Ex Vivo drug sensitivity in patient-derived 3D cultures in acromegaly and its association with clinical predictors

Abstract

Purpose

Personalized therapy in acromegaly is limited by interindividual variability in drug responses and the lack of robust markers predicting tumor shrinkage, rather than biochemical control alone. To test whether ex vivo drug-induced viability changes in patient-derived 3D cultures (Pd3D) of GH–secreting pituitary adenomas reflect tumor cell-intrinsic pharmacological sensitivity and align with established clinical predictors.

Methods

Spheroid-based Pd3D cultures were established from 27 patients with acromegaly. Cultures were exposed to octreotide, cabergoline, pasireotide, or vehicle control. We assessed cell viability changes; sample-level responder status (viability reduction vs vehicle, p < 0.05); and associations between responder status and known predictive markers, including clinical characteristics, MRI findings, dynamic drug tests, and pathological features. In 6 cases, AI-based digital image analysis quantified pre- and post-treatment SSTR2 expression in liquid-based cytology (LBC).

Results

All agents modestly reduced median cell viability (84–86%, p < 0.05), with responder rates of 33–40%. Concordance with established predictors was observed: octreotide responders correlated with T2 hypointensity (88% vs 44%, p = 0.04); cabergoline with positive bromocriptine tests (100% vs 45%, p = 0.03); and pasireotide with sparsely granulated patterns (64% vs 19%, p = 0.04). AI-based dynamic analysis demonstrated that ex vivo responders showed relatively stable SSTR2 expression after treatment, whereas nonresponders exhibited marked depletion.

Conclusion

Spheroid-based Pd3D ex vivo viability assays revealed modest but significant cohort-level effects and sample-level concordance with clinical predictors, suggesting its potential utility as an exploratory model. Additionally, AI-based quantification of SSTR2 dynamics captured functional receptor shifts.

Supplementary Information

The online version contains supplementary material available at 10.1007/s11102-026-01735-0.

Introduction

Acromegaly imposes a significant disease burden through excess systemic growth hormone (GH)/insulin-like growth factor-1 (IGF-1) and local tumor mass effects [1]. Although biochemical normalization is essential for improving survival, 40–60% of patients require adjuvant medical therapy post-surgery [2, 3]. Notably, biochemical control and tumor remission are not always achieved concurrently in clinical practice, and marked interindividual variability in drug response complicates therapeutic decision-making [4]. This trial-and-error approach places a considerable burden on patients and underscores the need to optimize therapeutic pathways [5–8]. While predictors of biochemical control have been extensively investigated, robust markers that directly forecast tumor shrinkage remain limited [9–11].

Across oncology, three-dimensional (3D) tumor culture systems have gained attention as tools to predict drug sensitivity because they mimic tumor architecture, cell–cell and cell–matrix interactions, and physiologically relevant nutrient and oxygen gradients, thereby providing more representative response readouts [12–16]. In contrast, traditional two-dimensional (2D) primary cultures often suffer from loss of native architecture and functional receptor expression [17]. Among 3D systems, organoids and spheroids represent two widely used but conceptually distinct models. Organoids typically self-organize and retain stem or progenitor compartments, whereas spheroids generally arise from the aggregation of primary cells or cell lines and lack self-organization. Although organoid models are valued for their higher intra-and inter-assay reproducibility and their ability to support long-term maintenance of lineage programs, their widespread application is constrained by higher cost and longer setup and expansion times. In the field of 3D models for pituitary adenomas, it remains to be established whether in vitro readouts accurately reflect actual clinical drug sensitivity. Furthermore, the construction and maintenance of organoids require considerable costs and highly specialized technical expertise, restricting their accessibility in routine clinical practice [17–20].

To address these challenges, we previously established a rapid and cost-conscious spheroid-based patient-derived 3D (Pd3D) cultures system from pituitary adenomas to perform ex vivo drug testing; in a small number of cases, ex vivo readouts from this platform aligned with clinical outcomes, including hormonal response and volumetric tumor shrinkage, although further research is required [21–23]. Although this approach may not fulfill all the rigorous criteria of organoids, we hypothesized that it represents a robust tumor spheroid model that preserves tumor-intrinsic drug responsiveness and is well suited for clinically oriented drug screening.

In the present study, we applied this Pd3D workflow to a consecutive series of patients with acromegaly. Using Pd3D cultures, we quantified drug-induced changes in cell viability and compared these readouts with established predictors of biochemical response to medical therapy. In addition, we examined the relationship between ex vivo drug responsiveness and somatostatin receptor (SSTR) expression levels assessed in liquid-based cytology (LBC) preparations, which provide a robust platform for immunostaining and ancillary molecular analyses in rare endocrine tumor samples [24]. To ensure objective quantification and reproducibility, we employed an Artificial Intelligence (AI)-based digital image analysis system [25]. This integrated approach allowed us to explore the feasibility and characteristics of ex vivo functional assays in the context of conventional hormonal markers. An overview of the study workflow is shown in Online Resource 1.

