Deep visual proteomics reveals DNA replication stress as a hallmark of signet ring cell carcinoma

Introduction

Signet Ring (SR) cell carcinoma (SRCC) is a rare and highly aggressive type of adenocarcinoma that can occur in multiple organs. While the stomach is the most common primary tumor site, SRCC has also been reported in the prostate, breast, lung, and bladder¹. Regardless of origin, it typically metastasizes rapidly to distant sites^2,3. Incidences of gastric SRCC have persistently increased over the last few decades^4,5.

If SRCC occurs from cells other than stomach glandular cells it may complicate disease classification in the affected organ^6,7,8. However, there is one pathological feature that characterizes SR cells as such: a high concentration of intracellular mucin that builds up in large vacuoles, pushing the nucleus to the periphery of the cell and giving it the distinctive shape of a signet ring⁹.

Despite clinical advances in gastric cancer classification, grading, and treatment, the SR cell carcinoma subtype remains a substantial clinical burden^9,10. Due to its rarity and a propensity for late symptom onset, SRCC patients are often diagnosed at an advanced stage, limiting treatment options and therapeutic efficacy^11,12. Surgical resection followed by postoperative chemotherapy and radiotherapy is the main management option for advanced disease¹³. However, these treatments have limited impact on overall survival and can have numerous negative effects that worsen patient well-being¹². The rarity of SRCC and the substantial knowledge gap regarding its fundamental biology and underlying signaling pathways thus combine to limit personalized therapeutic strategies for this distinct cancer subtype.

Investigations into SRCC biology have primarily revolved around this cancer’s inherently increased proliferation rate, characterized by aberrations of the RAS/RAF/MAPK¹⁴, HER2, or Wnt/β-catenin¹⁵ signaling pathways and mutation of the E-cadherin gene CDH1¹⁶. Microsatellite instability and strong lymphocyte infiltration have also been linked with colorectal SRCC, clinicopathological signatures typically rather associated with colorectal cancer than specifically with SRCC^17,18. It is also known that in colorectal SRCC, the SMAD complex triggers the epithelial-mesenchymal transition (EMT) in response to transforming growth factor (TGF)-β signaling, which accounts for the distinctive loss of epithelial cell junctions and polarity in SRCC of the colon^19,20.

So far, most research on SRCC has been limited to clinical observations, histological classifications^21,22, and obtaining genomic sequencing data specific to occurrences in affected organs^20,23, predominantly the colon. We reasoned that global molecular analyses at the protein level could contribute to elucidating the broader biological context and distinctive pathogenic mechanisms of SRCC. The spatial proteomics field has made significant strides in recent years, and is potentially able to address the above challenge^24,25. In particular, our group has developed the Deep Visual Proteomics (DVP) technology which combines high-resolution image acquisition with machine learning-guided segmentation and classification, followed by single-cell type-enriched high-sensitivity mass spectrometry (MS)-based proteomics²⁶.

In this study, we took a precision oncology approach by using DVP to examine SRCC in four different organs—the bladder, prostate, seminal vesicle, and lymph node—within a single patient. We reasoned that this spatial context would allow us to explore proteome differences and similarities of SR cells across tissues, offering valuable insights into tumor origin, potential mechanisms of metastasis and inform treatment recommendations.

Results

Patient Disease Background and Interventions

The patient was diagnosed with SRCC, with the bladder identified as the primary site of origin, following the removal of a suspicious mass on the bladder wall that was revealed by magnetic resonance imaging (MRI). Hematoxylin and eosin (H&E) staining of the mass revealed the typical signet ring morphology, and the patient was subjected to a radical cystectomy that removed the bladder (B.), prostate (P.), seminal vesicles (S.V.), and 14 lymph nodes (L.N.) (Fig. 1a; b). Post-surgery pathology of these organs revealed cells described as ‘adenocarcinoma with pure signet ring morphology’ in all organs and nine out of 14 lymph nodes. To enhance the therapeutic options for the patient, a genomic analysis was performed and a molecular tumor board report was filed, noting a microsatellite instability of only 0.8%, an ATRX (alpha thalassemia/mental retardation syndrome X-linked) frameshift mutation, MYCL and RICTOR (Rapamycin-insensitive companion of mTOR) amplification and KDM6A (Lysine-specific demethylase 6 A) biallelic loss. The patient underwent chemotherapy with a combination of oxaliplatin, which was discontinued after four months due to the onset of continuous neuropathy, and capecitabine, likewise discontinued after seven months, before being monitored by quarterly computed tomography (CT) scans (Fig. 1c). Twelve months after the cessation of chemotherapy, CT scans revealed suspicious enlargement of several lower abdominal lymph nodes. After further evaluation through a positron emission tomography (PET) scan, an accessible lymph node was removed by ultrasound-guided biopsy, in which pathology confirmed the presence of cells of signet ring morphology. The patient received a combination of immunotherapy with pembrolizumab, which is ongoing, and chemotherapy with carboplatin, which was again stopped after four months due to side effects (Fig. 1c). The tissues used in this study were obtained prior to any treatment.

