Metastasis and recurrence patterns in the molecular subtypes of urothelial bladder cancer

Urothelial cancer of the urinary bladder frequently metastasizes to lymph‐nodes, lungs, liver and bone. A taxonomy for molecular classification exists, but it is unknown if molecular subtypes show tropism for different organs. Here, we study 146 patients with de novo metastatic disease or recurrence after curative treatment. We classify primary tumors using two transcriptomic methods and immunostaining and identify enrichment and depletion of metastatic sites in molecular subtypes using permutation tests. We observed significant depletion of bone metastases in the Basal/squamous molecular subtype, whereas the Urothelial‐like subtype entailed an enrichment for metastases to bone. The Genomically unstable subtype was depleted of lung metastases, but enriched for atypical sites, including six out of seven patients with brain metastases. Stroma‐rich primary tumor samples were associated with local recurrence, but not with distant sites. Additionally, the proportion with brain or testis metastases differed between systemic chemotherapy regimens (GC vs MVAC) suggesting a sanctuary effect. In conclusion, molecular subtypes of urothelial bladder cancer are significantly associated with specific metastatic sites, suggesting that subtype‐specific molecular determinants could exist at various steps in the metastatic cascade.


| INTRODUCTION
Different cancer types metastasize preferentially to specific distant organs, and the most common sites of distant metastases are lungs, liver and bone. 1 Such metastatic patterns reflect anatomy, for example, blood and lymphatic vessel routes, but also the cancer cells' intrinsic affinity for specific microenvironments, that is, organotropism. 2 For urothelial cancer (UC) of the bladder no studies have systematically related the primary tumors' molecular profiles to the patients' metastatic sites, and the molecular underpinnings of organotropism are unknown.7][8] The subtypes capture discrete transcriptional states, for example, basal, mesenchymal and urothelial differentiation 6,9 associated with several genomic driver alterations, including mutations or copynumber alterations in CCND1, CDKN2A, ERBB2, FGFR3, RB1 and TP53. 7,8,10,11The most prevalent distant metastatic sites in UC are bone, lung and liver, which are estimated to make up 24% to 18% each, although lymph-node and local spread is also frequent (25% and 23%, respectively). 12,13Autopsy-studies have confirmed that these are the most frequent sites and revealed a strong association between lymph-node and distant metastases. 14ry few studies of UC have included both RNA-profiling of the primary tumor and information on metastatic sites.One study described nine patients with recurrences to bone after neoadjuvant chemotherapy (NAC) in the highly stroma infiltrated p53-like subtype, but not in the luminal-like or basal-like subtypes. 15Another study investigated RNA-profiles of metastatic biopsies from various sites but did not include primary tumors or link subtypes to sites. 16In the current study, we aim to establish if UC molecular subtypes are associated with different sites of recurrence and metastases, and to explore if such associations are influenced by other molecular or clinical factors.

| Patients and tumor samples
Patients with recurrent metastatic or primary metastatic bladder cancer were included in the study.The 146 patients in the study population include 57 patients with a recurrence from a previously published cohort of patients treated with ≥2 cycles of pre-operative cisplatin-based chemotherapy and radical cystectomy (RC) for muscle-invasive bladder cancer. 17 S1.The most frequent chemotherapy regimens were GC (gemcitabine, cisplatin) and MVAC (methotrexate, vinblastine, adriamycin and cisplatin) administered to 77 (53%) and 58 (40%) patients, respectively.Other regimens were used in nine patients (6%) including palliative treatment with gemcitabine only (three), a sequence of MVAC with subsequent ifosfamide and docetaxel (two), MVAC with subsequent gemcitabine (one) and Gemcitabine plus carboplatin (two).The three patients treated with curative radiochemotherapy received either GC (two) or cisplatin (one).All patients were treated within the Southern health care region (n = 63, 43%) or the Stockholm health care region (n = 83, 57%), and 38 palliatively treated patients (55%) had previously undergone a RC.All recurrence sites, and all metastatic sites for patients with primary M+ disease at diagnosis were recorded by clinical follow-up with chart review.Recurrence sites and metastatic sites were grouped as follows; "local", "lymphnode", "lung", "liver", "bone" and "other."Patient's age, sex and TNM-stage distribution at start of systemic treatment is shown in Table 1.