Materials and methods

Participants and data acquisition

Between November 2023 and November 2024, 27 consecutive patients with acromegaly who underwent transsphenoidal surgery at Moriyama Memorial Hospital were enrolled, and tumor specimens were collected intraoperatively. Acromegaly was diagnosed according to international guidelines [2, 3]. Clinical variables were retrospectively extracted from electronic medical records with a focus on previously proposed predictors of pharmacological responses. Extracted data included age, sex, serum GH levels, and age- and sex-adjusted IGF-1 standard-deviation scores (SDS). Additional parameters included basal and nadir GH levels during a 75-g oral glucose tolerance test (OGTT), the timing of the GH nadir, and results of octreotide and bromocriptine tests. The octreotide test was considered positive when GH decreased by ≥ 75% after a subcutaneous 100 µg [26]. The bromocriptine test was considered positive when GH decreased by ≥ 50% after oral administration of 2.5 mg [27]. Pituitary MRI variables included maximum tumor diameter, Knosp grade, and qualitative T2-weighted imaging (T2WI) signal intensity. MRI assessments were performed by endocrinologists, neurosurgeons, or radiologists. Histopathology variables, evaluated by an experienced pathologist, included GH immunohistochemistry, Ki-67 labeling index, semi-quantitative immunohistochemistry (IHC) expression scores for somatostatin receptor subtype 2 (SSTR2) and subtype 5 (SSTR5), and granulation pattern (densely or sparsely granulated) [28]. Owing to the retrospective design, not all clinical, radiological, and histopathological variables were available for every case.

Patient-derived 3D (Pd3D) cultures

Spheroid-based Pd3D cultures were established as previously described [21–23]. Freshly resected tumor tissue was immediately placed on ice in phosphate-buffered saline (PBS; Gibco, Waltham, MA, USA) and enzymatically dissociated in Dulbecco’s Modified Eagle Medium (DMEM; Gibco, Waltham, MA, USA) containing 0.3% bovine serum albumin (Sigma-Aldrich, St. Louis, MO, USA), 0.35% collagenase (FUJIFILM Wako, Osaka, Japan), 0.15% hyaluronidase (FUJIFILM Wako, Osaka, Japan), and 0.1% Y-27632 (FUJIFILM Wako, Osaka, Japan). Cell viability was in all patient samples confirmed using Trypan blue exclusion prior to plating to ensure that 10,000 viable cells were seeded per well in domes of growth factor–reduced phenol red–free Matrigel (Corning, NY, USA) in white flat-bottom 96-well plates (Thermo Fisher Scientific, Waltham, MA, USA) at a density of 10,000 cells per well. Matrigel domes were polymerized with the plates inverted at 37 °C for 10 min to promote gravitational settling and cell aggregation at the apex of each dome, thereby forming spheroids. Each well then received 50 µL of DMEM supplemented with 10% fetal bovine serum (FBS; Gibco, Waltham, MA, USA), 0.1% Y-27632 and penicillin–streptomycin (Gibco, Waltham, MA, USA). Plates were incubated at 37 °C in 5% CO₂ for 24 h to allow cells to recover from the stress of enzymatic digestion.

Drug exposure and cell viability readout

After 24 h of initial culture, the medium was replaced with fresh medium containing either vehicle (DMSO; Nacalai Tesque, Kyoto, Japan; final DMSO ≤ 0.1% [v/v]) or one of the following drugs at 10 nM (a standard concentration well-established in previous in vitro studies of primary human pituitary adenomas for evaluating receptor-specific responses without non-specific cytotoxicity [29–31]): octreotide (Selleck Chemicals, Houston, TX, USA), cabergoline (Tocris, Bristol, UK), or pasireotide (ChemScene, Monmouth Junction, NJ, USA). Each condition was tested in three to five replicate wells per sample, depending on cell yield. A single medium refresh was performed 72 h after drug addition, and cultures were maintained for 7 d from treatment initiation. Representative images at this endpoint are shown in Online Resource 2.

Cell viability was quantified with the RealTime-Glo™ MT Cell Viability Assay (Promega, Madison, WI, USA) according to the manufacturer’s instructions. This assay was selected because it measures the reducing potential of metabolically active cells without requiring cell lysis, thereby allowing continuous, sensitive monitoring of viability while preserving the intact 3D spheroid structure. Luminescence was recorded immediately before drug exposure (baseline) and at the end of treatment using an EnSpire Multimode Plate Reader (PerkinElmer, Waltham, MA, USA). For each well, endpoint luminescence was divided by its corresponding baseline value and multiplied by 100 to yield relative viability (%). For each treatment arm within a patient, the median relative viability (%) across replicate wells was used for analysis.

At the patient-sample-level, a drug was classified as yielding an ex vivo response (“responder”) if the median relative viability (%) in the treatment arm was significantly lower than that in the contemporaneous vehicle control at the endpoint (two-sided p < 0.05); otherwise, the sample was considered a “nonresponder.”

Immunocytochemistry of dissociated cells from Pd3D using the BD SurePath™

Freshly collected basal cell suspensions and 7-day cultured cells (harvested from Matrigel domes following vehicle or octreotide treatment using Cell Recovery Solution [Corning, NY, USA] according to the manufacturer’s instructions) were separately subjected to immunocytochemistry. All specimens were processed under identical conditions using the BD SurePath™ liquid-based cytology system (BD Biosciences, Franklin Lakes, NJ, USA).

