Abstract

   Current diagnostic methods for canine urothelial carcinoma (UC) are
   technically challenging or can lack specificity, hence there is a need
   for novel biomarkers of UC. To this end, we analysed the microRNA
   (miRNA) cargo of extracellular vesicles (EVs) from urine samples of
   dogs with UC to identify candidate miRNA biomarkers. Urine was
   fractionated using ultrafiltration combined with size-exclusion
   chromatography and small RNA sequencing analysis was performed on both
   the EV enriched and (EV free) protein fractions. A greater number of
   candidate miRNA biomarkers were detected in the EV fraction than the
   protein fraction, and further validation using droplet digital PCR
   (ddPCR) was performed on the EV enriched fraction of a second cohort of
   dogs with UC which indicated that miR-182, miR-221 and miR-222 were
   significantly overrepresented in dogs with UC when compared with
   healthy dogs and dogs with urinary tract infections. Pathway analysis
   confirmed that these three miRNAs are involved in cancer. In addition,
   their potential downstream gene targets were predicted and PIK3R1, a
   well-known oncogene is likely to be a shared target between miRNA-182
   and miRNA-221/222. In summary, this study highlights the potential of
   urinary EV-associated miRNAs as a source of biomarkers for the
   diagnosis of canine UC.

   Subject terms: Tumour biomarkers, Urological cancer, Medical research,
   Oncology, Urology

Introduction

   The diagnosis of urothelial carcinoma (UC), the most common type of
   bladder cancer in dogs^[38]1, is usually made on cytology from the
   tumour obtained by traumatic catheterisation, however this diagnostic
   method is invasive and not easily achieved in first-opinion veterinary
   practice. Hence, the development of non-invasive urine based diagnostic
   testing would be beneficial to allow early screening and diagnosis of
   dogs for UC. Several studies have investigated a variety of potential
   biomarkers and non-invasive diagnostic tests for canine UC, including
   protein biomarkers^[39]2,[40]3, metabolomic biomarkers^[41]4,[42]5 and
   DNA copy number aberrations^[43]6. However, the investigated biomarkers
   have been inadequately sensitive and/or specific to develop as
   diagnostic tests for UC. The most promising biomarker for UC available
   at present detects mutations in the BRAF gene (variant V595E) in cells
   obtained from urine samples of dogs^[44]7, however the test can have a
   sensitivity as low as 64%^[45]8, and hence false negative results will
   occur. Thus, there remains a need to identify highly sensitive
   non-invasive biomarkers for canine UC.

   In other fields, there is increasing interest in microRNA (miRNA)-based
   biomarkers, and changes in miRNA expression might also provide insights
   into the pathophysiological mechanisms underpinning disease processes,
   such as carcinogenesis. So far, two articles have reported altered
   miRNA profiles in canine bladder tissue in UC^[46]9,[47]10: Vinall et
   al. compared the miRNAome of transitional cell carcinoma with healthy
   dogs and dogs with inflammatory bladder disease, while Varvil et al.
   compared the miRNAome of normal canine urothelium to that of invasive
   UC. Both studies detected a few differentially expressed (DE) miRNAs in
   canine UC tissue, but the actual miRNAs differed. Subsequent urinary
   qPCR analysis of five differentially expressed miRNAs from one of the
   studies^[48]9 confirmed significantly reduced abundance of miR-103b and
   miR-16 in urine samples from dogs with UC compared to other lower
   urinary tract diseases^[49]11. Hence, miRNAs do appear to be
   differentially expressed in canine UC, although further studies are
   needed to establish if these are robust enough for clinical use.

   Extracellular vesicles (EVs) are nanoparticles released by all cells,
   which contain proteins, mRNAs and miRNAs from the cell of
   origin^[50]12. EVs carry intercellular messages, especially in cancer,
   and therefore their cargo has a high biomarker potential compared to
   the original biofluid^[51]13–[52]15. The majority of urinary EVs come
   from the kidney, prostate and bladder^[53]16, and so urinary
   EV-associated miRNAs should be specific for changes in the miRNome of
   the urinary tract. In addition, the lipid bilayer that surrounds the
   miRNA cargo of EVs will protect against degradation in the urine by
   RNAses, hence urinary EV-associated miRNAs should also be more
   sensitive for changes in the urinary miRNAome. However, large
   quantities of small RNAs co-isolate with non-vesicular urinary
   proteins^[54]17, which may or may not be representative of changes in
   the miRNAome of the urinary tract, therefore the use of techniques that
   separate EVs from non-vesicular proteins are important to optimise
   miRNA biomarker discovery. To date only one study has evaluated urinary
   EVs from dogs with UC, although this study investigated the abundance
   of Nuclear Matrix Protein-22 in urinary EVs of dogs with UC^[55]18,
   hence the potential of urinary EV-associated miRNAs as a source of
   biomarkers for canine UC is to date unexplored.