| RNA extraction and gene expression profiling
Data from 57 patients included in Sjödahl et al 17 GSE169455, and the 89 previously unpublished tumors are available via GSE222073.
Gene expression profiles were generated by first staining formalinfixed, paraffin embedded tissue blocks with hematoxylin and eosin.
A representative tumor-rich area free of necrosis and cauterization artifacts was selected for macrodissection.Depending on the size of the area, four to ten 10 μm sections were used for RNA isolation with the FFPET High Pure RNA isolation kit (Roche).Digestion with proteinase K was performed overnight at 56 C. Purified total RNA was amplified, labeled (pico kit, Thermo Fisher), and hybridized to Affymetrix Human Gene ST 1.0 microarrays (Thermo Fisher).Raw data was subjected to quantile normalization and adjustments for labeling batch using ComBat. 18Molecular subtype classification was performed by applying a single-sample Lund taxonomy (LundTax) classifier based on gene pair rules 19 to the raw expression data for each case.The MIBC Consensus single sample transcriptomic classifier was also applied. 8

| Molecular classification by immunohistochemistry
Two 1.0 mm tissue cores from the same tissue block used for RNA extraction was embedded in tissue microarrays and stained with a panel of 13 subtype-specific antibodies as previously described, 6 and with anti-p53 (Clone DO-7, M700101-2, Agilent).Stained cores were evaluated semi-quantitatively based on intensity and/or percentage of cancer-cell staining and were classified according to pre-specified LundTax subtype scores and cut-offs into the following five cancercell phenotypes: Urothelial-like, Genomically Unstable, Basal/Squamous-like, Mesenchymal-like and Small-cell/Neuroendocrine-like. 6aining of p53 was classified by the pattern of expression as wildtype, overexpression, complete absence or cytoplasmic as previously described. 20

| Statistical analyses
Chi-2 tests were used to compare proportions of categorical variables.
If the value of any cell was below five, Fischer's exact test was used.
To test associations between subtypes and recurrence or metastatic sites (RM-sites) it was necessary to consider that each patient was assigned one subtype but often had multiple RM-sites that may not be independent.Therefore, we applied a test based on 10 000 permutations of the data set, with the maintained subtype proportions randomly assigned to patients each time.The distribution of associations produced by the permuted data served as a null background against which the observed data was tested.For each combination of a subtype and a RM-site, the number of permutations that were more extreme than the observed data were divided by 10 000 to calculate an empirical P value.Both raw and Bonferroni-adjusted P values are reported.Differential gene expression analysis was performed using the Significance Analysis of Microarrays (SAM) with q < 0.01 considered as significant.In the heatmaps, significant genes were ordered according to the observed SAM score.Gene ontology term enrichment analysis for biological processes was performed using Panther (Pantherdb.org).3 | RESULTS

| Study cohort
We studied 146 UC patients who had a recurrence after curative treatment (n = 77) or primary metastatic disease (n = 69).Cohort characteristics are shown in Table 1.Recurrence or metastatic sites (RM-sites) for each patient was categorized as "Lymph node", "Local", "Lung", "Liver", "Bone" and "Other."Patients had an average of two RM-sites ranging from one to five (Figure 1A).The most frequent RMsite was "Lymph node", present in 80 patients (55%), whereas the least frequent was "Other", occurring in 21 patients (14%).