**Fig. 1: Signet ring cell carcinoma samples and timeline of medical interventions.**

A simple stain allows robust segmentation and classification for the DVP workflow

For spatial proteomics we sectioned formalin-fixed, paraffin-embedded (FFPE) tissue blocks of all four organs (bladder, prostate, seminal vesicle, and one lymph node) at three µm thickness using a microtome and mounted the tissue sections on polyethylene naphthalate (PEN) membrane-coated microscopy glass slides (Fig. 2a). Tissues were stained with DAPI for nuclear visualization. A crucial step in DVP is delineation of the cell plasma membrane for subsequent laser microdissection (Fig. 2c). For our samples, we found that staining with wheat germ agglutinin (WGA), a lectin that binds to specific carbohydrates in the plasma membrane, was sufficient for this purpose (Fig. 2b). In comparison to other staining methods, such as cytokeratin 1 (CK1) or conventional H&E, WGA staining proved superior in terms of efficiency and simplicity. Continuing with the DVP pipeline, we imaged the tissue slides with a standard immunofluorescent microscope (Zeiss Axio) and processed images with the Biology Image Analysis Software (BIAS)²⁶ (Fig. 2a; b). For cell segmentation we fine-tuned a pre-trained model in BIAS. We trained a machine learning model for cell classification, which involved manual annotation of more than 1000 SR cells and lymphocytes from each organ to capture morphological diversity, ensuring accurate classification across tissue types (Fig. 2b). Prediction accuracy of SR cells was 95% based on 10-fold cross validation and independent validation by a pathologist. Shapes were subsequently exported to a second microscope for semi-automated laser microdissection (Leica LMD7) (Fig. 2c). A total of 500 cell shapes were laser microdissected from FFPE tissue per sample in triplicates, corresponding to approximately 50 full-sized SR cells, as our FFPE sections represent only a thin fraction of the cell volume. For each organ, 1500 cell shapes (equivalent to approximately 150 full SR cells) were collected, lysed and enzymatically digested for subsequent MS-based proteomics (Fig. 2d). Peptides were separated by the Evosep One chromatography system²⁷, coupled to the Orbitrap Astral™ mass spectrometer²⁴. This was followed by protein identification and quantification using the DIA-NN software²⁸ (Fig. 2e; see Methods).

**Fig. 2: Deep Visual Proteomics workflow on WGA-stained tissues.**

Proteomic analysis identifies organ-specific SRCC and DDR protein signatures

Analyzing MS data from the equivalent of 50 SR cells in all four organs, and including non-cancerous epithelial prostate cells as controls, we quantified a median of 6638 different proteins (Fig. 3a), with an excellent coefficient of variation (CV) of approximately 11% across the tissues (Fig. 3b). A total of 4648 proteins were present across all triplicates and organs, and 7157 in at least 70% of samples of each organ, indicating high completeness of our data set (Fig. 3c). Across the four organs, we identified 4825 proteins as a common core proteome (Fig. 3d). As expected, proteins uniquely present in each organ mirror specific organ functions, such as semenogelin-2 (SEMG2) in the seminal vesicle which is responsible for gel matrix formation for spermatozoa²⁹ (Fig. 3d).

**Fig. 3: Proteomic depth and signatures of signet ring cells across tissues.**

Principal component analysis (PCA) clustered samples originating from the same tissue, but also cluster the cancerous SR cells away from the control (Fig. 3e). Likewise, SR cells from the lymph node, prostate, and bladder were clearly distinct from SR cells of the seminal vesicle (Fig. 3e). Well-known markers for prostate cancer, along with proteins associated with prostate cancer and involved in EMT, including dipeptidyl peptidase 4 (DPP4), transglutaminase 4 (TGM4), keratin 7 (KRT7), acid phosphatase 3 (ACP3), kallikrein-related peptidase 3 (KLK3) and solute carrier family 45 member 3 (SLC45A3)^{30,31,32,33,34,35}, were among the proteins driving the separation between SRCC and epithelial control in our PCA along loading component 1 (Fig. 3f)³⁶. Proteins that are notably enriched in the SR cells compared to the epithelial control cells include carcinoembryonic antigen-related cell adhesion molecule 5 (CEACAM5) and CEACAM6, mucins (MUC2, MUC5B), and calcium-activated chloride channel regulator 1 (CLCA1), which are classical markers for SRCC (Fig. 3f). Fatty acid binding protein 4 (FABP4) and glycerol-3-phosphate dehydrogenase 1 (GPD1) separate SR cells from those in other organs through principal component 2, likely due to tissue-specific differences in cellular proteomes, function, and interactions between SR cells and their tumor environment (Fig. 3f). Thus, DVP recapitulated expected or recently described physiological patterns while adding novel molecular players.