   The aim of the present study was to examine EV-associated and
   non-vesicular (protein-associated) miRNAs as potential biomarkers for
   canine UC.

Results

   The results of the study are reported in accordance with the ARRIVE
   guidelines.

Clinical cases

   The ‘sequencing’ cohort included 12 dogs with UC, 5 dogs with urinary
   tract infections (UTI) and 6 healthy control (HC) dogs. The
   ‘validation’ cohort included 8 dogs with UC, 8 dogs with UTI and 5 HC
   dogs. Signalment data for the groups are shown in Table [56]1.

Table 1.

   Baseline data for dogs in each of the cohorts included in the study.
   Group Age (median [range]) Sex Breed Total urine creatinine of sample
   volume from which EVs were extracted (median [range]) Small RNA content
   of 500uL of EV fraction (median [range]) miRNA content of 500uL of EV
   fraction (median [range])
   Sequencing UC (n = 12) 10 [8–14] years, 1 unknown

   1 male entire

   5 male neutered

   6 female neutered
   1 each of Basset de Bretagne, Staffordshire Bull Terrier, Jack Russell
   Terrier, Schnauzer, Scottish Terrier, Labrador, Polish Lowland
   Sheepdog, Pug, Border Collie, Poodle, Weimaraner, Cross-breed 17 [4–90]
   µmol 17 [2–53] ng 10 [1–18] ng
   Sequencing UTI (n = 5) 9 [6–10] years, 1 unknown

   1 male neutered

   1 male unknown status

   2 female neutered

   1 female entire
   2 Cocker Spaniel, 1 each of German Shepherd, Springer Spaniel and Boxer
   27 [2–106] µmol 13 [6–80] ng 2 [2–30] ng
   Sequencing HC (n = 6) 5 [2–9] years

   4 male entire

   1 male neutered

   1 female neutered
   1 each of Finnish Spitz, Bichon Frise, Labrador, Dachshund, German
   Shepherd and Cross-breed 59 [30–99] µmol 12 [8–33] ng 6 [3–16] ng
   Validation UC (n = 8) 12 [11–13] years

   4 male neutered

   3 female neutered

   1 unknown
   2 Labradors, 2 Staffordshire Bull Terriers, 2 West Highland White
   Terriers, 1 each of Papillon and Beagle 26 [13–30] µmol 26 [17–65] ng
   12 [8–40] ng
   Validation UTI (n = 8) 7 [0.25–10] years

   1 male entire

   1 male neutered

   1 female entire

   4 female neutered

   1 unknown
   2 Labrador, 1 each of Border Collie, Hungarian Viszla, Golden
   Retriever, Newfoundland, Staffordshire Bull Terrier and Pug 19 [9–54]
   µmol 24 [7–38] ng 9 [3–21] ng
   Validation HC (n = 5) 8 [2–12] years

   3 male neutered

   2 female neutered
   2 Cocker spaniels, 1 each of Boxer, Border Terrier and Pug 70 [26–82]
   µmol 27 [10–33] ng 11 [< 1–17] ng
   [57]Open in a new tab

EV characterisation

   Extracellular vesicles were separated from vesicle free proteins from
   canine urine as per our previously published protocol using
   ultrafiltration combined with SEC^[58]17 which was further optimised
   for canine urine (Fig. [59]1A, Supplementary Fig. 1). Transmission
   electron microscopy imaging (TEM) confirmed the presence of
   appropriately sized vesicles in fraction B (Fig. [60]1B) whereas
   non-vesicular protein fraction D had no visible vesicles (Fig. [61]1C).

Figure 1.