The cell suspensions were transferred into BD CytoRich™ Red preservative solution (BD Biosciences, Franklin Lakes, NJ, USA) and fixed at room temperature for at least 30 min in accordance with the manufacturer’s protocol. Fixed samples were centrifuged at 2,000 rpm for 3 min, the supernatant was discarded, and the resulting cell pellet was gently resuspended. An appropriate volume of the suspension was loaded into a BD SurePath settling chamber mounted on a BD SurePath-precoated glass slide (BD Biosciences, Franklin Lakes, NJ, USA). After allowing the cells to settle for 3 min, the chamber was removed and the slide was immediately transferred to 95% ethanol for additional fixation.

Immunocytochemistry was performed using a BOND-III automated immunostainer (Leica Biosystems, Nussloch, Germany) according to the manufacturer’s default immunohistochemistry program. Heat-induced antigen retrieval was performed using Epitope Retrieval Solution 1 (ER1; pH 6.0; Leica Biosystems, Nussloch, Germany) for 10 min. Because BD SurePath cytology slides were used rather than paraffin-embedded sections, deparaffinization steps were omitted.

Primary antibodies for somatostatin receptor included rabbit monoclonal anti-SSTR2 (clone UMB1, Abcam, Cat# ab134152, RRID:AB_2737601, Cambridge, UK; 1:1,500) and rabbit monoclonal anti-SSTR5 (clone UMB4, Abcam, Cat# ab109495, RRID:AB_10859946, Cambridge, UK; 1:20). Antibodies were applied according to the manufacturer’s datasheets, and a polymer-based horseradish peroxidase detection system supplied by the instrument’s manufacturer was used. Identical antibody panel and staining conditions were applied to basal, vehicle-treated, and octreotide-treated samples.

Digital image analysis for SSTR expression quantification

In a pilot subgroup, cytological whole-slide images (WSIs) derived from six tumor samples (Samples A–F) were analyzed. For each case, treated specimens were prepared under three conditions: immediately after dissociation (basal), and after 7-day exposure to vehicle or octreotide (10 nM). Digital image analysis was performed using the open-source software QuPath (version 0.5.1) [25]. To quantify the immunostaining intensity, color deconvolution was applied as a preprocessing step using standard optical density vectors: hematoxylin (nuclear stain) at (0.651, 0.701, 0.290), DAB (0.269, 0.568, 0.778), and residual (0.633, −0.713, 0.302).

Tumor cells were identified using a cell detection algorithm with the following parameters: pixel size, 0.5 µm; background radius, 8 µm; median filter radius, 1.5 µm; sigma, 1.8 µm; and threshold, 0.15. The maximum background intensity was set at 2. Detection was restricted to cell areas between 15 µm² and 150 µm², to exclude debris and non-specific aggregates. Automated detections were manually reviewed to remove dust particles, inflammatory cells, and doublets. Regions of interest (ROIs) were adjusted according to cellularity: the central area was analyzed for specimens with high cell density, whereas the entire slide area was evaluated for samples with low cell counts.

Cells were classified into four intensity categories based on mean DAB optical density across the entire cell area. In the absence of internal calibration controls for LBC specimens, standard intensity thresholds were utilized, referencing the established criteria for SSTR2 digital analysis [32]: negative (< 0.2, blue), 1 + (weak; 0.2–0.4, yellow), 2 + (moderate; 0.4–0.6, orange), and 3 + (strong; > 0.6, red) (Online Resource 3). Finally, the H-score (range: 0–300) was calculated for each sample using the standard formula: H-score = (1 × % of 1 + cells) + (2 × % of 2 + cells) + (3 × % of 3 + cells) [33]. For comparison, SSTR2 expression in corresponding formalin-fixed paraffin-embedded (FFPE) tissue sections was evaluated using the Volante score system (score 0–3), as previously described [34].

Statistical analysis

Continuous or ordinal variables were compared using the Mann–Whitney U test, and categorical variables were compared using Fisher’s exact test. Statistical significance was defined as p < 0.05. Data are presented as median with interquartile ranges (IQRs; 25th–75th percentiles). Analyses were performed using JMP Pro 18 software (SAS Institute Inc., Cary, NC, USA).

Results

Patient characteristics

Twenty-seven consecutive patients with acromegaly were included in this study (Table 1). The median age at surgery was 46.0 years [36.5–56.0], and 13/27 (48.1%) were female. At baseline, the median serum IGF-1 SDS was + 6.2 [+ 5.4 to + 7.4], and the median serum GH levels were 14.7 ng/mL [5.7–35.6]. An octreotide test was performed in 19 patients, of whom 15/19 (78.9%) met the predefined positivity criterion (≥ 75% GH reduction), with a median suppression of 82.6% [63.9–90.7]. A bromocriptine test was conducted in 18 patients, 12/18 (66.7%) were classified as positive, with a median suppression rate of 69.6% [39.5–88.1].

Table 1.