In the prostate, there was a clear and significant enrichment of MUC and CEACAM proteins between the epithelial control and SR cells (Fig. 3g). In contrast, we observed minimal differences in the levels of prostate and prostate cancer-associated proteins, including KLKB1, KLK2, KLK3, ACP3, and SLC45A3, which are markers of conventional adenocarcinoma of the prostate (Fig. 3g). This indicates that, despite originating in the prostate, the SR cells possess a distinct proteome consistent with SRCC morphology rather than typical prostate cancer. To control for SRCC-specific protein patterns and to investigate proteins with the most significant differential changes, we focused on two well-known protein families strongly associated with SRCC, mucins and CEACAMs.

MUC1, MUC2, and MUC13 showed the strongest – up to ten-fold – and most consistent enrichments in SR cells across all organs compared to the epithelial control cells of the prostate (Fig. 3h). MUC1 and MUC2 are already well known to be overexpressed in gastric cancers, however, MUC13, a transmembrane mucin, might play an as-yet-unknown role in cell signaling and epithelial barrier protection. MUC4, MUC5AC, MUC5B, and MUC12 had significant but fluctuating fold-changes between organs. SR cells in the seminal vesicles exhibited protein levels similar to those of non-cancerous control cells in the prostate. Mucin-like 1 protein (MUCL1) has structural similarities and glycosylation patterns to classical mucins, but, interestingly, its expression was not significantly changed across all tissues analyzed, demonstrating that changes and overexpression in SR cells are specific to classical mucins.

Regarding the CEACAM family, CEACAM5 and CEACAM6 expression increased up to ten-fold between cancerous and epithelial controls, with the sole exception of CEACAM6 in the SR cells of the seminal vesicle. CEACAM1 and CEACAM21, which have different functions and structures, remained uniform across the different tissues, supporting the notion that they are not directly involved in carcinogenesis³⁷.

We next asked if the proteins highly enriched in prostate SR cells could point us to any therapeutically relevant pathways. Indeed, the top pathways identified in gene ontology (GO) enrichment analysis, based on fold-change and statistical significance, were all related to DNA replication and DNA damage response (DDR), including ‘nucleotide excision repair (NER)’, ‘base excision repair (BER)’ and ‘mismatch repair (MMR)’ (Fig. 3i). We additionally found that the enrichment of MMR pathways is universal to all SR-positive tissues. The majority of the constituent proteins were upregulated, however, a number of prominent replication proteins (RPAs) were substantially downregulated (Fig. 3j).

Signet ring cells exhibit multiple DDR pathway deficiencies across organs

Following up on our observation that proteins of the DDR showed abundance changes between prostate SR cells and epithelial cells, we next investigated tissue-specific protein changes by correlating the fold-changes between them (Fig. 4a). Comparing two tissues at a time, we observed that Ly6/PLAUR domain-containing protein 8 (LYPD8) and UDP-glucuronosyltransferase 2B17 (UGT2B17) showed similar patterns to the aforementioned CEACAM5 and CEACAM6 proteins. LYPD8 is also involved in epithelial cell junction integrity, pointing to a dysregulation of cell-cell adhesion, as well as potential deficiencies in tissue protection. UGT2B17 is involved in the metabolism of steroid hormones and xenobiotics, which can alter the tumor microenvironment.

**Fig. 4: Proteomic profiling of signet ring cells in the context of DNA damage.**

S100 calcium binding protein P (S100P), MUC2, and CLCA1 also had similar expression patterns across the tissues (Fig. 4a), in line with CLCA1 affecting mucin secretion through Ca²⁺ signaling and its possible implications in cancer pathophysiology³⁸. KLK3 and TGM4, well-known prostate-associated proteins, consistently exhibited minimal or no fold change between tissues and non-cancerous control cells (Fig. 4a). Thus, signet ring cells may arise due to molecular mechanisms distinct from those of conventional prostate adenocarcinoma and metastases.

To globally examine protein patterns prevalent across all SR cells and contrast them with epithelial cells as a control, we performed unsupervised hierarchical clustering on the 1560 ANOVA significant proteins, which revealed two prominent clusters, those upregulated or downregulated with respect to control (upper, red cluster and lower, blue cluster in Fig. 4b).

We performed GO term enrichment analysis on the upregulated cluster using Reactome, NetPath, and Biological Processes, which highlighted diverse pathways active in SRCC cells. These included Wnt, leptin, epidermal growth factor (EGF) receptor and transforming growth factor β (TGFβ) receptor pathways, all well-known for their roles in various carcinomas including stomach, colorectal and SRCC, (Fig. 4c). Apart from these, the most prominent pathways were again associated with DNA replication and DDR (Fig. 4c).

