Development of a translational murine model for antibody-drug conjugate-induced interstitial lung disease

Establishment of a murine model to evaluate ADC-induced interstitial lung disease

To develop a robust murine model of ADC-induced interstitial lung disease (ILD), we selected trastuzumab deruxtecan (T-Dxd), a clinically validated ADC known to cause ILD in patients¹⁴. Dose selection was guided by previously described non-clinical studies, including those by Kumagai et al.¹². Based on FDA body-surface-area scaling (km ratio 12/3), a cynomolgus monkey dose of 78.8 mg/kg used in NHP studies translates to approximately 315 mg/kg in mice, underscoring the need for higher murine doses to achieve clinically relevant exposure. However, because ~300 mg/kg dose exceeds murine tolerability limits, a 200 mg/kg dose was selected as a high but feasible exposure level for initial ILD assessment in this model. The schematic representation of the dosing regimen used for initial ILD model development is shown in Fig. 1A. Fibrosis severity increased progressively with the number of T-Dxd doses, peaking after four weekly administrations (Group-1) (Fig. 1B) (Supplementary Table 2). Statistical analysis revealed significant increases in fibrosis scores in groups receiving three or four doses compared to trastuzumab controls at corresponding time points (p < 0.01). No significant fibrosis was observed in trastuzumab-treated animals across all dosing regimens. Histological distortion in T-Dxd-treated groups, characterized by alveolar septal thickening and fibrotic remodeling, while minimal to no pathological changes were observed in Trastuzumab-treated lungs (Fig. 1C). The observed histological features shown in Fig. 2B–J closely resemble those documented in clinical settings⁶ and non-human primate studies¹², supporting the translational relevance of this model. Notably, lung injury emerged only after multiple doses, indicating that repeated ADC exposure rather than a single high dose is essential to trigger ILD-associated changes. Whether this reflects attainment of a critical pulmonary concentration, a time-dependent biological process, or both cannot be distinguished in the current study.

The histopathology scoring system enables a comprehensive evaluation of ADC-induced ILD

Initial high-dose T-Dxd studies showed that repeated administration generated a range of pulmonary findings, including alveolar septal thickening, inflammatory cell infiltration, edema, hemorrhage, and evolving fibrotic remodeling. These observations indicated that ILD-like injury in mice reflects a continuum of inflammatory and remodeling events rather than a single discrete lesion. Accordingly, a structured and semi-quantitative scoring approach was developed to capture the multifactorial nature of ADC-associated lung injury and to support consistent, reproducible assessment across studies. This framework complements conventional toxicology endpoints and facilitates early identification of potential ILD risk in ADC programs, an emerging priority for translational safety evaluation¹⁵. To capture this complexity, we systematically assessed pulmonary toxicity across treatment groups and implemented a 12-parameter histopathology scoring system (Fig. 2A) that captures hallmark pathological features reported in both clinical¹⁶ and non-human primate studies of ADC-induced ILD¹². The parameters, illustrated in Fig. 2A and exemplified in panels B–J, span the spectrum of ILD pathology from acute inflammatory changes (e.g., alveolar inflammation, granulomas, lymphocytic infiltration, and mononuclear cell infiltration) to vascular and interstitial alterations (perivascular/intravascular edema, hemorrhage, pulmonary congestion, and microthrombi) and chronic remodeling (peribronchial thickening, fibrin deposition, and pulmonary fibrosis). Each parameter was scored semi-quantitatively based on whole-slide image analysis of lung sections, allowing for reproducible and granular evaluation of lung injury across cohorts. ILD severity was scored on a scale of 0–3 based on estimated tissue involvement across the whole slide image (WSI). Scores reflected the extent of histopathological features, with 0 (Normal) ≦ 5%, 1 (Mild) = 5–33%, 2 (Moderate) (33–66%), and 3 (Marked) (>66%). The composite ILD severity score was derived by summing individual parameter scores. This structured scoring approach allowed clear discrimination between ADCs with low versus high pulmonary toxicity and demonstrated sensitivity to changes in dosing frequency.