Baseline clinical characteristics of the study cohort

				available cases
Total number	(cases)	27
Baseline characteristics
Age at the time of surgery	(years)	46	[36.5–56]	27
Sex, female	(cases, %)	13 (48%)		27
Basal GH	(ng/mL)	14.7	[5.7–35.6]	27
Nadir GH on 75gOGTT	(ng/mL)	16	[6.5–33.2]	26
IGF-1 SDS		+ 6.2	[+ 5.4- + 7.4]	27
PRL	(ng/mL)	11.9	[7.4–19.1]	25
Dynamic endocrinological test
Octreotide suppression test, positive	(cases, %)	15 (78%)		19
Octreotide suppression test, suppression rate	(%)	82.6	[63.9–90.7]
Octreotide suppression test, nadir time	(hours)	4	[2–5]
Bromocriptine suppression test, positive	(cases, %)	12 (66%)		18
Bromocriptine suppression test, suppression rate	(%)	69.6	[39.5–88.1]
Bromocriptine suppression test, nadir time	(hours)	6	[4–6]
Pituitary MRI findings				27
Macroadenoma > 10 mm	(cases, %)	26 (96%)
Maximum tumor diameter	(mm)	19.5	[15.2–27.7]
Knosp grade, invasion	(cases, %)	7 (25%)
Knosp grade [0/1/2/3/4]	(cases)	[5/11/4/4/3]
MRI T2WI hypointensity	(cases, %)	16 (59%)
Histological findings				27
Granulated pattern, densely	(cases, %)	17 (62%)
SSTR2 [0/1/2/3]	(cases)	[1/1/8/17]
SSTR5 [0/1/2/3]	(cases)	[4/4/15/4]
Ki-67 labeling index	(%)	0.5	[0.3–1.2]

Data are presented as numbers (%) for categorical variables and as medians [25th–75th percentiles] for continuous variables. The numbers of patients with available data for each parameter are listed in the rightmost column

Abbreviations: GH growth hormone, OGTT oral glucose tolerance test, IGF-1 insulin-like growth factor-1, SDS standard deviation score, PRL prolactin, MRI magnetic resonance imaging, SSTR somatostatin receptor

On pituitary MRI, 26/27 (96.3%) patients had macroadenomas (> 10 mm). The median maximum tumor diameter was 19.5 mm [15.2–27.7]. Cavernous sinus invasion (Knosp grade 3–4) was observed in 7/27 (25.9%) patients, and MRI T2WI hypointensity was present in 16/27 (59.3%) patients. Histopathology analysis showed a densely granulated pattern in 17/27 (63.0%). The median Ki-67 labeling index was 0.5% [0.3–1.2]. Semi-quantitative immunohistochemical expression of SSTR2 tended to be higher than that of SSTR5 in the majority of evaluable samples. Data completeness varied across variables owing to the retrospective study design.

Overall, the characteristics of this cohort were broadly consistent with those reported in previous studies on acromegaly [35].

Association between known predictive markers and Pd3D cultures responsiveness

Across all 27 samples, all control cultures maintained robust viability from baseline to the endpoint (Online Resource 4). Each tumor sample was classified as an ex vivo responder or nonresponder in the Pd3D culture assay based on whether there was a statistically significant decrease in viability in the drug-treated arm compared to the vehicle control. To illustrate the magnitude of this effect, the median relative cell viability (%) at endpoint is shown for responders and nonresponders for all three agents: octreotide (84% [75–88] vs 100% [93–105], p < 0.01), cabergoline (84% [79–92] vs 94% [86–106], p = 0.04); and pasireotide (86% [85–89] vs 95% [88–100], p = 0.01) (Fig. 1). The proportion of responders was 9/27 (33%) for octreotide, 9/27 (33%) for cabergoline, and 11/27 (40%) for pasireotide.

Fig. 1.

Comparison of “responder” and “nonresponder” groups to each drug in patient-derived 3D cultures. Relative viability (%) at endpoint for “responder” and “nonresponder” groups for each treatment: octreotide, cabergoline, and pasireotide. Bars indicate medians (IQR 25th–75th percentiles). The number of samples in each group was as follows: octreotide (responder, n = 9; nonresponder, n = 18), cabergoline (responder, n = 9; nonresponder, n = 18), and pasireotide (responder, n = 11; nonresponder, n = 16). * p < 0.05, Mann–Whitney U test

Octreotide

Baseline characteristics, including age, sex, baseline GH level, and IGF-1 SDS, did not differ between the responder and nonresponder. Octreotide test results were also comparable between the two groups in the proportion positive rate, GH suppression magnitude, and time to nadir. On MRI, macroadenomas were present in 8/9 (88%) responders and in all 18/18 (100%) nonresponders. Maximum tumor diameter was 13 mm [13–27 mm] vs 21 mm [17–27] (p = 0.16), and cavernous sinus invasion (Knosp 3–4) was observed in 3/9 (33%) vs 4/18 (22%) (p = 0.65).

In contrast, MRI T2WI hypointensity, a previously reported marker of responsiveness to first-generation somatostatin receptor ligands (SRLs), was significantly more frequent among responders than among nonresponders (88% vs 44%, p = 0.04). Histopathological features, including granulation pattern and semi-quantitative SSTR2/SSTR5 expression scores, did not differ significantly between the two groups (p > 0.99, p = 0.11, respectively; Table 2a).

Table 2.