| Association between LundTax RNA-subtypes and metastatic sites
Each subtype's association with specific RM-sites is illustrated in Figure 1B (full crosstable in Figure S2).The associations suggested some subtypes were enriched or depleted for specific RM-sites.We tested the enrichment or depletion using a permutation test, comparing each of the 42 combinations (7 LundTax RNA-subtypes vs 6 RM-sites) to the same data with the retained subtype proportions randomly assigned to patients.
The directionality and strength of the subtypes' association to each RMsite is shown in Figure 1C.The most extreme associations observed were the Ba/Sq subtype being depleted of RM-site "Bone", and the GU subtype being depleted of RM-site "Lung" but enriched for RM-site "Other" (Figure 1D), all of which were significant.In addition, this analysis also indicates enrichment of RM-site "Bone" for the three biologically related subtypes UroA/UroB/UroC, which if grouped also resulted in a significant enrichment (Bonferroni-adjusted P < .05).For one of three Ba/Sq cases with a RM-site "Bone" we could obtain matched bone metastatic tissue, which revealed a concordant Ba/Sq IHC-classification (Figure 1E) with no signs of phenotypic heterogeneity in four analyzed 1 mm tissue cores.
Histograms comparing observed associations to permuted data for each subtype and RM-site are shown in Figure S3.

| Association between LundTax IHC-subtypes and metastatic sites
The same approach was applied to test associations using the IHCbased LundTax and the MIBC Consensus classifications.Cross comparisons of the three subtyping classifiers are shown in Figure S4.
While the associations between RNA-, and IHC-based LundTax classifiers were as expected (5 Â 5 concordance = 72%), there were two notable observations in the association between IHC-based subtypes and RM-sites.First, the enrichment of the Uro subtype to RM-site "Bone" was stronger with IHC-based subtyping, due to increased power when the Uro subtype is treated as a single large class of tumors instead of UroA, UroB and UroC.Second, the negative association between Ba/Sq and "Bone" was less strong for IHC-subtypes.

| Comparison of LundTax and consensus classifiers
Associations with the Consensus classifier and RM-sites were weaker than with either of the LundTax classifiers.The Consensus classification did identify enrichment of LumU for RM-site "Other", and of LumNS for RM-site "Bone", but no significance remained after Bonferroni correction (Figures 2A and S5).(Figure 2B).Thus, none of five LundTax Ba/Sq cases within Stromarich had RM-site "Bone", in line with the previous findings described above.Only one of 12 (8%) GU cases within Stroma-rich had RM-site "Lung", while this RM-site was otherwise common (42%).Of the 10 LundTax Uro cases within the Stroma-rich category, six (60%) had RM-site "Bone", again similar to Uro tumors and higher than Stromarich non-Uro tumors (19%).In conclusion, the Stroma-rich subtype is enriched for non-TURB tissue samples and for local recurrences, but unrelated to cancer-cell phenotypes and distant metastatic sites.Since 21% of the tumors in the cohort were classified as Stroma-rich, there were fewer cases remaining among the other Consensus subtypes, which decreased the power to identify associations with subtypedependent organotropism.

| Differential gene expression analysis of primary tumors by metastatic site
Next, we investigated if gene expression profiles from primary tumors contained any strong associations to RM-sites beyond what was captured by molecular subtypes.Heatmaps in Figure 3 show the expression of significant differentially expressed (Table S2) for each RM-site, with the cases ordered first by RM-status, then by molecular subtype.For RM "Local" (Figure 3A), 18 upregulated genes showed a coherent expression across samples and included several genes expressed in the Ba/Sq and Mes-like subtypes.One third of the upregulated genes were present in extracellular matrix and immune signatures, which are known to be higher in Ba/Sq and Mes-like subtypes.Conversely, genes downregulated in RM "Local" included several epithelially expressed genes and the urothelial differentiation factor GATA3.For RM "Lymph node", six upregulated genes were identified, but they were not coherently expressed, and no common biological function could established (Figure 3B).For RM "Lung", four upregulated and four downregulated genes were identified (Figure 3C).In contrast to the other RM-sites these genes did not show any shared subtypedependent expression, suggesting their involvement as candidate regulators of metastatic potential to the lungs in a subtype-independent manner.For RM-site "Liver", "Bone" and "Other", the heatmaps indicated that genes significantly downregulated in "Liver", upregulated in "Bone", and upregulated in "Other" were highly expressed in luminal-like subtypes (Uro and GU) (Figure 3D-F).Several of these genes, identified as upregulated in RM-site "Bone" (BCAS1, DHRS2, HPGD, PSCA, SNCG), overlapped with a canonically established urothelial differentiation signature that distinguishes luminal from non-luminal subtypes.Very few of the RM "Bone" upregulated genes seemed to deviate from this subtype-dependent pattern, although two possible exceptions being SLC23A2 and NR2F1.The genes downregulated in RM "Bone" all seemed to be via low expression in the Uro subtypes.Taken together, most differentially expressed genes and signatures identified for RMsites were proxies for the specific molecular subtypes that were enriched or depleted for that site.