Delayed onset of T-Dxd-induced interstitial fibrosis in the murine ILD model

The 200 mg/kg dose of T-Dxd was intentionally used as an initial dose-finding and feasibility study to establish whether ILD-like pulmonary pathology could be reproducibly induced in mice. After confirming the presence of ILD-like pathology under high-exposure conditions, we then refined the dosing regimen to more closely align with clinically relevant exposures to evaluate the sensitivity and translational applicability of the model. Recognizing that non-human primate (NHP) studies demonstrated delayed-onset ILD following repeated, clinically relevant T-Dxd exposure¹², we conducted a time-course study to characterize the temporal dynamics of pulmonary injury. To enhance translational relevance, the dose was reduced to 100 mg/kg, which approximates the human equivalent exposure range derived from the approved 5.4 – 8 mg/kg clinical dose when adjusted for species-specific pharmacokinetics¹. This lower dose was selected to elicit ILD-like pathology while minimizing acute toxicity and enabling observation of delayed pulmonary effects.

Mice received T-Dxd or Trastuzumab at 100 mg/kg (Q3D), and lungs were evaluated at 48 h, 1 week, and 2 weeks post final dose (Fig. 3A). Mean injury scores for fibrosis (Fig. 3B), (Supplementary Table 3) are elevated in T-Dxd treated mice compared to Trastuzumab controls after 1 week and 2 weeks, with the highest score at 2 weeks. Composite injury scores (supplementary Fig. S1) revealed that T-Dxd induced a broad range of lesions, including alveolar septal thickening, inflammatory infiltrates, pneumocyte hyperplasia, and early fibrotic changes, each contributing to overall pathology severity. Trastuzumab-treated animals maintained low injury and composite scores throughout. Histological analysis (Fig. 3C) (supplementary Fig. S2) confirmed these quantitative findings. A representative whole-slide image is provided in Supplementary Fig. S3. At 1-week post-last dose, T-Dxd-treated lungs exhibited diffuse alveolar damage with interstitial widening and inflammatory cell infiltration. 2 weeks after the last dose, tissue architecture was distorted, with ongoing inflammation and fibrosis. In contrast, Trastuzumab controls showed preserved alveolar structure and minimal pathology at all time points. Our murine findings are consistent with non-human primate data, where cynomolgus monkeys given T-Dxd at ≥30 mg/kg every 3 weeks (4 doses) developed interstitial pneumonitis with lymphocytic infiltration, alveolar edema, and fibrosis¹². The concordance between murine and NHP temporal patterns highlights the model’s translational utility for early detection of ADC-associated pulmonary risk.

Minimal lung toxicity observed with SN-38-based ADCs in preclinical ILD model

Previous time-course experiments demonstrated that ILD-like changes in mice were most consistently detectable approximately two weeks following the final ADC dose. Therefore, to evaluate ADC-induced lung injury, this interval was selected as the standardized observation window. We applied the same delayed-onset ILD assessment protocol to test ADC constructs. Mice were administered with ADCs at 100 mg/kg intravenously Q3D for 2 weeks (Fig. 4A). Study arms included Vehicle, Trastuzumab, T-SN-38-CL2A, and T-Dxd. Although pharmacokinetics is different between T-Dxd and SN-38-based ADCs, a uniform mg/kg dose (100 mg/kg) was used across groups to enable controlled cross-construct comparison. Lungs were collected 14 days after the final dose for histopathologic evaluation. Pulmonary pathology between T-Dxd and T-SN-38-CL2A treatment cohorts was compared under parallel dosing conditions. Mean injury scores for fibrosis, interstitial fibrinous deposition, hemorrhage, edema, and lymphocytic infiltration (Fig. 4B) and composite injury scores for several ILD parameters (Fig. 4C), (Supplementary Table 4) revealed no ILD-like pathology in Vehicle, Trastuzumab, and T-SN-38-CL2A-treated mice, with consistently low severity values across all 12 ILD parameters. Histopathological analysis two weeks post last dose confirmed preserved alveolar architecture in T-SN-38-CL2A treated lungs, with no evidence of peribronchial inflammation, septal thickening, or fibrotic remodeling, supporting their favorable lung safety-profile, even under conditions designed to uncover delayed-onset ILD (Fig. 4D). In contrast, T-Dxd treated mice exhibited markedly elevated composite and mean injury scores, reflecting a broad range of pathological features including alveolar septal thickening, inflammatory cell infiltration, pneumocyte hyperplasia, peribronchial inflammation, granuloma formation, and focal hemorrhage. These lesions were absent in all other treatment groups, highlighting a clear divergence in pulmonary outcomes between ADCs T-SN-38-CL2A and T-Dxd.

