data, as cells tended to stop growing at later passages.Therefore, these should be considered transient in vitro models with the lines being maintained through PDX passaging.The limited abilities to passage and retain the original tumour properties in vitro are common for patient-derived tissue, and further development of methodologies, such as culture media components and extracellular matrix composition, is required to achieve long-term growth of ACC cells in culture.
Identifying novel systemic therapies for ACC patients has been challenging.Our panel of ACC PDX lines add and complement the existing models available through support of the Adenoid Cystic Carcinoma Research Foundation for in vivo drug screening [20].However, the lack of in vitro models severely affects investigative Tumour ACC models predict cell-type-specific drug sensitivity approaches with a higher throughput.This has led to a historical hit-and-miss development of therapies based on rationale with limited preclinical data that have failed in clinical trials in ACC patients [53].Indeed, we have found limited sensitivity to many compounds that have been used in patients in our in vitro systems and therefore have focused on those where we see an appreciable effect.The ability to culture ACC cells in vitro allows the screening of multiple drug compounds to identify candidate novel therapies for this disease.
The high levels of MYB expression in ACC implies that it could be a good drug target, although evidence that tumours are truly dependent on MYB function for growth and survival is somewhat lacking.In the absence of developed MYB-specific inhibitory compounds, BET bromodomain inhibitors have been proposed to be effective in some ACC tumours [12].Small-molecule pan-BET inhibitors are currently in clinical trials but present significant challenges, such as modest (at best) clinical efficacy, drug resistance [54], and toxicity.Protein degrader compounds have shown great clinical promise and have been recently optimised for in vivo studies [55].BET bromodomain degraders were found to have a more profound anti-proliferative effect than the corresponding pharmacological inhibitors [56].This is consistent with the in vitro studies presented, which showed that dBET6

46
AJ Rose, MM Fleming et al impacted cell survival to a greater extent than JQ1 in ACC models.Our in vivo studies showed that the major effect of dBET6 was on a subset of cells within the tumours, namely the progenitor cell compartment.Our molecular analysis suggests that compounds like dBET6 can have anti-tumour effects at dose levels that do not achieve high target protein degradation, which would potentially lead to less treatment-related toxicity.Nevertheless, we cannot exclude that an off-target effect could contribute to drug action.The biphasic nature of ACC tumours is a pathognomonic feature of their histopathology.The maintenance of two cell-type populations during tumour progression suggests that they are interdependent for tumour growth and survival.The role of MYB in this interdependence is not clear as it is expressed in both populations.Our in vivo dBET6 studies show that inhibition of BRD4 leads to an increase in cells expressing ductal/epithelial markers within tumours.Whether this is due to the induction of transdifferentiation of myoepithelial marker-positive cells or a specific cell-type increase in proliferation of ductal/epithelial-positive cells is not clear, although high levels of Ki67 were not particularly observed in interstitial regions in treated samples.Transdifferentiation between compartments has been reported in ACC cells [50].
The data from this study show that dBET6 treatment is most effective in cells expressing myoepithelial markers within ACC tumours.Therefore, tumours that have a higher proportion of these cells, such as SG32 when compared with SG28, might be more sensitive to this compound.The status of p63 expression, a myoepithelial marker in ACC, has been proposed to be used, together with MYC, to subgroup ACC patients [57].Our results are consistent with tumours with high p63 levels being more sensitive to dBET6.The cell-type-specific effect might explain why some therapies have not been as effective as expected in clinical trials on ACC patients.The dual and interdependent nature of the cellular constituents of ACC tumours, and the potential mutual repopulation of one compartment from the other, makes the development of therapies particularly challenging.For example, a drug therapy that selectively depletes one cell type is likely to be unsuccessful if cells from the resistance, surviving cell compartment can repopulate the depleted population.Our studies suggest that combination therapies with compounds that target both cell populations should be explored.Choosing agents like dBET6, which can act at lower concentrations, might be a useful way to reduce the increased toxicity when combinations are considered.

