T1, June 2024. “What does the first biopsy show?”
Before the first scRNA-seq sampling at T1 (June 2024), Sid had been treated with SBRT, cisplatin plus doxorubicin, and then doxorubicin plus ifosfamide. After standard of care was exhausted, Sid received SQ3370 doxorubicin-biogel therapy, JLF and CeGaT peptide vaccines, DeltaRex-G (a tumor-targeted gene therapy), pazopanib, and trabectedin. Circulating tumor DNA had been low for a period, but disease later recurred at the Th4 vertebra. A biopsy at this site (T1, UCLA) yielded the first scRNA-seq timepoint.
We began by auto-annotating cells. In some osteosarcomas, tumor cells express RANKL, which activates RANK on osteoclast precursors. The T1 annotation identified osteoclast-labeled cells that expressed RANK and classic myeloid markers. TNFSF11 (RANKL) RNA was low in annotated tumor, stromal, and T cells. Cells annotated as tumor also expressed CAF genes. Sid received denosumab after T1. Protein-level RANKL is still an open question: T1 has no antibody-derived tag (ADT) matrix, and later CD254 counts are nearly empty.
At T1, macrophages had a higher M2/reparative score than at T2. The LAM/foamy score stayed nearly flat, with small version-dependent shifts and a slightly higher T2 score in the full merge. The inflammatory signal remained concentrated in classical monocytes. An MHC-II-high APC-like state was also present. Dexamethasone overlapped the T1 biopsy, while all six exploratory steroid-response genes were higher at T1. Steroid exposure may help explain the T1 signal. Together, these states put CSF1R, TREM2, CD40, and STING/IFN on the map for mechanistic follow-up.
At T1, annotated tumor cells showed low MHC-I core, antigen-processing, and IFN-response RNA programs. Expression of cancer-testis antigens was modest and similar between annotated tumor and stromal cells. These scores measure RNA, leaving functional antigen presentation as the key next experiment.
Low MHC-I can reduce inhibitory self signals for NK cells. However, T1 tumor cells were also weak for activating MICA/MICB/ULBPs and PVR/NECTIN2 programs. NK and CD8 effector-memory cells expressed cytotoxic machinery, while proliferation was low and exhaustion markers were not high. This creates a clear next experiment: direct NK-killing assays with defined activating inputs.
T1 therefore provides a lower-antigen, more reparative baseline for the two later biopsies.
T2, January 2025. “What changed after T1?”
Between T1 and T2, Sid received proton therapy and intratumoral AdaPT-001 (an oncolytic virus). He also received SNK-01 (autologous NK-cell therapy), Anktiva/N-803 (an IL-15 agonist), dostarlimab (anti-PD-1), ipilimumab (anti-CTLA-4), and denosumab (anti-RANKL). With several treatments administered during the interval, T2 captures the biological state after dense treatment exposure.
We merged T1 and T2 to compare their expression states. Reclustering changed cell labels across versions, so we used broad compartments for longitudinal scoring. Within the broad stromal compartment, T2 had higher MHC-I core, antigen-processing, and IFN-response scores than T1. The CAF contrast placed HLA-A, HLA-B, HLA-C, B2M, PSMB9, HLA-DRA, HLA-DRB1, HLA-DRB5, CD74, STAT1, and IRF1 in the T2 direction. Its group-member lists are empty, so we treat this as stromal evidence rather than a malignant-cell comparison.
T2 TAMs made the clearest switch. They upregulated antigen-presentation and IFN genes. Their IFN/APC score nearly doubled, from 1.4812 at T1 to 2.9242 at T2. The M2/reparative score was higher at T1. The LAM/foamy score stayed nearly flat. This is the clearest macrophage state shift in the series.
T2 T cells had higher cytotoxic, exhaustion, and chronic-stimulation RNA scores than T1. Their stem-like/memory score decreased. The T2-versus-T1 contrast placed IFN-response genes in the T2 direction. T2 cells expressed mixed T-cell and NK-like identity markers and checkpoint genes such as TIGIT and LAG3. The IL-15/STAT5 program was modest. Together, these features define a more cytotoxic, checkpoint-rich T-cell RNA state at T2.