Next, by comparing the SR cells to the epithelial control cells, we ran a gene set enrichment analysis (GSEA) on proteins that showed a significant enrichment following pairwise proteomic comparison. Remarkably, seven of the top ten pathways are part of DDR, namely ‘Activation of [the] pre-replicative complex’, ‘Activation of ATR (ataxia-telangiectasia mutated and Rad3-related), a pathway triggered by perturbations affecting DNA replication dynamics characterized by replication stress (RS)’, ‘PCNA-dependent long patch base excision repair (LP-BER)’, ‘gap-filling DNA repair synthesis and ligation in global-genome nucleotide excision repair (GG-NER)’, and ‘DNA strand elongation’ (Fig. 4d; e).

To validate our proteomic results regarding ATR signaling activation, we stained all SRCC-positive tissues for phospho-ATR (pATR), the activated form of the protein kinase which phosphorylates downstream key proteins involved in DDR^39,40,41,42. Our staining results confirmed the presence of pATR across our tissue samples, with the highest positivity observed in the seminal vesicle tissue (Fig. 4f).

Pathways implicated in metabolic processes such as ‘glycogen metabolism’ and signaling mechanisms such as the ‘Ca²⁺ pathway’ and ‘G-protein beta:gamma signaling’ were negatively enriched (Fig. 4d; e). Downregulation of these pathways in SR cells likely reflects metabolic reprogramming of cancer cells, alterations in calcium signaling to support uncontrolled growth and survival, and specific adaptations of SRCC to facilitate mucin production and secretion.

Given the observations of significant changes in protein abundances related to DDR pathways and the ATR signaling axis in SR cell-positive tissues compared to epithelial control cells, we further investigated proteins involved in stalled replication fork (RF) protection and repair of complex DNA lesions formed in the event of replication fork collapse, a key part of the cellular response to DDR. These included proteins of the Fanconi Anemia (FA) pathway specifically the FA group D2 protein (FANCD2) and its interactor⁴³, Fanconi Anemia complementation group I (FANCI)⁴⁴. Additional mediators of the same process including DNA unwinding RecQ-like helicase 5 (RECQL5), Werner syndrome helicase (WRN), and helicase-like transcription factor (HLTF) all displayed a similar positive fold change (Fig. 4g). Our data provide strong indications of an ongoing RS and of the subsequent response of the SR cells to maintain their genomic stability by upregulating various RF protection mechanisms.

Upon persistent RS and prolonged RF stalling, replisome structure is impaired and RFs collapse, leading to the emergence of single-end double strand breaks (seDSBs), the most deleterious form of DNA lesions. Cells then trigger the highly error-prone break-induced replication pathway (BIR) to deal with this threat^45,46. GSEA on our proteomic data showed a significant enrichment of this mutagenic pathway (Fig. 4d). Moreover, DNA polymerase delta subunit POLD3, an essential subunit of DNA polymerase delta upon BIR, together with POLD2 and DNA polymerase epsilon (POLE) showed a positive fold change enrichment comparing SR cells of the prostate, the seminal vesicle, the lymph node, and the bladder to the epithelial control (Fig. 4g).

APOBEC3s, members of the Apolipoprotein B mRNA-editing enzyme catalytic polypeptides (APOBECs) superfamily, exhibit overexpression across various cancer types, notably bladder^47,48,49 and prostate cancer^50,51. The induced hyper-mutations of long stretches of single-strand DNA (ssDNA) formed during BIR (with APOBEC3A and APOBEC3B being the major mutators) through deamination, foster genome instability in cancer cells, a phenomenon referred to as ‘Kataegis’. However, proteins of the APOBEC3 family of enzymes were markedly reduced in abundance in SR cells of every tissue (Fig. 4g), possibly as a protective feedback mechanism to mitigate the mutational burden and maintain genomic stability^52,53,54,55.

Collectively our analysis of the proteome changes of SR cells from the bladder, identified as the primary tumor site, as well as from metastatic sites, namely the prostate, seminal vesicle, and lymph node revealed consistent patterns of a severe dysregulation of multiple DNA repair mechanisms, with a potential negative impact on genome integrity.

Enrichment of Complement System and PD-1 Signaling Proteins in Signet Ring Cells result in a cytotoxic T lymphocyte infiltration

Mutations and expression profiles of DDR genes have recently been associated with alterations in immune regulatory gene expression and CD8-positive T cell infiltration in the tumor microenvironment, serving as a predictive marker of immune checkpoint blockade (ICB) therapy efficiency^56,57. We therefore hypothesized that our unique protein signatures could indicate a higher immunogenicity and a greater mutational burden in SRCC. The Reactome-curated ‘Complement system’ pathway displayed a positive fold change across all tissues, with the most marked increases seen in the C1q subcomponent subunits A (C1QA), B (C1QB), and C (C1QC) (Fig. 5a). A similar expression pattern was observed in immunoglobulins and proteins involved in the programmed cell death protein 1 (PD-1) signaling pathway (Fig. 5a).