Together, these findings suggest that the pulmonary effects observed are not solely explained by topoisomerase-I inhibition from the ADC payload. While this data demonstrates that the model can distinguish ADCs with differing pulmonary responses, mechanistic conclusions regarding specific structural drivers cannot be inferred from this comparison alone. SN-38 and Dxd differ in potency, biodistribution, and pharmacologic properties, and therefore direct attribution of the observed differences to linker or payload components is not appropriate. Instead, these results highlight the capability of this model to detect ADCs with differential pulmonary safety profiles and support future mechanistic work using matched payload-linker constructs to define structural contributors to ILD risk more precisely.

Evaluation of structural features contributing to ADC-associated ILD

ADC-induced interstitial lung disease (ILD) is thought to arise, at least in part, from dose-dependent but target antigen-independent uptake of ADCs in lung tissue via Fcγ receptor-mediated internalization by alveolar macrophages¹⁷. We hypothesized that structural elements of ADCs, including linker design, molecular architecture, and physicochemical properties, may modulate target-independent pulmonary uptake and/or local payload release, thereby contributing to ILD liability. Similar considerations regarding the role of ADC architecture in pulmonary toxicity have recently been proposed in the context of CEACAM5-exatecan ADC development by Sloot et al.¹³. To explore whether the murine ILD model could differentiate pulmonary responses across ADC constructs with distinct architectures, we generated an exploratory ADC incorporating the same GGFG (Gly-Gly-Phe-Gly) cleavable sequence and immolation motif used in T-Dxd but replacing the cytotoxic payload with a non-cytotoxic cyclohexylamine (“dummy” payload). This construct T-MC-GGFG-Cy (structure shown in Table 1) was synthesized to enable evaluation of structural contributions under otherwise comparable dosing conditions. A dose response viability assay showed that both the Trastuzumab-dummy payload ADC (T-MC-GGFG-Cy) and unconjugated Trastuzumab produced comparable effects on cell viability, with overlapping concentration response profiles, confirming that the dummy payload did not contribute to the cytotoxicity (Fig. 5B). The cytotoxicity profiles were independently confirmed using a confluence-based proliferation assay measured with the Incucyte S3 live-cell imaging platform. Representative data are shown in Supplementary Fig. S4. We then compared lung injury induced by T-MC-GGFG-Cy, T-Dxd, Trastuzumab, and vehicle controls using our delayed-onset ILD assessment protocol (Fig. 5A). All groups were dosed under identical schedules and comparable mg/kg exposures. The mean injury scores revealed substantial differences in ILD severity between the treatment groups. Both ADCs with cleavable peptide linkers with specific chemical motifs, such as the GGFG (T-Dxd and T-MC-GGFG-Cy), induced pronounced pulmonary pathology, including lymphocytic infiltration, perivascular edema, hemorrhage, and fibrosis (Fig. 5C). Importantly, these differences were independent of the presence of cytotoxic payload, as observed with ADC containing a non-cytotoxic dummy payload. The composite lung injury scores in mice treated with T-MC-GGFG-Cy approached those of T-Dxd and were significantly higher than Trastuzumab or vehicle controls (p < 0.01) (Fig. 5D), (Supplementary Table 5). Histopathology assessment (Fig. 5E) revealed that Vehicle and Trastuzumab-treated lungs preserved alveolar architecture with no significant lesions. In contrast, T-MC-GGFG-Cy-treated lungs exhibited ILD-like changes, including alveolar wall thickening, interstitial inflammatory infiltrates, and fibrotic foci. T-Dxd-treated lungs showed a comparable ILD injury pattern with hemorrhage and diffuse parenchymal remodeling. The observed ILD pathology between T-Dxd-based ADCs and T-SN-38-CL2A confirms that the ADCs carrying the same cytotoxic payload class but conjugated through different linkers yielded divergent outcomes, with non-GGFG linker architectures producing minimal lung injury.