T2 annotations included proliferating exhausted states and cytotoxic and exhaustion co-expression. Our similarity search found related CD8 states in a melanoma dataset and a head-and-neck cancer trial. These matches place Sid's T2 state in a broader map of checkpoint-associated CD8 biology.
The TCR data sharpen the picture: T2 and T3 share 498 clonotypes. The dataset also contains expanded and mutation-linked clones. Together, they form a focused candidate set for functional recognition and tumor-killing experiments.
By T2, antigen, IFN, cytotoxic, and exhaustion-related RNA programs had all risen. RNA, Xenium, spatial CD3D, and TCR counts converged on a T-cell-rich later state. T3 combines CD45-positive and CD45-negative preparations, so its captured fractions are sample-specific. The orthogonal assays support the same overall pattern.
T3, April 2025. “Which expression programs persisted or changed?”
Between T2 and T3, Sid received FAP-targeted radioligand therapy (177Lu-FAPi and 225Ac-FAPi). This therapy is intended to deliver radiation to FAP-expressing cells. Sid also received more SNK-01, dostarlimab, and intratumoral AdaPT-001. The T3 biopsy captures the combined state after this treatment interval. The case has no dosimetry, so radioligand delivery remains unmapped.
We scored the same macrophage programs across T1, T2, and T3: M2/reparative TAM, LAM/foamy TAM, IFN/APC TAM, and inflammatory chemokine. The T3 IFN/APC and inflammatory RNA scores increased from T2. The top-100 comparison contained four of 20 IFN/APC genes in the T3 direction. Spatial HLA-DRA and raw HLA ADT decreased from T2, revealing a mixed MHC picture across RNA, spatial, and protein layers.
The T3 T-cell exhaustion/counter-regulation score increased from T2. The tissue-resident and chronic-stimulation score increased slightly. The T3-versus-T2 contrast placed CTLA4 in the T3 direction. The stem-like/memory score flipped direction with label mapping: +0.0222 in the merged analysis and −0.0817 with exact standalone T-cell labels. Checkpoint ADT shifts were batch-wide and tracked higher isotype background, so they cannot confirm a protein-level exhaustion change. This makes exhaustion the clearest focus for T3 functional follow-up.
Stromal programs fell from T2 to T3. At T2, stromal cells had a matrix-CAF state. From T2 to T3, the FAP/CAF score decreased by 0.2677. The osteosarcoma program score decreased by 0.1595. In the truncated top-100 stromal contrast SPARC, one of 15 FAP/CAF genes, pointed toward T2. Some osteogenic genes, including RUNX2, BGLAP, and SPP1, increased despite the lower aggregate score. FAP detection decreased by 10.7 percentage points, and Xenium FAP also decreased. The signal is clear: FAP expression fell. Connecting that decline to FAPi now depends on tumor dosimetry and direct response measurements.
T3 held the T-cell-rich, IFN-active state while exhaustion RNA rose and both aggregate stromal programs moved down.
What the case revealed
Across three biopsies, the strongest signal is a coordinated transition. T1 is more reparative and antigen-low. T2 and T3 are more T-cell-rich and IFN-active. T3 adds stronger exhaustion RNA and lower stromal FAP/CAF and osteosarcoma programs. TCR, WGS, and HLA connect persistent clones to candidate antigens and create a focused set for functional follow-up. The result is a high-resolution map of longitudinal biology across combined treatment intervals, with focused hypotheses ready for functional testing.
Glass turned a fragmented multi-omic case into an explorable timeline. In a few clicks, we annotated cells, compared timepoints, scored programs, and connected the major findings to live UMAPs and gene-level evidence. Marble turned custom contrasts and literature search into one interactive workflow. This is the kind of analysis we built Glass for: fast, traceable, and ready to grow as new data arrives. The next test is scale: more patients, independent cohorts, and prospective outcomes.