**Fig. 5: SRCC shows immunogenicity and cytotoxic T cell infiltration.**

To confirm our hypotheses derived from our proteomic analyses, which pointed to tumor immunogenicity and DDR pathways, we immunostained for PD-1 and CD8-positive cytotoxic T cells in bladder tissue (primary tumor site), and in lymph nodes (metastasis) (Fig. 5b). These tissues exhibited moderate to substantial infiltration of PD-1-positive T cells (Fig. 5b), which we further confirmed through quantification of the PD-1 signal on CD4- and CD8-positive T cells (Supplementary Fig. 1). We also observed a pronounced upregulation of programmed cell death ligand protein 1 (PD-L1) on SR cells of the bladder (Fig. 5c). This suggests that immunotherapy, particularly PD-1/PD-L1 inhibitors, could be a promising therapeutic approach for targeting these tumors.

In line with our findings, the PD-1 inhibitor pembrolizumab had indeed been recommended and administered as a therapy following recurrence rather than chemotherapy. Our results indicate that the latter would have been unlikely to be effective while having the usual adverse effects. Initiated in 2022, the pembrolizumab ICB therapy on our patient has successfully halted tumor progression, with MRI scans conducted quarterly confirming tumor stasis.

Based on our results, we propose a model in which DNA damage repair mechanisms and replication stress responses take center stage in SRCCs (Fig. 5d). These SR cells hyperactivate the epidermal growth factor receptor (EGFR) pathway, leading to hyper-proliferation. Increased DNA replication, combined with defective MMR, results in numerous unrepaired post-replication DNA lesions across the genome. The repair of these lesions relies on excision repair mechanisms, including base excision repair (BER) and nucleotide excision repair (NER), which we observed to be upregulated at the protein level. Increased occurrence of such lesions, along with a higher rate of DNA replication, are major drivers of replication stress, activating the ATR signaling pathway. SRCC cells appear to respond to this stress by upregulating proteins involved in stalled replication fork protection and collapsed replication fork repair, as indicated by our proteomic analysis. Additionally, the elevated number of DNA lesions resulting from these alterations, combined with DNA damage induced by chemotherapy, may activate the cyclic GMP-AMP synthase (cGAS)-stimulator of interferon genes (STING) pathway, transforming the tumor microenvironment into an immunologically ‘hot’ state and enhancing responsiveness to immune checkpoint inhibitors⁵⁸.

Discussion

In this study, we employed Deep Visual Proteomics (DVP) to investigate the proteome landscape of Signet Ring Cell Carcinoma (SRCC) across primary and metastatic sites from a single patient. Our analysis of approximately 50 SR cells per organ yielded up to 7700 proteins, providing unprecedented insights into the tumorigenic properties and potential signaling pathways of SRCC.

We identified both shared and organ-specific protein patterns in SR cells, with a clear distinction from normal epithelial control cells. Key drivers of this difference include mucins, CLCA1, CEACAM5, and CEACAM6. Mucins, particularly MUC1, MUC2, and MUC13, showed significant enrichment in SR cells across all organs, directly contributing to the characteristic signet ring morphology⁵⁹. CLCA1, closely linked to mucin production, can significantly alter the tumor microenvironment, affecting cell adhesion and migration^38,60. The upregulation of CEACAM5 and CEACAM6, immunoglobulin-related glycoproteins and adhesion molecules, is notable. While CEACAMs are known to facilitate cellular connection and are frequently elevated in various cancers^61,62,63, their specific role in SRCC has not been previously emphasized. Their overexpression may contribute to the distinctive morphology and aggressive behavior of SRCC through promotion of invasion and metastasis^64,65,66,67.

Our data revealed significant alterations in DNA damage response (DDR) pathways across SR cells in different organs. We observed changes in excision repair mechanisms, including DNA mismatch repair (MMR), base excision repair (BER), and nucleotide excision repair (NER). The upregulation of most MMR proteins, coupled with the downregulation of MLH1, suggests a defective MMR pathway, consistent with previous studies linking microsatellite instability to colorectal SRCC^17,18.

A key finding of our study is the upregulation of the ATR signaling axis, indicating ongoing replication stress – a recognized hallmark of cancer-driven genome instability⁶⁸. Our proposed model suggests that replication fork stalling and collapse result from an increasing load of post-replicative lesions combined with increased proliferation and DNA replication rates. In response to this stress, the ATR signaling pathway is activated, triggering mediators of stalled replication fork protection, and collapsed replication fork repair and restart (Fig. 4g; f; 5d)⁶⁹. Single-ended double-strand breaks, the most deleterious form of DNA lesions formed upon replication fork collapse, are addressed by the break-induced replication (BIR) pathway^70,71. We demonstrated that BIR is upregulated in SR cells across all four organs examined. The error-prone nature of BIR has been associated with high mutation rates, gross chromosomal rearrangements (GCRs), and loss of heterozygosity, further fostering genomic instability^52,72.