Table 1.

Test articles evaluated in studies

Open in a new tab

Test articles evaluated in this study.

Schematic representation of antibody–drug conjugates (ADCs) used in the study, showing antibody scaffold, linker, and payload structures. The drug-to-antibody ratio (DAR) was fixed at 8.

Evaluation of linker–payload engineered ADC variants with preserved antitumor activity and markedly reduced ILD liability

Motivated by prior observations that a GGFG-linked dummy payload elicited pronounced ILD features relative to T-Dxd, we sought to determine whether modifying linker–payload architecture could mitigate ADC-associated pulmonary injury without compromising antitumor activity. To this end, we applied a stepwise evaluation strategy using our delayed-onset ILD protocol to assess four novel ADCs: NJB-T-PL10, NJB-T-PL30, NJB-T-PL90, and NJB-T-PL119 (Table 2). We initially generated and evaluated NJB-T-PL10 and NJB-T-PL30, both of which incorporate a GGFG-based cleavable linker but differ in their topoisomerase I inhibitor payloads. In efficacy studies using the JIMT-1 xenograft model, both NJB-T-PL10 and NJB-T-PL30 demonstrated measurable antitumor activity (Fig. 6A). However, NJB-T-PL10 exhibited reduced efficacy compared with NJB-T-PL30. Despite these differences in efficacy, both constructs produced consistently elevated composite ILD scores under the delayed-onset protocol, reflecting increased fibrosis, interalveolar and interstitial inflammation, intravascular fibrinous deposition, peribronchial and peribronchiolar inflammation, microthrombi, vascular congestion, hemorrhage, perivascular edema, lymphoid hyperplasia, granulomatous changes, and alveolar injury relative to vehicle and Trastuzumab-treated controls (Fig. 6D, E), (Supplementary Table 6). These findings indicate that ILD liability is retained across different payloads with a shared GGFG linker and immolation motif. Next, to assess whether altering linker–payload architectural features could modulate this phenotype, we synthesized NJB-T-PL90, in which the NJB-T-PL30 payload was preserved while the GGFG linker was replaced with a non-GGFG linker. This novel ADC variant showed tumor efficacy comparable to T-Dxd (Fig. 6B) and showed reduction in composite ILD scores approaching levels observed in vehicle and Trastuzumab treatment groups (Fig. 6F), suggesting attenuation of lung injury with changes in linker–payload architecture. To further extend this observation, we generated NJB-T-PL119, which retained the same non-GGFG linker used in NJB-T-PL90 but incorporated a structurally distinct topoisomerase I inhibitor payload. The resulting ADC, NJB-T-PL119, showed significant efficacy against JIMT-1 tumors (Fig. 6C) and exhibited the lowest composite ILD scores among the tested ADCs, with minimal histopathologic features of lung injury (Fig. 6G). All four novel ADC variants demonstrated tumor growth inhibition, indicating that the observed ILD scores were not associated with a loss of antitumor activity. Collectively, these findings provide preliminary evidence that ADC linker–payload architecture is associated with differences in ILD liability, with GGFG-based cleavable constructs showing higher composite lung injury scores and non-cleavable architectures showing reduced pathology in our ILD model.

Table 2.

Novel ADCs evaluated in this study

Open in a new tab

Schematic representation of the four HER2-targeted ADCs evaluated in this study.

NJB-T-PL10, NJB-T-PL30, NJB-T-PL90, and NJB-T-PL119, highlighting differences in linker–payload architecture. NJB-T-PL10 and NJB-T-PL30 incorporate a cleavable MC-GGFG linker, whereas NJB-T-PL90 and NJB-T-PL119 utilize non-cleavable, non-GGFG linker architectures.

While these exploratory observations raise the possibility that ADC structural features may influence ILD-like outcomes, definitive attribution of lung toxicity to any specific component (e.g., linker structure, payload class, or other molecular properties) requires further mechanistic investigation. Construct-matched controls (e.g., an exatecan-based ADC with a CL2A linker) and dedicated pharmacokinetic and tissue-distribution analyses will be needed to precisely dissect structure-toxicity relationships, which are beyond the scope of the present study. However, the data indicate that modifying the nature of the linker, immolation motif, and payload is key to attenuating off-target activity in the lung while preserving antitumor potency.

