The spatially informed mFISHseq assay resolves biomarker discordance and predicts treatment response in breast cancer

Background: Breast cancer (BCa) is a heterogeneous disease requiring precise diagnostics to guide effective treatment. Current assays fail to adequately address the complex biology of BCa subtypes/risk groups and accurately predict responses to treatments like antibody-drug conjugates (ADCs). To address these limitations, we developed and validated a novel diagnostic, prognostic, and predictive tool, mFISHseq. Methods: Our approach, mFISHseq, integrates multiplexed RNA fluorescent in situ hybridization with RNA-sequencing, guided by laser capture microdissection. This technique ensures tumor purity, allows unbiased profiling of whole transcriptome data, and explicitly quantifies intratumoral heterogeneity. Results: In a retrospective cohort study involving 1,082 FFPE breast tumors, mFISHseq demonstrated high analytical validity with 93% accuracy compared to immunohistochemistry across training and test sets. Our consensus subtyping approach provided near-perfect concordance with other molecular classifiers (κ > 0.85) and reclassified 30% of samples into subtypes with distinct prognostic implications. Consensus risk groups mitigated misclassification of single samples and provided prognostic information about both early and late relapse. High risk patients had enriched innate and adaptive immune signatures, which predicted response to neoadjuvant immunotherapy. Furthermore, we identified patients responsive to ADCs, as evidenced by a 19-feature classifier for T-DM1 sensitivity, validated on the multicenter, phase II, prospective I-SPY2 trial. To demonstrate the clinical potential, we deployed mFISHseq as a research use only test on 48 patients, revealing insights into the efficacy of novel targeted therapies, such as CDK4/6 inhibitors, immune checkpoint inhibitors, and ADCs. Conclusion: The mFISHseq method solves a long-standing challenge in the precise diagnosis and classification of BCa subtypes/prognostic risk groups, and allows accurate response prediction for patients, including those treated with immunotherapies and ADCs. Out of a starting cohort of 1,082 breast samples, we excluded one sample for revoked informed consent, four samples for damaged FFPE blocks or sections rendering them unable to be processed, 63 samples because pathology review revealed benign/healthy tissue or DCIS/LCIS, and one sample had missing clinical data. This left a cohort of 1,013 breast tumors available for later analyses. The published 1,254 breast cancer samples are comprised of 1,014 patients with invasive breast cancer (1 sample has no clinical data), 99 subtype samples from patients who had an extra region of interest (ROI) collected by laser capture microdissection (LCM), 25 patients with in situ carcinoma (24 DCIS/1 LCIS), 24 no tumor tissues (i.e., tissues dissected from tumor specimens that contained only healthy, atypical ductal hyperplasia, or other benign cells upon pathological review), 12 true healthy samples, 41 scroll samples used for the LCM vs. no LCM experiment, and 39 positive control samples. The Macherey Nagel NucleoSpin total RNA FFPE XS kit was used for RNA isolation. After RNA isolation, RNA quantity was measured using the Qubit RNA HS (High Sensitivity) Assay Kit with a Qubit 4 Fluorometer and RNA quality using the Agilent High Sensitivity RNA ScreenTape with an Agilent 4150 TapeStation. The DV200 value of the sample (i.e., the percentage of fragments more than 200 bases in length) was calculated as recommended by Illumina. Samples with DV200 values over 15% were considered viable samples for library preparation. We used the Takara SMARTer Stranded Total RNA-Seq Kit v3 - Pico Input Mammalian kit to prepare total RNA-SEQ libraries following the manufacturer's instructions. We included a single natural positive control sample in each library preparation batch and a synthetic spike-in control in each sample to control for batch library preparation effects. Following library preparation, the quantity and fragment size range of the library was assessed using both the Qubit dsDNA HS kit (Qubit 4 Fluorometer) and the Agilent High Sensitivity DNA ScreenTape kit (Agilent 4150 TapeStation). Successfully prepared libraries contained sufficient library to pool on an Illumina NovaSeq 6000 sequencing instrument and fragment range spanning 200 - 1,000 bp, with a local maximum of 250 - 350 bp. Depending on pool size, individual sequencing libraries were pooled and sequenced on an Illumina NovaSeq 6000 using SP, S1, S2, or S4 flow cells. Pooled libraries were spiked with 10% PhiX as recommended by both Illumina and Takara for low-complexity libraries sequenced on patterned flow cells. Paired-end sequencing (2 x 100 bp) was conducted to obtain approximately 100 million reads per sample.