Comparative analysis of established predictors between Ex Vivo responders and nonresponders for (a) Octreotide, (b) Cabergoline, and (c) Pasireotide

(a) Octreotide
		responder	(n)	nonresponder	(n)	p value
Total number	(%)	9 (33%)		18 (67%)
Baseline characteristics
Age at the time of surgery	(years)	46 [40,52]	9	48 [32,59]	18	0.64
Sex	(female, %)	4 (44%)	9	9 (50%)	18	> 0.99
Basal GH	(ng/mL)	14 [4.0,29]	9	15 [7.1,40]	18	0.75
IGF-1 SDS		+ 6.1 [+ 5.6, + 6.4]	9	+ 7.0 [+ 5.3, + 7.4]	18	0.95
Dynamic endocrinological test
Octreotide suppression test, positive	(cases, %)	4 (80%)	5	11 (78%)	14	0.53
Octreotide suppression test, suppression rate	(%)	90.8 [80.8,94.4]	5	82.6 [53.9,89]	14	0.16
Octreotide suppression test, nadir time	(hours)	4 [4,6]	5	4 [2,4]	14	0.3
Pituitary MRI findings
Macroadenoma > 10 mm	(cases, %)	8 (88%)	9	18 (100%)	18	0.33
Maximum tumor diameter	(mm)	13 [13,27]	9	21 [17,27]	18	0.16
Knosp grade, invasion	(cases, %)	3 (33%)	9	4 (22%)	18	0.65
Knosp grade [0/1/2/3/4]	(cases)	[4/1/1/1/2]	9	[1/10/3/3/1]	18	0.59
MRI T2WI hypointensity	(cases, %)	8 (88%)	9	8 (44%)	18	0.04
Histological findings
Granulated pattern, densely	(cases, %)	6 (66%)	9	11 (61%)	18	> 0.99
SSTR2 [0/1/2/3]	(cases)	[0/0/2/7]	9	[1/1/6/10]	18	0.22
SSTR5 [0/1/2/3]	(cases)	[0/0/8/1]	9	[4/4/7/3]	18	0.11
(b) Cabergoline
		responder	(n)	nonresponder	(n)	p value
Total number	(%)	9 (33%)		18 (67%)
Baseline characteristics
Age at the time of surgery	(years)	43 [30,52]	9	49 [40,59]	18	0.17
Sex	(female, %)	5 (55%)	9	8 (44%)	18	0.69
Basal GH	(ng/mL)	12 [10,14]	9	24 [5.0,40]	18	0.44
IGF-1 SDS		+ 6.0 [+ 5.6, + 6.4]	9	+ 7.1 [+ 4.9, + 7.7]	18	0.62
PRL	(ng/mL)	13 [9.0, 24]	8	10 [7.4,15]	17	0.48
Dynamic endocrinological test
Bromocriptine suppression test, positive	(cases, %)	7 (100%)	7	5 (45%)	11	0.03
Bromocriptine suppression test, suppression rate	(%)	86.0 [69.6,89.8]	7	39.7 [20.6,73.2]	11	0.06
Bromocriptine suppression test, nadir time	(hours)	6 [5,6]	7	6 [3,6]	11	0.46
Pituitary MRI findings
Macroadenoma > 10 mm	(cases, %)	8 (88%)	9	18 (100%)	18	0.33
Maximum tumor diameter	(mm)	19.5 [13.5,23.8]	9	20.4 [16.3,27.9]	18	0.73
Knosp grade, invasion	(cases, %)	2 (22%)	9	5 (27%)	18	> 0.99
Knosp grade [0/1/2/3/4]	(cases)	[3/2/2/0/2]	9	[2/9/2/4/1]	18	0.74
MRI T2WI hypointensity	(cases, %)	7 (77%)	9	9 (50%)	18	0.23
Histological findings
Granulated pattern, densely	(cases, %)	5 (55%)	9	12 (66%)	18	0.68
SSTR2 [0/1/2/3]	(cases)	[0/1/1/7]	9	[1/0/7/10]	18	0.34
SSTR5 [0/1/2/3]	(cases)	[0/0/7/2]	9	[4/4/8/2]	18	0.03
(c) Pasireotide
		responder	(n)	nonresponder	(n)	p value
Total number	(%)	11 (40%)		16 (59%)
Baseline characteristics
Age at the time of surgery	(years)	46.0 [30.0,51.0]	11	49.5 [42.2,60.5]	16	0.13
Sex	(female, %)	6 (54%)	11	7 (43%)	16	0.70
Basal GH	(ng/mL)	13.1 [9.03,39.0]	11	15.4 [4.26,32.33]	16	0.53
IGF-1 SDS		+ 6.2 [+ 5.5, + 7.9]	11	+ 6.5 [+ 5.3, + 7.4]	16	0.94
Pituitary MRI findings
Macroadenoma > 10 mm	(cases, %)	11 (100%)	11	15 (93%)	16	> 0.99
Maximum tumor diameter	(mm)	18.1 [13.9,23.0]	11	21.0 [16.4,30.6]	16	0.41
Knosp grade, invasion	(cases, %)	2 (18%)	11	5 (31%)	16	0.66
Knosp grade [0/1/2/3/4]	(cases)	[3/5/1/1/1]	11	[2/6/3/3/2]	16	0.23
MRI T2WI hypointensity	(cases, %)	7 (63%)	11	9 (56%)	16	> 0.99
Histological findings
Granulated pattern, sparsely	(cases, %)	7 (64%)	11	3 (19%)	16	0.04
SSTR2 [0/1/2/3]	(cases)	[0/0/3/8]	11	[1/1/5/9]	16	0.31
SSTR5 [0/1/2/3]	(cases)	[1/2/6/2]	11	[3/2/9/2]	16	0.64

Data are presented as numbers (%) or medians [25th–75th percentiles]. Comparisons between groups were performed using the Mann–Whitney U test for continuous and ordinal variables and Fisher’s exact test for categorical variables

Abbreviations: GH growth hormone IGF-1 insulin-like growth factor-1, SDS standard deviation score, MRI magnetic resonance imaging, T2WI, SSTR, somatostatin receptor

Cabergoline

Baseline clinical characteristics were similar between cabergoline responders and nonresponders. In contrast, the bromocriptine test was positive in all responders but in less than half of nonresponders (100% vs 45%, p = 0.03), a difference that reached statistical significance. Neither the GH suppression rate nor the time to GH nadir during the bromocriptine test differed between the groups.