| Subtype specific biological characterization of organotropism
In total, 34 patients with a urothelial-like subtype (UroA, UroB or UroC) had RM-site "Bone" compared to 22 expected.Since this is an enrichment involving the largest subtype, there were sufficient numbers to explore differentially expressed genes (DEGs) within the subtype.Uro tumors from patients with RM "Bone" showed significant downregulation of five genes (SCARNA9L, SOD1, PDIA6, RSF1 and KDM1A), of which SOD1 and PDIA6 are involved in the cellular response to oxidative stress and protein folding (Figure S6A-B).SOD1 has also been previously linked to the CXCL12-CXCR4 known signaling axis in bone tropism. 21We also performed gene ontology (GO) term analysis for the top differentially expressed genes in each direction.This revealed that genes upregulated in Uro tumors with RM-site "Bone" were significantly enriched for seven GO terms all involving cartilage, bone and vasculature development (Figure S6C).Several upregulated genes (eg, COL1A1, NPPC, ROR2, TMEM119, COMP, ITGA11, HAND2 and SOX6) were driving this enrichment of skeletal system development terms, each with a modest fold change (1.17-1.45).Furthermore, except for NPPC and SOX6, these genes are  S6D).To examine if these cases were classified with high confidence, we examined the RNA-classifier delta scores (difference between highest and second highest subtype prediction scores).The three Ba/Sq cases with RM-site "Bone" had similar delta scores to the other Ba/Sq cases in the cohort indicating a confident classification.
Only seven patients with a GU subtype had RM-site "Lung", compared to 14 expected.A molecular pathological summary for these cases is shown in Figure S6D.Interestingly, five of seven GU cases with lung metastases were classified as Uro by IHC, a significantly higher proportion of discrepant cases than GU without lung metastases (P = .03).To determine if this is due to subtype intra-tumor heterogeneity or classification instability, we again examined RNAclassification stability for the five GU cases classified as Uro by IHC.
Three cases were found to have low delta scores (0.029-0.096), two being close to a Uro and one being close to a Sc/NE classification.
The other two had high delta scores (0.51-0.53), indicating that intratumor heterogeneity may underlie the discrepant RNA and IHC classification for these two cases.

| Clinical significance of atypical metastatic sites
Despite the strong enrichment for RM-site "Other" seen for the GU subtype, no specific biological determinant could be found for this category.We observed three main locations of RM-site "Other"-the brain/CNS, skin/soft tissue and abdominal carcinomatosis (Table 2).
T A B L E 2 Clinical data and molecular classification for 21 patients with RM-site "Other"