In line with this model, SR cells exhibited a considerable increase in the abundance of poly(ADP-ribose) polymerase (PARP), a key player in DNA repair, and a decreased abundance of APOBEC (apolipoprotein B mRNA-editing catalytic polypeptide-like) enzymes compared to adjacent non-tumorigenic epithelial cells. This protein profile suggests a complex interplay between DNA damage accumulation and repair mechanisms in SRCC.

The activation of DDR and repair pathways likely contributes to the high mutation rate and genomic instability observed in SRCC, which is linked to microsatellite instability and increased tumor immunogenicity. This provides a mechanistic explanation for the observed enrichment of immune-related protein signatures and the efficacy of immunotherapy^17,18, as confirmed by immunofluorescence imaging showing T lymphocyte infiltration and PD-1 expression²². Alterations in the DDR pathways observed in our proteomic analysis align with known genomic changes in SRCC. Specifically, mutations in the ATRX gene locus, a critical regulator of chromatin remodeling and telomere maintenance, are associated with replication stress and the activation of the alternative lengthening of telomeres (ALT) pathway^73,74,75. ATRX mutations also influence immune pathways by modulating the tumor microenvironment, promoting immune cell infiltration while upregulating immune checkpoint proteins like PD-L1 and PD-L2, which may facilitate immune escape^58,76. These findings provide a mechanistic link between DDR alterations and the immunogenic profile of SRCC, highlighting their potential role in the observed response to combination therapy with pembrolizumab and chemotherapy^77,78. Additionally, alterations related to the complement cascade pathway, as previously reported⁷⁹, were observed in SRCC tissues, further highlighting the complex interplay of immune signaling and tumor biology. However, immune infiltration and PD-1 signals varied across organs. While pembrolizumab achieved tumor stasis for 20 months, exceeding typical SRCC survival times⁸⁰, tumor recurrence in the neobladder may reflect insufficient T cell responses in this organ. These findings demonstrate how SRCC can adapt to and escape from therapy, underscoring the need for dynamic proteomic profiling to guide treatment adjustments.

Our results provide a rationale for the observed clinical response to pembrolizumab immunotherapy in this patient^57,81,82,83, despite initial sequencing results showing only 0.8% unstable microsatellite sites. This highlights the potential of proteomic analysis in guiding treatment decisions, especially in cases where genomic data alone may not fully capture the tumor’s biology. The identification of replication stress as a central feature of SRCC opens new avenues for targeted therapies. Our findings suggest that targeting the ATR pathway or exploiting vulnerabilities in DNA repair mechanisms could be promising strategies. Additionally, the overexpression of CEACAMs points to potential targets for antibody-drug conjugates or other targeted therapies.

This study highlights the power of spatial proteomics in uncovering the molecular intricacies of rare cancers such as SRCC. By providing a comprehensive view of the proteome across different organs, this approach identified common features of SR cells that transcend tissue origin, as well as organ-specific adaptations. DVP provided unique insights into tumor biology by tracking proteomic signatures to specific tissue locations while preserving cell identity and tissue integrity, which may not be achievable with genomic or transcriptomic analyses alone. Because this analysis is based on a single patient, limiting its generalizability, larger studies involving diverse cohorts will be essential to validate these findings and explore the diagnostic and therapeutic potential of this approach. Additionally, functional assays targeting disrupted proteins of the DNA repair pathways could provide deeper insights into their molecular functionality, complementing the proteomics data and advancing the understanding of SRCC. Thus, future studies integrating proteomic data with genomic and transcriptomic profiles could yield even more comprehensive insights into SRCC biology, potentially leading to improved diagnostic and therapeutic strategies for this aggressive cancer subtype.

In conclusion, our DVP-based analysis of SRCC reveals a complex interplay of DNA damage response, replication stress, and immune signaling pathways, offering insights that suggest potential therapeutic strategies. These findings deepen our understanding of SRCC biology and elucidate the success of pembrolizumab therapy, which halted tumor progression for an extended period. This retrospective explanation by our proteomic data underscores the potential of precision oncology approaches guided by comprehensive molecular profiling. A recent CT scan identified a new tumor in the neobladder, potentially linked to low T cell infiltration and PD-1 expression observed in the original bladder sample. Residual SR cells from the metastasis of the original bladder tumor, observed in abdominal CT scans in 2022, which are possibly less susceptible to immunotherapy, may have regrown and metastasized to the neobladder. Local factors, such as urine exposure or tumor location in the neobladder space, might further hinder antibody access. Despite these challenges, the patient remains active and continues treatment, highlighting how molecular insights can guide therapy for aggressive cancers like SRCC.

