Supplementary Figure S21 shows Venn diagrams of CNV-confirmed cancer cells and their coexpression of DLL3, SEZ6, B7H3 for tumors in Cohort C.

Supplementary Figure S24 shows the distribution of DLL3 positive cells based on cell type for Cohort B and patient 37.

Supplementary Figure S17 shows graphs of CD4+ T cells phenotypic profiles for memory, activation and exhaustion.

Supplementary Figure S5 shows single-cell transcriptional landscape of SCLC tumor biopsies and CNV for Cohort B.

Supplementary Figure S2 shows the CTC-iCHIP workflow and images of staining optimization using control cancer cell lines.

Supplementary Figure 7 shows CNV of SCLC tumor biopsies for patient-2.

Supplementary Data S1 contains characteristic descriptions for Cohort B.

Supplementary Figure S6 shows the CNV of SCLC tumor biopsies for Cohort C.

Supplementary Figure S9 shows that the fraction of DLL3-positive tumor cells does not significantly correlate with the distribution of SCLC molecular subtypes in primary tumors across cohorts.

Supplementary Figure S8 shows fraction of DLL3 negative tumor cells based on single-cell RNA-seq “drop out” of rare transcripts.

Supplementary Figure S12 shows isoform characterization of DLL3 including protein structure, expression and isoform profiles for specific patients.

The retinoblastoma protein (RB) is a well-characterized repressor of E2F transcriptional activity that controls genes involved in cell proliferation during the G1 phase. Here, we examine the effect of RB on chromatin organization and uncover a non-canonical role for RB in which it promotes the removal of cohesin from insulators and thereby induces the expression of adjacent genes. We identify that RB colocalizes with cohesin across the human genome. During mitosis, RB facilitates the removal of cohesin from CTCF sites, with this effect persisting into early G1, impacting loop formation and extrusion at topologically associating domain (TAD) boundaries. Chromosome conformation capture assays reveal that RB reduces insulation at TAD boundaries and promotes enhancer-promoter interactions marked by histone 3 lysine 27 acetylation. In this way, RB enhances the transcription of hundreds of genes beyond the E2F transcriptional program, a finding that expands our understanding of this key tumor suppressor and cell cycle regulator.

Supplementary Figure S4 shows the size distribution of the CTCs based on their marker expression profiles.

Supplementary Table S1 shows patient demographics and clinical characteristics for Cohort A.

Supplementary Figure S16 shows CD4+ T cells phenotypic profiling for individual markers in control sample HD2.

Markers of T-cell dysfunction in patients with acquired tarlatamab resistance despite persistent DLL3 expression. A, Box plot representing flow cytometric analysis of PBMCs showing the percentage of CD8+ (left) or CD4+ (right) live T cells. Blood specimens analyzed from healthy donors (HD: gray, N = 3), tarlatamab-untreated advanced SCLC (Pre-Tx: orange, N = 5), SCLC cases at the time of tarlatamab response (Resp.: blue, N = 4), and patients with SCLC with acquired resistance to tarlatamab despite persistent DLL3 expression in CTCs (Prog.: red, N = 2, open circles denote two distinct measurements from one case). For A–E and G–I, multiple measurements from a single patient (where available) were averaged for statistical comparison using the Student unpaired t test between Resp. vs. Prog. groups. B, Percentage of central memory (CM) CD8+ T cells (phenotype: CD45RA−CCR7+CD28+) versus effector memory subtype 1 (EM1) CD8+ T cells (phenotype: CD45RA−CCR7−CD28+). C, Percentage of KLRG1+ cells within activated (PD-1+) CD8+ T cells. D, Percentage of GZMB+ cells within activated (PD-1+) CD8+ T cells. E, Percentage of CD39+ cells within activated (PD-1+) CD8+ T cells. F, Percentage of GZMB+ cells within PD-1+CD8+ T cells, with and without CD39 expression, across all SCLC samples. Statistical comparison between CD39+ (dark red) and CD39− (gray) cells was done using the Student paired t test (N = 9 SCLC cases, including Pre-Tx, Resp., and Prog.). Samples with fewer than 50 recorded T cells in either the CD39+ or the CD39− group were excluded. G, Percentage of PD-1+ cells within total CD8+ T cells. H, Percentage of TIM3+ cells within activated (PD-1+) CD8+ T cells. I, Percentage of TOX+ cells within activated (PD-1+) CD8+ T cells.

Supplementary Figure S19 shows Venn diagrams of CNV-confirmed cancer cells and their coexpression of DLL3, SEZ6, B7H3 for tumors in Cohort B.

The establishment of distinct transcriptional states entails coordinated transcriptional coregulation of several functionally disparate biological processes (fd-BPs), involving thousands of genes scattered throughout the genome. Linear clustering of genes in a single BP is one strategy for their transcriptional coregulation. However, whether such gene clustering also plays a role in transcriptional coherence of several fd-BPs remains unexplored. Here, by analyzing the genomes of eukaryotes ranging from yeast to plants to humans, we find thousands of conserved and species-specific clustered fd-BP pairs, many of which are transcriptionally correlated in normal human tissues. Strikingly, our results reveal that often this system-level transcriptional coordination is achieved in part by the genic proximity of regulatory nodes of fd-BPs (such as transcription factors) whose coregulation drives the transcriptional coherence of their respective pathways. This, we hypothesize, is one strategy for creating coregulated, tunable modulons in eukaryotes.

Differential CTC expression of DLL3 upon acquired resistance to tarlatamab. A, Representative case (patient 1) showing total CTC enumeration and DLL3 staining in a DLL3High patient with initial profound response to tarlatamab, whose CTCs show complete loss of expression of the targeted epitope at the time of acquired resistance and disease progression. CTC counts are normalized to 20 mL, with the percentage DLL3-positive (orange) versus DLL3-negative (green) shown for each time point during the course of treatment (arrowheads: tarlatamab cycles). B, Another representative tarlatamab-responsive DLL3High case (patient 8) in which increasing numbers of CTCs at the time of acquired drug resistance show universally persistent expression of the targeted DLL3 epitope. C, Relative fraction of DLL3Pos patients who initially benefited from tarlatamab (PR or SD), whose CTCs at the time of cancer progression are either above (positive) or below (low) the median 25% cutoff for DLL3 expression (N = 5).

Supplementary Figure S13 displays a gallery of CTFs in specific patient samples following the onset of tarlatamab therapy.