MRI-derived variables, including tumor size, Knosp grade, and T2WI signal intensity, were comparable between responders and nonresponders, as were granulation patterns (Table 2b). Notably, semi-quantitative SSTR5 expression was significantly higher in responders than in nonresponders (p = 0.03), whereas SSTR2 expression did not differ between groups (p = 0.34).

Pasireotide

Baseline characteristics and MRI findings did not differ between the groups. Tumors with sparsely granulated patterns were more common in responders than in nonresponders (64% vs 19%, p = 0.04), whereas SSTR2/SSTR5 expression scores did not differ between the groups (Table 2c).

AI-assessed temporal dynamics of SSTR2 expression in LBC specimens

An additional six patients with acromegaly were included for whom both ex vivo drug sensitivity assays and AI-based immunocytochemical analysis were performed. The basal SSTR2 H-scores generally corresponded to the diagnostic Volante scores obtained from the corresponding FFPE tissues (Table 3). Following octreotide treatment, distinct quantitative changes in SSTR2 expression were observed between responders and nonresponders. Ex vivo responders (Samples A, B, and C) showed relatively stable SSTR2 expression, in contrast to the marked depletion observed in nonresponders (Fig. 2a). Samples A and C exhibited increased H-scores (+ 30.8% and + 45.9%, respectively), accompanied by a shift in intensity fractions, in which negative (0) cells decreased, whereas 1 + (weak) and 2 + (moderate) positive cells increased (Fig. 2b). Sample B, which had the highest baseline expression, maintained a high proportion of 3 + (strong) cells (approximately 60%) despite a slight increase in negative (0) cells, distinguishing it from the depletion pattern observed in nonresponders (Fig. 2b, Table 3). In contrast, nonresponders (Samples D, E and F) exhibited a marked reduction in H-scores compared to vehicle-treated controls (Fig. 2a). Sample D showed a 30.3% reduction in H-score (167.0 to 116.4), with 3 + (strong) cells decreasing from 18.5% to 7.7%. Sample E showed a 60.8% decrease (145.9 to 57.1), characterized by a marked increase in negative 0 cells from 3.2% to 63.5% and a depletion of 3 + (strong) cells (Fig. 2b). Similarly, Sample F demonstrated a 44.7% reduction in H-score (131.7 to 72.8), with 3 + (strong) cells decreasing from 36.7% to 14.2% (Fig. 2b, Table 3).

Table 3.

Clinical characteristics and AI-based SSTR2 dynamics in the pilot subgroup (n = 6)

Sample			A	B	C	D	E	F
Status			responder	responder	responder	nonresponder	nonresponder	nonresponder
Age		(yo)	25	51	45	51	47	51
Sex			M	M	F	M	F	M
GH		(ng/mL)	13.0	48.1	81.2	3.3	49.8	11.6
IGF-1		(ng/mL)	616	775	896	367	316	797
IGF-1 SDS			+ 5.5	+ 8.1	+ 9.8	+ 3.9	+ 3.7	8.2
MRI	Size	(mm)	13	21	47	6	13	18
	Knosp grade		1	1	2	0	3	3
	T2WI intensity		Low	Low	Low	Low	High	High
FFPE	Volante SSTR2 score		3 +	3 +	2 +	2 +	3 +	2 +
LBC H-score	Basal		102.8	60.3	32.4	106	150.3	74.8
	Vehicle		104.6	245.8	43.6	167	145.9	131.7
	Octreotide		136.9	233	63.7	116.4	57.1	72.8
	delta H-score	(%)	30.8	−5.2	45.9	−30.3	−60.8	−44.7

Status refers to the classification of responders or nonresponders based on ex vivo assays. Sex was denoted as F (female) or M (male). The FFPE SSTR2 score indicated the conventional immunohistochemical grade (Volante score 0–3 +) assessed using formalin-fixed paraffin-embedded surgical tissue. H-scores (range, 0–300) were quantified using AI-based digital image analysis of the liquid-based cytology (LBC) specimens. The delta H-score (%) was calculated as the percentage change in the H-score after octreotide treatment relative to that of the vehicle control. GH, growth hormone; IGF-1, insulin-like growth factor-1; SDS, standard deviation score; MRI, magnetic resonance imaging; (T2WI); LBC, liquid-based cytology

Fig. 2.

Quantitative dynamics of SSTR2 expression in response to octreotide treatment. a Trajectories of SSTR2 H-scores. Individual lines represent the shift in H-score from vehicle-treated controls to octreotide-treated conditions for each case (Samples A–F). Green lines indicate ex vivo responders, and blue lines indicate nonresponders. The H-score (range 0–300) reflects the total intensity and quantity of receptor expression. b Shifts in SSTR2 staining intensity distribution. 100% stacked bar charts illustrate the percentage of cells in each intensity category (negative: 0; weak: 1 +; moderate: 2 +; and strong: 3 +) for each sample. Paired bars (Veh and Oct) show changes before and after treatment. Colors indicate receptor expression levels: gray (negative, 0), yellow (weak, 1 +), orange (moderate, 2 +), and red (strong, 3 +) Abbreviations: Veh, vehicle; Oct, octreotide

Although statistical significance could not be determined owing to the limited sample size, these findings suggest an association between the quantitative temporal dynamics of SSTR2 expression and drug responsiveness.