Figure 1 .
Figure 1.Genetic and histological analysis of ACC tumours.(A) Summary of identified mutations from ACC sample whole-exome DNA sequencing.Corresponding pathways are indicated on the left of the table.L1 indicates an MYBL1 fusion.(B) Diagram showing the wild-type MYB, MYBL1, and NFIB proteins with conserved domains indicated, and the fusion proteins found in each translocated sample.The fusion proteins for each ACC sample are depicted with the total length of MYB/MYBL1 indicated in the right-hand column, plus the additional length of NFIB.(C) A cluster plot to show the distances in similarity between each patient's ACC sample based on RNA-seq data.Distances were calculated by the Euclidean method.(D) FFPE sections of ACC PDX lines stained with H&E and for MYB, p63, and KIT by immunohistochemistry.

Figure 2 .
Figure 2. Generation of a 2D ACC model.(A) Schematic of the experimental procedure for the generation of a 2D cell model from PDX tissue.PDXs are dissociated to single cells before mouse stromal cells are removed and cells are plated.Created with BioRender.com.(B) Western blot for MYB expression in SG32 PDX samples compared with SG32 cells grown in 2D.B-ACTIN is a loading control.(C) RT-qPCR comparison of MYB and the downstream target of MYB; EN1 gene expression in SG32 PDX samples and SG32 2D cells.(D) Differential expression of genes involved in several signalling pathways between SG32 2D cells and PDX.(E) Changes to the expression of MYB and EN1 in SG32 2D cells following the addition of BMP7, WNT3a, and the TGFβ inhibitor A83-01 to ACC 2D medium established by RT-qPCR.(F) Western blot demonstrating MYB expression in SG32 2D cells following the addition of A83-01 to media.B-ACTIN is a loading control.(G) Changes to the expression of MYB and EN1 in SG37 2D cells following the addition of BMP7, WNT3a, and the TGFβ inhibitor A83-01 to ACC 2D medium established by RT-qPCR.(H) Western blot demonstrating MYB expression in SG28 2D cells following the addition of A83-01 to medium.B-ACTIN is a loading control.Data in C, E, and G are expressed as mean ± SD and statistical analyses were performed using two-way ANOVA with Šidák's multiple comparisons test; ***p < 0.0005, ****p < 0.0001.Only statistically significant differences are annotated.

Figure 3 .
Figure 3. Generation of a 3D ACC model.(A) Schematic of the experimental procedure for the generation of a 3D cell model from PDX tissue.PDXs are dissociated to single cells before mouse stromal cells are removed and ACC cells are plated in Matrigel.Created with BioRender.com.(B and C) Changes to the expression of MYB and EN1 in SG28 (B) and SG32 (C) organoids following the addition of the TGFβ inhibitor A83-01 to ACC 3D medium established by RT-qPCR.(D) Increased organoid growth following incubation at low oxygen concentration.Organoids for four lines were grown in 20% O 2 concentration and 2% O 2 for 10 days, following which the number of organoids that grew for each condition was counted.(E) Brightfield images of the differential growth of organoids at different oxygen concentrations.(F) FFPE sections of SG28, SG32, and SG69 stained with H&E comparing PDX tissue with organoids of the same sample.Brightfield images of the organoid lines are also shown.(G) FFPE PDX sections stained by immunohistochemistry for MYB compared with organoids from the corresponding lines stained for MYB.Data in B, C, and E are expressed as mean ± SD and statistical analyses were performed using two-way ANOVA with Šidák's multiple comparisons test; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.Only statistically significant differences are annotated.