   [62]Figure 1
   [63]Open in a new tab

   Schematic presentation of the EV isolation steps, and SEC fractions
   collected during purification. Transmission electron microscopy images
   of (B) Fraction B including EVs and (C) Fraction D including
   vesicle-free proteins. (D) Scatter plot of urinary EV numbers
   normalised to total creatinine content of sample (µmol). Dogs with UTI
   and UC had significantly higher numbers of urinary EV compared to dogs
   in the HC group: there was no significant difference between UC and UTI
   groups. *p < 0.05.

   Nanoparticle tracking analysis confirmed the presence of particles with
   a size profile consistent with EVs (Supplementary Fig. 1C). Urine EV:
   urine total creatinine ratio (of ‘sequencing’ and ‘validation’ cohorts
   combined) were significantly greater in dogs with UC (P = 0.04) and UTI
   (P = 0.03) compared to healthy controls, however there was no
   significant difference between the UC and UTI groups (P = 0.63)
   (Fig. [64]1D).

Small RNA sequencing revealed differentially expressed miRNAs in UC

   For the sequencing cohort samples, RNA from EV and protein
   SEC-fractions was subjected to small RNA sequencing. First, the size
   profiles of the two libraries collected for each sample were compared
   for both length and clustering analysis and the profiles were very
   similar (Supplementary Fig. 2). The libraries were combined for each
   sample to maximise read number and the sensitivity for detection of
   candidate biomarkers in the later analyses. A total of 569,062,706
   reads were obtained with a mean of 10,161,834 reads per sample
   (Supplementary Table 1). Out of processed and filtered 16–24 nt long
   reads, 1–29% (median 6.3%) aligned to mature canine miRNAs in EV
   fractions and 0.06–8% (median 0.5%) in protein fractions (Supplementary
   Table 1). In total, of 455 known mature canine miRNAs in
   miRbase^[65]19, 159 were identified from the samples. Of these, 158
   were identified in the EV fractions and 150 were present in the protein
   fractions.

   Differential expression analysis was performed separately for EV and
   protein fractions from the sequencing cohort. For each analysis, the HC
   and UTI groups were combined as a control group, and miRNA abundance
   was compared with that of the UC group. The EV fraction included 11
   miRNAs that were more abundant and 4 miRNAs that were less abundant in
   the UC group (adjusted p-value < 0.05) (Fig. [66]2A, Supplementary
   Table 2), hereafter defined as the differentially abundant miRNAs
   (DA-miRNAs). Protein fractions included 5 significantly upregulated
   mature miRNAs and 2 downregulated miRNAs in the UC group (Fig. [67]2B,
   Supplementary Table 3). All the miRNAs with differential abundance in
   the protein group had base mean 60 or lower, indicating relatively low
   expression level, whereas 7 out of 15 DA-miRNAs in the EV fraction had
   base mean higher than 60. In addition, as the EV fraction yielded a
   higher percentage of reads aligned to miRNAs and a greater number of
   DA-miRNAs than the protein fraction, we chose to continue to further
   analyse the EV fractions only. Principal component analysis (PCA) using
   the DA-miRNAs from the EV fraction separated the control groups from UC
   samples (Fig. [68]2C). Normalised read counts for the DA-miRNAs in the
   EV fraction are presented in Fig. [69]2D.

Figure 2.

   [70]Figure 2
   [71]Open in a new tab

   Small RNA sequencing results. Volcano plot of differentially expressed
   (DE) miRNAs in (A) EV fraction and (B) protein fraction when UC samples
   were compared to controls (healthy controls and dogs with UTI).
   DE-miRNAs with a log[2] fold change value < 0 were less abundant in UC
   dogs relative to healthy controls and those with a log[2] fold change
   value > 0 were more abundant in UC dogs relative to healthy controls.
   (C) Principal component analysis using all DE miRNAs from EV fraction
   separated UC samples from HC and UTI groups. (D) Normalised read
   numbers of each DE miRNA in EV fractions.

ddPCR validation of candidate miRNAs

   In addition to potential miRNA candidate biomarkers, the sequencing
   data from the EV fractions were explored to identify miRNAs that had
   similar expression across the UC and healthy control groups, and from
   this group we selected individual miRNAs with high, medium or low
   expression level against which to normalise our candidates
   appropriately. As a result, three miRNAs with different expression
   levels were selected: cfa-let-7a (mean expression: 8702 reads, adjusted
   p-value: 0.55), cfa-miR-26a (mean expression 609 reads, adjusted
   p-value: 0.28) and cfa-miR194 (mean expression 40 reads, adjusted
   p-value: 0.51). Their abundance was further explored with ddPCR which
   confirmed stability of expression of these miRNAs across the cohorts
   (Fig. [72]3A–C). We performed ddPCR for these normalisation miRNAs, in
   both the sequencing and validation cohorts.