Test articles

The purchased reagents were used as is without further purification. Deruxtecan and CL2A-SN-38 were purchased from Medchem Express. Deuterated solvents were purchased from Cambridge Isotope Laboratories (Andover, MA). NMR spectra were recorded on a Varian Unity INOVA 500, and samples calibrated for: ¹H NMR (DMSO, δ = 2.5 ppm). Trastuzumab (anti-HER2) (T) was purchased from Selleckchem (Cat. No. A2007, Selleckchem, Houston, TX). Novel ADCs NJB-T-PL10 and NJB-T-PL30 were synthesized according to methods disclosed in a published patent (WO2024/248944 A1). ADCs NJB-T-PL90 and NJB-T-PL119 were synthesized according to methods disclosed in a published patent (WO2025/222127 A2). ADCs T-Dxd, T-SN-38-CL2A, and T-MC-GGFG-Cy were prepared in 1× histidine buffer.

Antibody–drug conjugation

Trastuzumab at a concentration of 5.1 mg/mL was fully reduced using 9–12 molar equivalents of TCEP, followed by the addition of 12–13 molar equivalents of linker–payload in 10–20% v/v dimethylacetamide (DMA). The resulting conjugates were purified by size-exclusion chromatography and/or tangential flow filtration (TFF) into 20 mM histidine buffer (pH 5.5) over 6–10 diavolumes. The final conjugates were concentrated, sterile filtered through a 0.2 µm PTFE filter, and stored at the working concentration (Supplementary Table 1). The monomer content was confirmed by SEC using a TSKgel G3000SWXL column (7.8 × 300 mm). DAR was verified by LC-MS using a Waters Xevo G2-XS QTof mass spectrometer coupled to an ACQUITY UPLC system with a PLRP-S column (1000 Å, 8 µm, 2.1 × 50 mm). Chemical structures of the linker payloads used for conjugation are shown in Tables 1 and 2.

Cell viability assays

HER2-positive SKBR-3 breast cancer cells (ATCC HTB-30) were cultured in McCoy’s 5A medium supplemented with 10% FBS at 37 °C with 5% CO₂. For the cytotoxicity assay, Cells were seeded in 96-well plates at 2 × 10³ cells/well and allowed to adhere overnight. Serial dilutions of test articles (PBS Vehicle, Trastuzumab, and T-MC-GGFG-Cy) were prepared in assay medium (final concentration range: 2.5 × 10⁻¹⁴ – 1.4 × 10⁻⁷ M). Cells were treated for 144 h, after which viability was assessed using CellTiter-Glo® (Promega) according to the manufacturer’s protocol. Luminescence was measured on a BioTek Synergy HTX plate reader. Curves were generated with a three-parameter nonlinear regression model in GraphPad Prism (v10). Data represent mean ± SD of triplicate wells. For confluence measurement, cells were treated in 96-well plates and placed in an Incucyte S3 live-cell imaging system and imaged every 12 h. Images were acquired using a 10× objective, capturing five fields per well. Cell confluence was quantified using Incucyte software (version 2024B) with the AI-Confluence analysis module. A size filter was applied to include objects larger than 500 µm² to exclude dead cells and debris. Confluence values were normalized to the baseline measurement at time 0. Data from the 144 h time point were exported and plotted against treatment concentration in GraphPad Prism (v10 or later). Curves were generated using a three-parameter nonlinear regression model.