Methods

Study design and ethical permission

This single-patient case study was conducted after direct patient contact and full, informed consent (HREC reference: UTS ETH22-7236). Following consultation with the Nepean Blue Mountains Local Health District Human Research Ethics Committee (HREC), no formal application was required, as there was no clinical intervention or risk to privacy. The patient provided active consent for the storage and future use of personal data and biological materials, the sharing of data with research institutions within and outside the EU/EEA, and the inclusion of personal data in scientific publications without anonymization. Additionally, the patient consented to the public release of their proteomics raw data on PRIDE. Approval for histological slides and tissue blocks was granted by the Director of Pathology, and communication from the Nepean Blue Mountains Local Health District serves as evidence of the waiver of the need for further HREC approval. All procedures in this study are consistent with the principles outlined in the Declaration of Helsinki.

Immunohistochemistry and high-resolution microscopy

A detailed protocol for FFPE tissue mounting and staining on membrane PEN slides 1.0 (Zeiss, 415190-9041-000) is provided in the original Deep Visual Proteomics (DVP) article²⁶.

The tissue sections were initially subjected to deparaffinization and hydration through three cycles involving xylene and decreasing ethanol concentrations from 99.6% to 70%. For Wheat Germ Agglutinin (WGA) labeling, sections on membrane PEN slides were incubated with WGA staining solution (Biotium, 29023; diluted 1:1000) in a light-protected environment at 37 °C for 10 min. For pan-cytokeratin (CK), CD8, PD1, PDL1 and pATR staining, antigen retrieval was achieved by immersing the tissue sections on glass slides in EDTA buffer (Sigma, E1161; pH 8.5) at 90 °C for 30 min. Following this, the tissue sections were blocked with TBS protein-free blocking buffer (LI-COR, 927-80000) for 20 min at room temperature. For CD8/PD1/CK triple staining, the sections underwent overnight incubation at 4 °C with anti-CD8 antibody (Abcam, ab17147; 1:100), followed by slide washing and subsequent incubation with Alexa Fluor® 647 goat anti-mouse antibody (Invitrogen, A-21235; 1:1000) for one hour at room temperature. After rinsing, the slides were further incubated overnight at 4 °C with anti-PD1 antibody (Miltenyi Biotec, 130-117-384; 1:100) and anti-CK antibody (Invitrogen, 53-9003-82; 1:500). For PDL1/CK double staining, slides were incubated with anti-PDL1 antibody (Invitrogen, 12-5983-42; 1:100) and anti-CK antibody (Invitrogen, 53-9003-82; 1:500) overnight at 4 °C. For pATR staining, slides were incubated with anti-pATR antibody (GeneTex, GTX128145; 1:500) overnight at 4 °C, followed by slide washing and subsequent incubation with Alexa Fluor® 647 donkey anti-rabbit antibody (Invitrogen, A-31573; 1:1000) for one hour at room temperature. Finally, we used DAPI (Abcam, ab228529; 1:1000) for nuclear counterstaining for 5 min at room temperature and the slides were mounted with Anti-Fade Fluorescence Mounting Medium (Abcam, ab104135) before examination under an AxioScan7 microscope (Zeiss, for WGA, CK, CD8, PD1, and PDL1 imaging) or PANNORAMIC 250 Flash III (3Dhistech, for pATR imaging).

Tumor regions were identified using CK staining, WGA staining, or simply by including the autofluorescent signal of mucin.

Cell segmentation and classification

Microscopy images were imported into BIAS (Biology Image Analysis Software, single-cell-technologies.com), for machine learning-based cell segmentation, classification, and subsequent single-shape export for semi-automated laser microdissection. For SR cell segmentation, we utilized a pre-trained deep neuronal network on our IF WGA-stained tissues. Detection confidence was set to 60% and the contour confidence to 20%. Cell shapes with a larger area than 1000 µm² were excluded. To accurately classify SR cells, we trained a BIAS-integrated multilayer perceptron (MLP) feedforward neural network on manually identified SR cells across all four tissues. We set the weight scale and the momentum parameter to 0.01, and the number of iterations to 10,000. Subsequently, reference points were set, and SR cell contours were exported for semi-automated laser microdissection²⁶.

Image quantification

Tissue sections immunofluorescently stained for CD4, CD8, and PD-1, with a DAPI nuclear counterstain, were imaged using an AxioScan7 microscope at 20x magnification, resulting in .czi format files. The images were analyzed in QuPath software. A full-image annotation was first created to encompass the entire tissue section, and the ‘Positive Cell Detection’ algorithm was applied with the following settings: detection image set to optical density sum, requested pixel size of 0.5 microns, sigma at 1.5 microns, minimum cell area of 5.0 microns, maximum cell area of 200.0 microns, detection threshold of 100, maximum background intensity set to 1.0, shape splitting enabled, nucleus inclusion enabled, boundary smoothing enabled, and measurement collection enabled. All other parameters were left as default. The analysis was performed in batch mode and exported as ‘detection’ data, allowing extraction of mean cellular intensity values from CD4-positive or CD8-positive T cells expressing PD-1 for subsequent quantification.