Figure 3.

   [73]Figure 3
   [74]Open in a new tab

   ddPCR validation using both the discovery cohort and validation cohort.
   (A–C) Three baseline miRNAs used in normalisation of the potential
   biomarker miRNAs. Droplet count was normalised to the number of
   particles in EV fraction/µmol creatinine in the urine. (D–J) Droplet
   counts of potential biomarker miRNAs normalised to the closest baseline
   miRNA and total µmol of creatinine. Receiver operating characteristic
   curves of the three validated miRNAs with (K) UC vs. healthy controls
   and UTI combined and (L) UC vs. UTI samples. *p < 0.05, **p < 0.01,
   ***p < 0.001.

   Validation with ddPCR was performed for nine candidate biomarker miRNAs
   in the EV fraction: miR-92b, miR-93, miR-143, miR-145, miR-150,
   miR-182, miR-199, miR-221 and miR-222. Of these, miR-199 and miR-143
   had low droplet count in ddPCR, meaning no or few individual droplets
   per sample were detected, therefore, we excluded those miRNAs from
   further analysis. However, for the other candidate miRNAs the ddPCR
   data were normalised to the normalisation miRNA with the closest
   expression level to that of the candidate miRNA and a second
   normalisation to the total creatinine amount in the sample was then
   performed (Fig. [75]3D–J). In the sequencing cohort, the ddPCR results
   confirmed significant upregulation of the miRNAs of interest
   (p < 0.05).

   In the independent validation cohort, three miRNAs (miR-182, miR-221
   and miR-222) demonstrated increased expression in the urinary EVs of UC
   dogs compared to the control groups (UTI and HC combined,
   F[76]ig. [77]3G–I).

   The biomarker potential of these three candidate miRNAs was evaluated
   using receiver operating characteristic (ROC) curve analysis. The
   analysis included two different comparisons: UC dogs versus control (HC
   and UTI groups combined) (Fig. [78]3K) and UC dogs versus UTI dogs
   (Fig. [79]3L). Normalised miR-222 had the highest area under the curve
   (AUC) for discriminating UC dogs from all control dogs (AUC = 0.88, 95%
   CI 0.78–0.98; P < 0.001) and UC dogs from UTI dogs alone (AUC = 0.86,
   95% CI 0.72–1.00; P = 0.001) both indicating excellent discrimination.
   The normalised and indexed expression of miR-182 and miR-222 had
   acceptable (moderate to good) discrimination (Table [80]2).

Table 2.

   Results of receiver operating curve analysis evaluating the diagnostic
   accuracy of the 3 candidate urinary EV-associated miRNAs as a
   diagnostic biomarker for urothelial carcinoma (UC).
   miR candidate AUC for UC vs. all controls 95% CI for UC vs. all
   controls Sig AUC for UC vs. UTI 95% CI for UC vs. UTI Sig
   miR-222 0.88 0.78–0.98 p < 0.001 0.86 0.72–1.00 p = 0.001
   miR-221 0.79 0.66–0.92 p = 0.001 0.68 0.49–0.87 p = 0.09
   miR-182 0.85 0.73–0.96 p < 0.001 0.78 0.62–0.83 p = 0.008
   [81]Open in a new tab

   Two control groups were used; all controls represented a population of
   healthy controls and dogs with diagnosed urinary tract infection (UTI),
   and UTI represented the group of dogs with UTIs only.