Animal studies

All animal experiments were performed in compliance with institutional guidelines, approved by the Stony Brook University Institutional Animal Care and Use Committee (IACUC) (Protocol #2021-00074) and conducted in accordance with institutional guidelines and Code of Federal Regulations 9 CFR 2.31. Female Balb/C (RRID:MGI:2161072) mice 6–8 weeks (about 2 months) old were purchased from Taconic Biosciences (Germantown, NY). The mice were acclimated for at least 7 days prior to study initiation and housed under specific pathogen-free conditions with a 12-hour light/dark cycle, ad libitum access to food and water, and environmental enrichment. For initial ILD model development, Trastuzumab and T-Dxd were administered at 200 mg/kg (5 mL/kg) once weekly for 4 weeks. Mice were assigned to four staggered dosing cohorts and administered intravenous doses on Days 0, 7, 14, and 21. Group 1 received doses on Days 0, 7, 14, and 21 (four doses total), Group 2 on Days 7, 14, and 21 (three doses), Group 3 received doses on Days 14 and 21 (two doses), and Group 4 received a single dose on Day 21 (Fig. 1A). All animals were euthanized on Study Day 28, by exposure to carbon dioxide gas in a rising concentration, followed by cervical dislocation to enable uniform histopathological assessment across groups and evaluate ILD-like changes as a function of repeated dosing. As a vehicle, 1× PBS pH 7.0, or HBS buffer (10 mM Histidine, 10% Trehalose, 0.02% Polysorbate 20, pH 5.5) was administered at the same volume as the ADCs. For the delayed ILD-onset protocol, the animals were randomized into treatment groups based on body weight. The test articles T (Trastuzumab) and ADCs T-Dxd, T-SN-38-CL2A, T-MC-GGFG-Cy, NJB-T-PL10, NJB-T-PL30, NJB-T-PL90, and NJB-T-PL119 were administered intravenously at 100 mg/kg on a Q3D schedule for two weeks. A maximum of four doses was administered at three-day intervals, unless otherwise specified. Body weights and clinical signs were monitored at regular intervals to assess tolerability and systemic toxicity. Animals were euthanized two weeks post last dose or earlier if humane endpoint criteria were met.

Tumor efficacy studies

Female NOD.CB17-Prkdc<scid > /J (NOD-SCID) mice were purchased from Jackson Laboratory (Bar Harbor, Maine, USA). The mice were acclimated prior to study initiation and housed under specific pathogen-free conditions with a 12 h light/dark cycle, ad libitum access to food and water. JIMT-1 human breast cancer cells (ACC 589, DSMZ cell lines bank, Germany) cultured in 90% DMEM + 10% Hi FBS were harvested during logarithmic growth, counted, and resuspended in sterile PBS (+Matrigel, 1:1) at a final concentration of 3 × 10⁷ cells/mL. Mice were inoculated subcutaneously in the right flank with 3 × 10⁶ JIMT-1 cells in a total volume of 100 µL per mouse. For all in vivo tumor studies, power calculation to determine sample sizes (5% significance, 90% statistical power) is based on a two-tailed, Mann–Whitney U test. Stratified randomization has been performed for all efficacy studies. An independent researcher was assigned to blindly measure parameters such as tumor size to reduce or eliminate some sources of experimental bias. Tumor growth was monitored using digital calipers, and tumor volume (TV) was calculated using the formula TV = (length × width²)/2. Once tumors reached the predetermined enrollment range (typically ~200–250 mm³), mice were randomized into treatment groups to ensure comparable mean tumor volumes at baseline. The test ADCs were administered at a dose of 10 mg/kg, given once weekly for 2 weeks (Q7D × 2) by intravenous injection. Control animals received Vehicle and/or Trastuzumab according to the same schedule. Tumor measurements and body weights were recorded at least twice a week throughout the study. Animals were monitored daily for clinical signs of distress, and predefined humane endpoints (maximum tumor volume or health criteria) were applied in accordance with institutional animal care and use committee (IACUC) approved protocols. Tumor growth inhibition and response rates were calculated relative to baseline and vehicle-treated controls.