| DISCUSSION
We identify four significant patterns of metastatic organotropism occurring in the three main UC molecular subtypes (Figure 4).Among 31 patients with a Ba/Sq subtype tumor, only three (10%) had a bone metastasis, and conversely, patients with a Uro subtype tumor were overrepresented for bone metastases (34/61, [56%]).Additionally, we found two associations for patients with a GU subtype tumor, that is, an underrepresentation of lung metastasis (7/39 [18%]), and an overrepresentation of metastases to an atypical site ("Other" 12/39 [31%]), including the central nervous system.
Prior data on this topic in the literature is very limited, but one recent study performed subtype classification on biopsies from UC metastases and included information about metastatic sites. 16However, no association between subtypes and metastatic sites was presented, and any such analyses would be limited by the scarcity of biopsies from bone metastases (six cases) and the absence of lung as a specific metastatic site.Still, large proportions of the Consensus LumP and LumU subtypes in that study were reported in biopsies from liver metastases (47% and 45%), whereas the proportion of Ba/ Sq and Stroma-rich cases occurred in the liver were lower (19% and 14%).Instead, Ba/Sq and Stroma-rich subtypes had higher proportions in Lymph node metastases and the "Other" category.In our data, we saw no tendency for LumP and LumU to be enriched in patients who develop liver metastases, although the studies used different source material for the subtyping (primary bladder tumor tissue and metastatic biopsies, respectively).Thus, the results would only be comparable if molecular subtypes are concordant between primary tumors and metastases, which is still unknown.A second study based on a trial of neoadjuvant chemotherapy plus bevacizumab, identified that nine patients with bone metastases within 2 years were all of the p53-like subtype. 15Like the Stroma-rich Consensus category, the p53-like subtype is characterized by a high content of stromal cells and has no direct counterpart in the Lund Taxonomy, which is based on the intrinsic cancer-cell phenotype. 6,19Although the classifiers used were different, this result would not likely be obtained in our were upregulated in cases with bone metastases without a clear link to the Uro subtype.These genes are candidates be involved in the organ-specific steps of the metastatic cascade.However, most of variation UC metastatic sites was not explained by variation in the primary tumor's bulk expression profile.A possible interpretation is that there exists a sub-population of metastasis-initiating cells that shares the molecular subtype with the bulk profile, but also expresses private specific organotropic determinants which may not be detected by bulk gene expression profiling.In other cancer types, cells with increased metastatic potential have been defined by stemness, for example, CD44 + CD24 À breast cancer cells, 22 metabolic plasticity, for example, CD36 expression in oral cancer, 23 adhesiveness, 24 or expression of organotropic factors such as PTHrP for bone metastases in breast cancer. 25Nevertheless, despite limited knowledge regarding the metastasis initiating cell population, our results indicate that the molecular context provided by subtyping still is important for metastatic organotropism.In addition, host characteristics, immune-or stromal microenvironment, may also partially explain organ-specific metastases.Such information can theoretically be retrieved also from bulk gene expression data by applying cell-type deconvolution, 26 but would require corresponding tissue from matched metastases which was not available in the current study.
We also observed different proportions of atypical metastatic sites within the RM-site "Other" category between health care regions.Patients in the southern health care region had a lower proportion with metastases in this category, and all were emerging in the brain or testis.In contrast, patients treated in the Stockholm region in other organs except the brain, leading to a relative increase in brain metastases in MVAC-treated patients. 27In early studies on MVAC, 16% of initial responders later developed brain metastases. 28Similarly, the blood-testis barrier might cause a similar sanctuary site in the testicles. 29Gemcitabine, on the other hand, has a small molecular size and does pass into the brain. 30udy limitations include that we correlated the molecular profile of urothelial cancer with known metastatic sites at the time of recurrence or metastasis.This is a compromise between considering only the first metastatic site, and considering all sites identified by autopsy.
Furthermore, the fact that African American patients are more likely to develop lung, liver and bone metastases compared to Caucasian patients 31 should be considered as another study limitation with the vast majority of included patients being Caucasians.Additionally, since our cohort consisted of both recurrences after curative treatment and primary metastatic cancers, we did not model organ-specific metastasis in a time-dependent manner.Patient selection was from a chemotherapy treated cohort, thus excluding patients not receiving such treatment.Furthermore, without detailed genomic data from corresponding metastatic sites, we cannot be certain if the various metastases derive from the bulk or main clone of the primary tumor.
Nor can we establish if tumors change their subtypes as they metastasize to specific organs.These are important questions which likely require unique model systems, and/or large series with samples represented from both primary tumors and metastases, such as in the TRA-CERx Renal study. 32 conclusion, this is the first systematic investigation showing sitespecific metastases by UC molecular subtypes.The Ba/Sq subtype is less likely to develop bone metastases compared to the Uro subtype, whereas the GU subtype is less prone to develop lung metastases but overrepresented in atypical sites including the central nervous system.
Future research should identify the mechanisms that link these specific phenotypic states of cancer cells to organotropism in UC.
Of the additional 89 patients, 20 were previously treated with curative intent (single cycle of neoadjuvant or induction chemotherapy and RC, n = 5; radio-chemotherapy, n = 3; RC and adjuvant chemotherapy, n = 12) and 69 received upfront palliative chemotherapy for primary metastatic or locally advanced disease.Patients with recurrence or metastasis only to lymph nodes were not included.A flow chart with details about excluded patients is shown in Figure S1.The last available tissue sample before initiation of systemic therapy was used for RNA and protein analyses.The number of tumors with variant histology upon pathologist's examination is listed in Table