Laser microdissection

After aligning the reference points using the LMD7 (Leica) microscope, we imported the shape contours to facilitate semi-automated laser microdissection, which was conducted with the following parameters: laser power at 34, aperture set to one, cutting speed at 28, the middle pulse count to three, final pulse to one, head current at 47%, pulse frequency at 2600 Hz, and an offset of 180. For each type of organ tissue, SR cell shapes were excised in triplicates, and collected into 384-well plates, deliberately omitting the outermost rows and columns. After microdissection, we spun down the plate at 1000 g for 10 min, and the dissected cell shapes were preserved by freezing at –20 °C for later processing.

For the MS analysis, a total of 500 cell shapes were laser microdissected from FFPE tissue per sample, corresponding to approximately 50 full-sized SR cells due to the reduced volume represented by FFPE sections. For each organ, 1500 cell shapes were collected as triplicates, with 500 shapes pooled for a single MS injection. Based on literature and our measurements during laser microdissection, SR cells exhibit variable shapes and sizes. Nevertheless, the dimensions of the excised shapes consistently represented about one-tenth the volume of a full SR cell. This approximation allows for sufficient sample collection and analysis across different tissues.

MS sample preparation

The entire MS sample preparation protocol was adapted from the original DVP paper²⁶. After protein digestion, samples were vacuum dried, resuspended in 20 µL Evosep Buffer A (0.1% formic acid v/v) and directly loaded on Evotips (https://www.evosep.com/).

LC-MS

Subsequently after Evotip loading, our low input samples were analyzed on our Orbitrap Astral mass spectrometer (Thermo Fisher Scientific) connected to the EvoSep One chromatography system (https://www.evosep.com/). We utilized a commercial analytical column (Aurora Elite TS, IonOpticks) and an EASY-Spray™ source to run our samples with the 40 Samples Per Day (’40 SPD’) method (31-min gradient). All samples were recorded in DIA (data independent acquisition) mode. The Orbitrap analyzer of the mass spectrometer was utilized for full MS1 analyses with a resolution setting of 240,000 within a full scan range of 380–980 m/z. The automatic gain control (AGC) for the full MS1 was adjusted to 500%. For the acquisition of our low-input FFPE DVP samples, we set the MS/MS scan isolation window to 3 Th (200 windows), the ion injection time (IIT) to 5 ms, and the MS/MS scanning range to cover 150−2000 m/z. Selected ions were fragmented by higher-energy collisional dissociation (HCD)⁸⁴ at a normalized collision energy (NCE) of 25%.

MS data analysis

Raw files were first converted to the mzML file format using the MSConvert software (https://proteowizard.sourceforge.io/) from Proteowizard, keeping the default parameters and selecting ‘Peak Picking’ as filter. Afterwards, mzML files were quantified in DIA-NN²⁸ (version 1.8.1) using the FASTA (2023, UP000005640_9606, with 20,594 gene entries) from the UniProt database and a direct-DIA approach. The enzyme specificity was set to ‘Trypsin/P’ with a maximum of two missed cleavages. Parameters for post-translational modifications were set to including N-terminal methionine excision, methionine oxidation and N-terminal acetylation were all activated, and a maximum of two variable modifications were allowed. Precursor FDR was set to 1%, and both mass and MS1 accuracy were set to 15 ppm. ‘Use isotopologues’, ‘heuristic protein inference’, ‘no shared spectra’ and ‘match between runs’ (MBR) were enabled. Protein inference was set to ‘genes’ and the neural network classifier run in ‘single-pass mode’. We chose the ‘robust LC (high precision)’ as quantification method and a retention time-dependent cross-run normalization strategy. SRCC samples, microdissected across all four organs, were searched together.

Bioinformatic analysis

After quantification in DIA-NN, the protein group matrix was imported into Perseus⁸⁵, and samples were annotated according to the organ of origin (bladder, lymph node, prostate, and seminal vesicle). Proteins with 70% of quantitative values present ’in at least one group’ were then kept for imputation of missing values based on their normal distribution (width = 0.3; downshift = 1.5). Further, all statistical tests were corrected for multiple hypothesis testing, applying a permutation-based false discovery rate (FDR) cutoff either 5% or 1%.

Gene Set Enrichment Analysis (GSEA) was conducted using Python (version 3.9.7) and the GSEApy package (documentation: https://github.com/zqfang/GSEApy, version 1.0.4).

For the purpose of data visualization, our analyses were performed using the Python programming language (version 3.9.7), and essential libraries such as NumPy (version 1.20.3), Pandas (version 1.3.4), Matplotlib (version 3.4.3), and Seaborn (version 0.12.2). Additionally, the ShinyGo web tool (documentation: http://bioinformatics.sdstate.edu/go/), version 0.77, was used to perform gene ontology (GO) term enrichment analysis.

No custom code was developed for this study; only minor adjustments, such as parameter tuning, file path modifications, and format adaptations, were made. The modified code originates from the open-source Python package AlphaPeptStats (Github repository: https://github.com/MannLabs/alphapeptstats)⁸⁶.