Discussion

Using our previously reported simple ex vivo spheroid-based Pd3D culture assay, we evaluated whether tumor-specific cell viability responses to octreotide, cabergoline, and pasireotide reflect clinical drug responsiveness in a consecutive cohort of patients with acromegaly. Although the overall reduction in viability was modest, substantial intertumoral variability was observed, and these responses showed concordance with established clinical response indices. This modest reduction is consistent with the primary pharmacological mechanisms of SRLs and dopamine agonists, which involve inhibition of hormone secretion, cell cycle arrest, and reduction in cell volume, rather than acute massive cytotoxicity. Furthermore, the ex vivo responder rates of 33–40% observed in our study align well with the clinical literature, which reports significant tumor shrinkage in approximately 30–50% of patients treated with these agents[1, 4]. These findings provide initial evidence suggesting the potential feasibility of this assay and suggest that it captures relative pharmacological susceptibility. Specifically, the significantly lower relative viability in the responder group compared to the nonresponder group suggests that this assay can reflect tumor-intrinsic drug sensitivity.

Regarding the baseline clinical characteristics, the proportion of patients showing a positive acute octreotide test in our cohort was relatively high (78.9%). This high responsiveness likely reflects the unselected, consecutive nature of our surgical cohort, which preserves the natural proportion of treatment-sensitive tumors and avoids the overrepresentation of medically refractory cases. Nevertheless, we acknowledge that the relatively small sample size may have introduced some selection bias.

Pd3D culture systems have been shown to capture patient-specific drug responses, and are increasingly recognized as translational functional assays [12–16]. In particular, cell-viability–based readouts from various patient-derived 3D tumor models have been reported to predict radiographic tumor responses, although discordant results have also been described [36, 37]. The present findings support the rationale for future prospective studies examining whether ex vivo effects correlate with subsequent tumor shrinkage and biochemical remission in the same patients.

In clinical practice, responsiveness to medical therapies varies significantly, highlighting the need for reliable predictive markers. While several clinical, radiological, and histopathological predictors of biochemical control have been identified for these agents [38–45], predictors of actual tumor shrinkage remain limited. Because biochemical control does not invariably coincide with radiographic response, there is a critical need for functional approaches that can anticipate both outcomes.

In our study, an ex vivo Pd3D culture assay generated sample-level viability readouts for each drug and demonstrated partial concordance with established clinical predictors, including octreotide responsiveness with T2WI hypointensity, cabergoline responsiveness with bromocriptine testing, and pasireotide with a sparsely granulated pattern. At the same time, discordant cases were also observed. This concordance suggests that Pd3D cultures preserve key tumor features relevant to drug responsiveness, supporting the face validity and feasibility of the assay. These features likely included preservation of SSTR2/5 and dopamine receptor (D2R) expression, the histopathological granulation pattern that contributes to the tumor’s T2WI MRI signal, downstream signaling pathways, and the cell—matrix context. Together, these features may underline clinical indicators such as responsiveness in dynamic drug tests. Although the clinical use of these dynamic suppression tests remains controversial, we utilized them as an immediate in vivo biological reference to validate whether our culture system successfully preserved the native functional receptor phenotypes. In addition, the 3D architecture facilitates the re-establishment of native cell—cell and cell—matrix interactions and microenvironmental gradients that influence receptor trafficking and signaling fidelity, features emphasized in contemporary pituitary 3D model work [17], and may stabilize SSTR/D2R localization and Gi/o-coupled signaling, improving the capture of patient-intrinsic drug sensitivity in our assay.

Complementing these viability assays, quantitative analysis of LBC specimens captured pharmacological dynamics of SSTR2 in Pd3D models. This methodology has previously been applied to hormone receptor assessment in breast cancer and immunocytochemical characterization of neuroendocrine carcinomas, supporting its utility as a robust foundation for molecular analysis in endocrine oncology [46–48]. In the present study, the concordance between baseline LBC-derived H-scores and FFPE-based Volante scores supports the validity of this approach as a surrogate for conventional pathology. Furthermore, digital image analysis of SSTR2 expression in FFPE tissues has been shown to correlate strongly with manual pathological scoring, providing an objective basis for receptor quantification [32]. Our findings in LBC specimens extend this concept to cytological samples. The observed dynamic IHC approach revealed functional shifts—specifically the upregulation or maintenance of high intensity fractions in responders and the depletion of receptor-positive cells in nonresponders—offering a more functional prediction of therapeutic response compared to static evaluation. However, as an important biological caveat, the mere presence of the receptor at the cell membrane does not guarantee the intracellular effect. To our knowledge, this is the first study to demonstrate the feasibility of AI-based quantitative cytometry in capturing drug-induced receptor dynamics in patient-derived pituitary models. By addressing the challenge of low cellular yield inherent in ex vivo assays, this dynamic LBC approach serves as a preliminary and promising exploratory proof-of-concept for integrated functional and pathological assessments, though further large-scale validation is warranted to confirm its predictive value.