Pathology and histological evaluation

At necropsy, lungs were inflated via the trachea with 1 mL of 10% neutral buffered formalin (NBF) using a 3 mL syringe fitted with a 27-gauge needle to preserve alveolar architecture. The tissues were immersion-fixed for 48 h at room temperature and then transferred to 70% ethanol for storage prior to processing. Whole lungs, including both left and right lobes, were processed and sectioned at 5µm thickness, placed on a glass slide, and then stained with hematoxylin and eosin (H&E) using standard protocols. For each animal, the entire lung section slides were digitally scanned using Aperio AT2 (MTC) or GT450 (H&E) scanner (Leica Biosystems, Inc.) to generate Whole Slide Images (WSIs), which were uploaded to the cloud platform for evaluation. Histopathologic evaluation was performed on the full cross-sectional area of the lung section, allowing assessment of all lobes present within the section and minimizing regional sampling bias. Lesions were scored semi-quantitatively across the entire lung section under blinded conditions. WSIs were reviewed in their entirety by a senior board-certified veterinary pathologist in a blinded manner. Previously annotated slides, Excel datasets, and pathology reports were reviewed and correlated with current findings. Interstitial lung disease (ILD)-like changes were scored using a semi-quantitative system based on published criteria^25,26. Composite grading incorporated key pathological features, including alveolar septal thickening, inflammatory cell infiltration, pneumocyte hyperplasia, and fibrosis, with severity defined on a 3-point scale. Representative lesions were imaged under brightfield microscopy, and high-resolution, non-compressed images suitable for publication were acquired at 40 × magnification with a 50 µm scale bar. For archival and data sharing, images were uploaded to a secure cloud repository. In selected samples, additional staining such as Masson’s trichrome was performed to assess collagen deposition and fibrotic remodeling.

ILD histopathology scoring system

A structured 12-parameter histopathology scoring system was applied to whole-slide images (WSIs) of lung sections to capture the spectrum of ADC-induced interstitial lung disease (ILD) pathology. Each parameter was graded semi-quantitatively on a 0 – 3 scale, where 0 = normal (<5% of lung parenchyma involved), 1 = mild (5 – 33%), 2 = moderate (33 – 66%), and 3 = marked (>66%). The system encompassed three domains of pathology. Acute inflammatory features included alveolar inflammation, defined as neutrophils and activated macrophages within alveolar spaces and septa, lymphocytic infiltration in the interstitium or peribronchial regions, mononuclear cell infiltration, characterized by foamy or pigment-laden macrophages and granulomas, defined as well-formed epithelioid histiocyte aggregates with or without giant cells, scored separately as 0 = none, 1 = rare, 2 = multiple small foci, or 3 = numerous/confluent. Vascular and interstitial alterations included perivascular or interstitial edema, pulmonary congestion (capillary and venous engorgement), hemorrhage (extravasated erythrocytes in alveolar or interstitial spaces), and microthrombi, defined as fibrin-rich occlusive thrombi in small vessels and scored as 0 = none, 1 = 1 – 2 thrombi per lobe, 2 = 3 – 5 per lobe, or 3 = >5 per lobe or confluent distribution. Chronic and remodeling features included peribronchial wall thickening due to smooth muscle and stromal expansion, intra-alveolar or interstitial fibrin deposition (confirmed by special stains where necessary), pulmonary fibrosis defined as collagenized interstitium with architectural distortion, and type II pneumocyte hyperplasia, scored by the proportion of alveolar surface lined by reactive cuboidal epithelium. Scoring was performed on WSIs at low power with high-power confirmation, surveying all lobes. Scores per animal reflected the aggregate of all lobes. If lobes were scored separately, values were averaged per animal. A composite ILD severity score (range 0–36) was calculated by summing the 12 parameters, and a mean injury score (range 0–3) was derived by dividing the composite score by 12. All slides were digitally coded and reviewed by a board-certified veterinary pathologist blinded to treatment assignment to minimize bias. For focal or binary lesions such as granulomas and microthrombi, explicit count-based categories were used when lesions were sparse, while percent area scoring was applied when lesions were extensive or confluent. To ensure consistency, airways >2 mm and large vessels were excluded from parenchymal calculations, while terminal bronchioles were included. When edema and hemorrhage co-localized, each was scored independently if thresholds were met. For fibrosis, only regions with architectural distortion were prioritized, while perivascular stromal thickening alone was not scored as fibrotic.