F I G U R E 1
Metastatic sites are unevenly distributed among UC molecular subtypes.(A) Histogram showing the distribution of the number of RMsites in the cohort.(B) Association between tumor molecular subtypes and RM-sites illustrated in a Sankey diagram.Note that each tumor has one subtype but may have more than one RM-site.(C) Dotchart showing permutation test results with the LundTax RNA-classifier.Dots placed on the left axis indicate a depletion, whereas dots placed on the right axis indicate an enrichment for a specific site.Vertical dotted lines show unadjusted, and Bonferroni adjusted P value thresholds.(D)The observed number of patients (red lines) with a specific subtype and RM-site is compared to the distribution in permuted data (histograms).Patients with the Ba/Sq subtype had fewer instances of RM-site "Bone" than expected.Patients with the GU subtype had fewer instances of RM-site "Lung" than expected.Patients with the GU subtype had more instances of RM-site "Other" than expected.(E) IHC of a primary Ba/Sq bladder tumor (TURB sample) and a matched metastasis from the depleted RM-site "Bone" show concordant KRT14 and GATA3 expression with no signs of intra-tumor heterogeneity.Images show 300 Â 300 μm.
Furthermore, the Consensus classifier identified a strong enrichment of Stroma-rich tumors for RM-site "Local" indicating that a high stromal content is associated with local recurrence or dissemination, but not with specific distant sites.Further analyses showed that only 13 of 31 (42%) of Stromarich tumors were sampled from TURB tissue, compared to 93 of 115 (81%) among the other Consensus subtypes (P = 4.4e-05).This indicates that subtyping in non-TURB tissue (cystectomy, lymphadenectomy or distant metastatic samples) may show systematic differences in stromal content, which may not always reflect an intrinsic tumor characteristic (Figure 2B).Since the LundTax classifies tumors based on expression states of the cancer cells proper, the 31 cases classified as Consensus Stroma-rich represented a mix of LundTax subtypes (10 Uro, 12 GU, 5 Ba/Sq and 4 Mes-like).Within the Stroma-rich class, patients' RM-sites were consistent with their LundTax classification Associations with Stroma-rich tumors according to consensus classification.(A) Dotchart showing permutation test results with the MIBC Consensus RNA-classifier.Dots placed on the left axis indicate a depletion, whereas dots placed on the right axis indicate an enrichment for a specific site.Vertical dotted lines indicate unadjusted and Bonferroni adjusted P value thresholds.(B) Stroma rich tumors were enriched for non-TURB Sample-type and enriched for local recurrence, likely via selection due to sampling.LundTax cancer-cell phenotypes within Stroma-rich maintained the same associations seen in non-Stroma-rich tumors, suggesting that specific tropism is due to cancer-cell intrinsic biology.