An additional observation in this dataset was the association between cabergoline responsiveness and higher SSTR5 expression. This finding suggests receptor-level crosstalk as a potential mechanistic link. Hetero-oligomerization between D2R and SSTR5 can potentiate Gi/o signaling and alter desensitization/internalization, offering a rationale for enhanced dopamine agonist efficacy in SSTR5-high tumors [49]. Thus, this ex vivo assay is a promising translational tool for identifying previously unrecognized predictors and mechanisms of drug sensitivity.

Conversely, we observed a partial discordance between ex vivo responses to SRLs and established predictive markers, such as the lack of a significant difference in static SSTR2/5 expression between responders and nonresponders. However, our AI-based LBC analysis suggests that this discrepancy highlights the limitation of conventional static assessments rather than a weakness of the ex vivo model. Static post-operative SSTR evaluations only capture a baseline "snapshot" and may not reflect the tumor's ability to maintain receptor expression during actual drug exposure. Indeed, our dynamic analysis revealed a marked depletion of SSTR2 after treatment in nonresponders. While molecules such as β-arrestin are known to mediate receptor internalization and recycling, the observed receptor depletion in nonresponders suggests that intracellular degradation pathways may also be actively involved. Coupled with the biological caveat that the mere initial presence of the receptor at the membrane does not guarantee its intracellular efficacy, these findings suggest that our dynamic ex vivo assay may have the potential to predict pharmacological responses more accurately than conventional static pathological evaluations, although further validation is required.

Limitations of this study are as follows. First, we employed a single-dose screening strategy (10 nM). While this concentration is widely validated by previous landmark in vitro studies as optimal for capturing receptor-mediated efficacy in pituitary cultures [29–31], this approach inherently fails to capture the interpatient variability in dose–response (titration) curves. The relatively modest decrease in viability observed between responders and nonresponders may partly reflect sub-optimal dosing for individual tumors, highlighting the need for future studies incorporating dose-titration assays. Second, the model-intrinsic features of patient-derived spheroids can introduce selection and context effects; unlike organoids which maintain complex tissue architecture, our spheroid culture mainly enriches sphere-forming subpopulations and depletes stromal, vascular, and immune compartments. Extracellular matrix composition/mechanics and serum-borne cues may shift differentiation state and receptor trafficking such that in-culture SSTR/D2R abundance or granulation patterns may not perfectly reflect bulk-tumor readouts [17]. Third, interobserver variability in radiologic and pathologic classifications and spatial intratumoral heterogeneity can decouple imaging surrogates and receptor/secretory phenotypes from cell-intrinsic drug sensitivity. Fourth, given our sample size, chance variation may have contributed to the observed discordance with established predictive markers. This limitation also extended to the AI-based LBC analysis, where a small number of cases and the absence of a predefined cutoff value precluded formal statistical verification, underscoring the exploratory nature of this pilot sub-analysis and necessitating further validation of the observed trends in larger cohorts. Fifth, although our responder definition lacked a predefined biological threshold, the tightly converged effect sizes across all three agents (median relative viabilities of 84–86%; Fig. 1) strongly suggest that our statistical approach effectively captured a biologically distinct phenotype rather than random technical noise. Building on these results, establishing an optimal cutoff value to redefine responsiveness will be a crucial next step to further refine this assay and enhance its clinical utility. Finally, in this cohort, many patients achieved oncologic remission by surgery, and additional follow-up was required before medical therapy was indicated for residual or recurrent disease. As these patients were not treated with the tested drugs during the study period, a conclusive answer regarding actual clinical drug responsiveness is lacking. This precludes the direct correlation within the present study of ex vivo effect sizes with subsequent tumor shrinkage or biochemical remission, emphasizing that our findings currently remain correlative with established predictors rather than definitive clinical outcomes. While this pilot study does not immediately alter current clinical guidelines, applying AI-based analysis to the entire cohort and integrating multiomics analyses would provide deeper molecular insights and establish a critical foundation for future precision medicine, which remains a vital next step.

In conclusion, this is a pilot study that we propose a novel methodological approach that integrates patient-derived 3D (Pd3D) cultures with liquid-based cytology (LBC) to investigate the pathophysiology of pituitary adenomas. This integrated platform demonstrates potential utility as an experimental system for elucidating pharmacological pathophysiology, serving as a promising exploratory translational model to advance precision medicine in acromegaly.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

All authors contributed to the study conception and design. Material preparation, data collection and analysis were performed by Y.T., A.I. and H.F. The first draft of the manuscript was written by Y.T. and H.F., and all authors commented on previous versions of the manuscript. All authors read and approved the final manuscript.

Funding

Open Access funding provided by Kobe University. This work was partially supported by a Grant-in-Aid for Scientific Research from the Japan Society for the Promotion of Science (KAKENHI; grant numbers 19K09003 (HF), 22K08654 (HF), 25K02700 (HF), and 23K15412 (HS)); the Program for Forming Japan's Peak Research Universities (J-PEAKS) from the Japan Society for the Promotion of Science (JSPS) (TA); and by AMED under Grant Number JP256f0137011 (TA).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Ethics approval

All procedures complied with the protocol approved by the Research Ethics Committee of Kobe University Hospital and Moriyama Memorial Hospital (IRB #1363 and B240223, respectively) and adhered to the Declaration of Helsinki.

Consent to participate

Written informed consent was obtained from all participants prior to surgery and sample collection.

Competing interests

The authors declare no competing interests.