Predicted gene targets of miR-182, miR-221 and miR-222

   We used both miRDB^[82]20 and TargetScan v8.0^[83]21 for biological
   target prediction for miR-182 and miR221/222 and the overlapped genes
   were extracted for further analysis, of which 377 genes were found for
   miR-182 and 121 were found for miR-221/222. Hypergeometric tests were
   performed to determine to what extent the two databases agreed with
   each other and the resulting p-values were both less than 0.001,
   indicating that the number of overlapping genes is more significant
   than expected. Pathway enrichment analysis using Enrichr^[84]22
   revealed that the target genes of both miR-182 and miR221/222 are
   highly associated with a wide range of cancer types (Supplementary
   Fig. 3B and C), including endometrial cancer, ovarian cancer,
   pancreatic ductal cancer, breast cancer, prostate cancer, glioblastoma
   and hepatocellular carcinoma with the underlying molecular processes
   involving the FoxO and the ErbB pathways (Fig. [85]4A,B).

Figure 4.

   [86]Figure 4
   [87]Open in a new tab

   Gene set enrichment analysis of the target genes of miR-182 (A) and
   miR-221/222 (B), respectively, using the Enrichr web tool. Enriched
   results associated with the signaling pathways from the Elsevier
   Pathway Collection, the KEGG 2021 Human and the BioCarta 2016 databases
   were shown. P values were adjusted using the Benjamini-Hochberg (BH)
   method. (C) A volcano plot that indicates the differentially expressed
   (DE) genes between the 22 canine urothelial carcinoma tissue (iUC)
   samples and the 8 normal control samples (Nom) using the published
   datasets DRA005844 and SRP217874 obtained from GEO. The overlaps
   between the DE genes and the miR-182 target genes (GL82) were coloured
   in blue whereas the ones between DE genes and the miR-221/222 target
   genes (GL221) were coloured in orange. DE genes were determined using
   the edgeR Bioconductor package in R. (D) Violin plots showing the
   expression of shared overlapped DE genes between GL182 and GL221/222,
   which are differentially expressed between the Nom and the iUC samples.

   Since target prediction alone cannot reveal the exact downstream effect
   of miRNA changes, we collected differentially expressed gene lists from
   three recently published studies of dog UC^[88]23–[89]25 and used them
   to prepare 5 gene sets to further validate our results. The sizes of
   these gene sets were 30, 342, 361, 888 and 1182 after unrecognised gene
   symbols were filtered out. The gene targets of miRNA-182 were
   significantly enriched with down-regulated genes compared between UC
   and normal (adj.p < 0.05), suggesting that miR-182 is likely to play an
   inhibitory role in UC. As a comparison, the gene targets of miR-221/222
   consist of both up- and down-regulated genes in UC, reflecting a more
   mixed (activation/inhibitory) role of miRNA-221/222 (Supplementary
   Fig. 3D).

   We also downloaded two available datasets from NCBI SRA
   ([90]https://www.ncbi.nlm.nih.gov/sra), including 22 bladder tissue
   samples from dogs with UC and 8 normal controls (Supplementary
   Fig. 3E). 3451 upregulated and 3269 downregulated genes in UC compared
   to normal controls were defined with cutoffs of |logFC|< 1 and False
   Discovery Rate (FDR) < 0.05. Further overlapping with target genes of
   miRNA-182 and miRNA-221/222 resulted in 129 differentially expressed
   (DE) genes for gene list 182 (GL182; 39 up-regulated and 90
   down-regulated) and 34 DE genes for gene list 221/222 (GL221/222; 14
   up-regulated and 20 down-regulated), respectively (Fig. [91]4C).
   However, there are only 2 up-regulated (ARF4 and FOXN2) and 4
   down-regulated (RIMS3, PCMTD1, ADAM22 and PIK3R1) genes in UC shared
   between miRNA-182 and miRNA-221/222 (Fig. [92]4D). Consistent with the
   pathway enrichment analysis, PIK3R1, a component in the PI3K signalling
   pathway that is associated with the regulation of cell growth,
   proliferation and survival^[93]26, appeared as an important miRNA
   target gene involved in canine UC development.

Discussion

   The primary aim of the present study was to identify novel urinary
   EV-based miRNA biomarkers for canine UC, and we identified three
   candidate miRNA biomarkers, namely miR-182, miR-221 and miR-222. ROC
   analysis showed that increased normalised abundance of miR-222 was
   significantly associated with UC and demonstrated excellent
   discrimination between dogs with UC vs. dogs with UTI and vs. a
   combined group of healthy dogs and dogs with UTI. Normalised expression
   of miR-221 and miR-182 associated with EVs was also associated with
   good discrimination of UC dogs from the other groups.