Figure 4 .
Figure 4. PROTAC inhibitor of BRD4, dBET6, demonstrates efficacy in ACC in vitro models.(A and B) Four PDX lines grown in 2D were drugged with the BRD4 inhibitor JQ1 (A) and the BRD4 PROTAC dBET6 (B) over 10 days and cell viability was measured using CellTiter-Glo.(C) SG28 and SG32 organoids were drugged for 10 days with increasing concentrations of dBET6, following which cell viability was measured using 3D CellTiter-Glo.(D) Western blot for BRD4 expression following addition of IC 50 concentrations of JQ1 and dBET6 to SG28 cells grown in 2D for 24 h.B-ACTIN is a loading control.(E) Western blot for BRD4 expression following addition of IC 50 concentrations of JQ1 and dBET6 to SG32 cells grown in 2D for 24 h.B-ACTIN is a loading control.(F and G) Differential expression of MYB, EN1, and BRD4 target MYC in SG28 2D cells (F) and SG32 2D cells (G) following 12 h drug treatment established by RT-qPCR.Each drug was used at IC 50 concentrations for the relevant ACC sample.Data in A, B, and C are expressed as mean ± SD with curves generated using non-linear regression.Data in F and G are expressed as mean ± SD and statistical analyses were performed by two-way ANOVA with Šidák's multiple comparisons test; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.Only statistically significant differences are annotated.

Figure 5 .
Figure 5. dBET6 inhibits SG32 tumour growth in vivo.(A) Average tumour volume of mice treated with vehicle (n = 8) compared with mice treated with 5 mg/kg dBET6 (n = 6) over 30 days.(B) Comparison of the change in tumour volume over the experiment in vehicle-and dBET6-treated arms.(C) Mean body weights of mice in both arms of the study.(D) Western blots for BRD4 and MYB expression in 4-day shortterm vehicle-and dBET6-treated tumours.B-ACTIN is a loading control.(E) Gene expression RT-qPCR analysis for MYB, EN1, and MYC of vehicle-treated tumours compared with short-term dBET6-treated tumours.(F) FFPE sections of short-term treated tumours were stained for Ki67 by immunohistochemistry. Quantification of the percentage of Ki67-positive nuclei from three independent areas on each of three tumours in both arms is shown.(G) Heatmap showing the top 100 differentially expressed genes between vehicle-treated and dBET6-treated tumours.(H) A comparison of the differentially expressed genes in dBET6-treated tumours and previously published ChIP-seq data of MYB bound genes.(I and J) Differential expression between vehicle-and dBET6-treated tumours of previously published ductal-like (I) and myoepithelial-like (J) gene sets from a single-cell analysis study in ACC.The ductal-like cell marker KIT and myoepithelial-like cell markers TP63 and ACTA2 are indicated.(K) Increase in the number of cells expressing the ductal marker KIT in short-term dBET6-treated tumours was established by immunohistochemical staining of FFPE sections.Data in A, B, D, and F are expressed as mean ± SD.Statistical analyses in A were performed using an ANOVA; ****p < 0.0001.Statistical analysis in B was performed using a one-way ANOVA with Dunnett's multiple comparisons test; ****p < 0.0001.Only statistically significant differences are annotated.

Figure 6 .
Figure 6.ACC 3D model indicates that dBET6 inhibits ACC progenitor cell abilities and reveals a cell-type-specific sensitivity.(A) Schematic of the experimental procedure to determine differential effects of dBET6 on ACC organoid progenitor and proliferation abilities.ACC single cells are plated in Matrigel and drugged with dBET6 on day 0 for the progenitor assay and on day 3 for the proliferation assay.Organoid development is recorded on day 10 and compared with control treated organoids.Created with BioRender.com.(B) Brightfield images and quantification of the comparison of the number of organoids that develop following dBET6 treatment on day 0 or day 3. (C and D) Differential sensitivity of cells expressing the myoepithelial marker CD49f and cells expressing the ductal cell marker KIT to dBET6 treatment from day 0 in SG32 organoids (C) and SG28 organoids (D).(E) Differential gene expression of ductal-like cell markers and myoepithelial-like cell markers between SG28 and SG32 PDXs based on RNA-seq.(F) Differential protein expression of KIT and p63 between SG28 and SG32 PDXs was confirmed using immunohistochemistry. Data in B, C, and D are expressed as median and statistical analyses were performed by two-way ANOVA with Šidák's multiple comparisons test; ***p < 0.001, ****p < 0.0001.Only statistically significant differences are annotated.