Synthesis of MC-GGFG-Cy

To the solution of cyclohexyl amine (10 mg, 0.101 mmol, 1 eq) and MC-GGFG-CH₂-O-CH₂COOH (CAS: 1599440-25-1) (69.1 mg, 0.112 mmol, 1.1 eq) in anhydrous DMF at RT, NMM (0.014 mL, 0.126 mmol, 1.25 eq) was added, followed by DMTMM (55.8 mg, 0.126 mmol, 1.25 eq). The reaction was stirred at room temperature until the consumption of the starting material. After 10 min, diluted with (1:1) water: acetonitrile and purified via reverse phase C18 (50 g) using water: acetonitrile (0.05% FA). The product fractions were combined and lyophilized to get MC-GGFG-Cy (46 mg, 65% yield) as a white solid. (ESI) m/z: [M + H] ⁺ = 698.41. ¹H NMR (500 MHz, DMSO) δ 8.65 (t, J = 6.6 Hz, 1H), 8.35 (t, J = 5.9 Hz, 1H), 8.16 (d, J = 7.9 Hz, 1H), 8.08 (t, J = 5.7 Hz, 1H), 8.02 (t, J = 5.7 Hz, 1H), 7.54 (d, J = 8.2 Hz, 1H), 7.28 – 7.21 (m, 4H), 7.20 – 7.15 (m, 1H), 7.00 (s, 2H), 4.56 (d, J = 6.6 Hz, 2H), 4.52 – 4.44 (m, 1H), 3.81 (s, 2H), 3.74 (ddd, J = 9.5, 8.9, 4.7 Hz, 3H), 3.67 (t, J = 5.0 Hz, 2H), 3.59 (dd, J = 16.7, 5.5 Hz, 2H), 3.38 – 3.35 (m, 2H), 3.06 (dd, J = 13.8, 4.4 Hz, 1H), 2.80 (dd, J = 13.8, 9.8 Hz, 1H), 2.14 – 2.07 (m, 2H), 1.74 – 1.61 (m, 4H), 1.57 – 1.42 (m, 5H), 1.28 – 1.15 (m, 6H), 1.12 – 1.02 (m, 1H).

Calculation of drug to antibody ratio (DAR)

To determine the DAR of the conjugates by RP-LC/MS, 0.5 µg of ADC was injected onto a PLRP-S column (1000 Å, 8 µm, 2.1 × 50 mm) on a Waters Acquity UPLC running a 25 – 45% gradient (ACN  :Water + 0.1% formic acid; 0.35 mL/min flow) connected to a Xevo G2-XS QTof mass spectrometer. Monomeric purity of the ADC’s was determined by size exclusion chromatography (SEC); 20 µg of ADC was injected on a TSKgel3000SWXL column running an isocratic method of PBS pH 7.4 with 10% IPA at a flow of 0.8 mL/min. Purification of ADCs was performed either by preparative SEC followed by tangential flow filtration (TFF), or TFF alone. SEC purification was performed on an AKTA Avant using a Hiload^TM 26/600 Superdex^TM 200 pg column with isocratic 1× PBS as the mobile phase at 2.6 mL/minute. TFF was performed with a Sartorius 30,000 MWCO HY cassette. ADCs were buffer exchanged into 20 mM histidine, pH 5.5, with 6 – 10 diavolumes. Following purification, ADCs were concentrated with 30 KDa MWCO spin concentrators to the working concentration before being passed through a 0.2 µm filter and aliquoted for analysis and storage. Final analysis was performed as previously written, with the additional step of reducing 10 µg of ADC with dithiothreitol for DAR analysis by LC–MS. LC–MS spectra was deconvoluted and signals were used in the below formula (Eq. 1) for DAR determination.

LC–MS DAR Calculation- where n is the number of linker–payloads attached. Ln and Hn refer to the signal intensities of the light(L) and heavy(H) species with respect to the number of linker–payloads attached.

Statistical analysis

All statistical analyses were conducted using GraphPad Prism (v10.0 or later). Data are presented as mean ± standard deviation. Prior to hypothesis testing, data distributions were assessed for normality using the Shapiro-Wilk test. For group-wise comparison, a Kruskal–Wallis test followed by Dunn’s post hoc multiple comparison test was applied. A p-value of <0.05 was considered statistically significant. In figures, significance levels are indicated as follows: (*) represents p values between 0.01 and 0.05, and (**) represents p-values between 0.001 and 0.01.