F I G U R E 3
Differentially expressed genes for RM-sites show subtype specific expression patterns.Heatmaps showing significant differentially expressed genes (DEGs) between primary tumors from patients with or without specific RM-sites, ordered within groups by LundTax subtype.(A) RM-site "Local", (B) RM-site "Lymph node", (C) RM-site "Lung", (D) RM-site "Liver", (E) RM-site "Bone" and (F) RM-site "Other."members of a coherent bladder cancer extracellular matrix signature, suggesting an induced expression in fibroblasts, than in the cancer cells.We also performed a similar analysis for the GU subtype and RM-site "Other."In total 12 patients with a GU tumor had RM-site "Other" compared to five expected.No significant DEGs or enriched GO terms were identified.Only three patients with a Ba/Sq subtype had RM-site "Bone", compared to 11 expected.Examination of IHC-stained tissue cores and Hematoxylin and Eosin-stained TURB whole sections together with molecular-pathological data in these three tumors, verified a typically Ba/Sq-like appearance, including squamous differentiation, trabecular growth pattern and expected molecular alterations (altered CDKN2A[p16], RB1 and TP53 inferred by IHC staining) (Figures1Eand Of seven patients with metastases to brain/CNS, six were of the GU IHC subtype (P = .001).Several cases with soft tissue metastases and/or carcinomatosis were of GU subtype, but this category also included patients with a Uro or Sc/NE subtype.Finally, a pattern was observed when analyzing these data in relation to health care region, where five of six patients with RM-site "Other" from the Southern health care region had brain/CNS metastases and the last patient had a testicular metastasis.All instances of soft tissue/carcinomatosis were among patients from the Stockholm health care region.The main treatment differences between the regions during the study period are that open surgery and MVAC was used in the Southern health care region, whereas robotic surgery and GC was used in the Stockholm health care region.However, the type of surgery could not likely explain the observed differences, since patients from the Stockholm region with abdominal carcinomatosis or skin/soft tissue metastases were operated both with open (n = 5) and robotic (n = 4) techniques.However, treatment with different chemotherapy regimens before recurrence is a possible explanation, since 18 of 21 (86%) of the cases with RM-site "Other" were recurrences after chemotherapy.These observations prompted us to study RM-sites in two subgroup analyses.First, patients with recurrence after RC (n = 111) showed different proportions with local recurrence when stratified by region (FigureS7A).Second, patients with recurrence after curative chemotherapy treatment (n = 74) showed a trend toward different sites depending on type of chemotherapy regimen (FigureS7B).
cohort, since the similar Consensus subtype, Stroma-rich, showed no enrichment for bone metastases.The diverging results could possibly be explained by the exposure to anti-VEGFR therapy (bevacizumab) in addition to chemotherapy affecting the natural tropism of UC cells or be related to the relatively low number of observations with bone metastases.The finding that most DEGs for specific metastatic sites were proxys for molecular subtypes associated with those specific sites support our hypothesis of subtype-dependent metastatic organotropism.Possible exceptions from these observations were DEGs for lung metastatic tumors, where a few genes (including cell adhesion molecules ITGA6 and DSC3) were significantly upregulated without any coherent subtype association.Furthermore, SLC23A2 and NR2F1, F I G U R E 4 Summary of the proportional differences in distant metastatic sites between the three major molecular subtypes of UC.Given a tumor of a Urothelial-like (Uro), Genomically unstable (GU) or Basal/Squamous (Ba/Sq) subtype, the relative proportions of distant metastatic sites are shown (excluding lymph node and local recurrences).One star is shown for nominally significant associations (in Figure1C), and two stars are shown for the Bonferroni significant association.A plus-sign indicates significant enrichment, whereas a minus sign is shown for RMsites significantly depleted in the subtype.
had very few metastases to brain, but more frequently to soft-tissue, skin or in the form of carcinomatosis.Neither extent of lymphadenectomy (similar between health care regions) nor surgical technique (open vs robotic cystectomy) could explain the difference, since only a minority of the cases with atypical sites differed by the type of surgery.One of many possible explanations for this geographical observation is the different chemotherapy regimens used.MVAC does not cross the blood-brain barrier and a sanctuary effect has been suggested by which MVAC treatment achieves efficient systemic control