   Detection of the BRAF mutation has shown promise in diagnosis of UC and
   prostate cancer in dogs^[94]7 although sensitivity can be as low as
   64%^[95]8, and it is possible that miRNA biomarkers, especially from
   EVs, could represent complementary detection methods for UC. Future
   studies comparing the sensitivity and specificity of our individual or
   combined candidate EV-associated miRNA biomarkers and the detection of
   BRAF mutations for the diagnosis of canine UC would be of interest.

   miR-182, miR-221 and miR-222 are expressed in several tissue types and
   are not specific to bladder tissue^[96]27, however previous studies
   have found a strong correlation between miR-221/222 and cancer^[97]28
   and combination treatment including antagomiRs targeting miR-221-3p and
   miR-222-3p (along with palladium allyl complex 4d) induced apoptosis of
   colon cancer and glioblastoma cancer cells^[98]29, further supporting
   their potential role in the survival of cancer cells. Another recent
   study showed that miR-221/222 targeted ANXA protein, regulated cell
   cycle progression and affected chemosensitivity in breast
   cancer^[99]30, which also indicates their importance in tumour
   pathophysiology. miR-182 is also upregulated in prostate cancer tissues
   and cells, and lncRNA LINC01018 (which targets miR-182) was shown to
   regulate its expression^[100]31. These studies support the biological
   relevance of increased expression of miR-182, miR-221 and miR-222
   within urinary EVs and we hypothesise that the altered abundance of
   these miRNAs within canine urinary EVs of dogs with UC reflects
   increased expression of these miRNAs within canine UC cells themselves.

   Our target prediction and pathway enrichment analysis revealed that the
   targets of our candidate miRNA biomarkers are highly associated with
   cancer tissues, and that the gene targets of our candidate miRNAs are
   significantly enriched within canine UC gene datasets. Pathway
   enrichment analysis implicated the FoxO and the ErbB pathways, which
   regulate cell proliferation, migration, differentiation, apoptosis and
   cell motility by mediating the PI3K-Akt pathway, the MAPK pathway and
   the mTOR pathway^[101]32–[102]34, in canine UC. These data support
   reports that ErbB/MAPK signalling and the AKT-mTOR-S6K cascade could be
   therapeutic targets in canine UC^[103]23,[104]35.

   From the key genes overlapping with these cancer-associated pathways,
   two are shared between miR-182 and miR221/222: PIK3R1 and PPP3R1. Both
   are well-known oncogenes^[105]36,[106]37, and according to our
   prediction results, they could also play an important role in canine UC
   biology. The findings of our target prediction and pathway enrichment
   analyses further supports the biological relevance of increased
   miR-182, miR-221 and miR-222 abundance in urine EVs of canine UC
   patients and suggests that increased expression of these biomarkers is
   also apparent within canine UC tissues. Interestingly, the DA-miRNA
   biomarkers we identified have not been reported in previous studies of
   the miRNAome of canine UC tissues^[107]9,[108]10, although similar
   findings discrepancies have been reported in humans^[109]38. Our
   findings are strengthened by the inclusion of a second validation
   cohort of dogs, which led to the exclusion of several candidate
   biomarker miRNAs. We hypothesise that the altered abundance of
   EV-associated miRNAs observed are reflective of changes in the UC cells
   themselves rather than elsewhere in the urinary tract and further
   investigation of the expression levels of these candidate miRNAs in
   canine UC tissues would be of interest.

   Unfortunately, two of the miRNAs selected for the ddPCR validation,
   miR-143 and miR-199, did not produce reliable data due to sensitivity
   issues with the ddPCR assays available, therefore we were not able to
   validate their expression by ddPCR or evaluate their potential as a
   biomarker. We suggest that they are not discounted but require further
   technical developments, given that both have previously been linked to
   bladder cancer in humans^[110]39–[111]42.

   One previous study identified reduced abundance of miR-16 and miR-103b
   in the urine of dogs with UC^[112]11, however neither of these miRs
   were identified having reduced abundance in the EV or protein fractions
   of our UC cohort, which may reflect differences in the sample type
   evaluated between the two studies (individual EV/protein fractions vs.
   total urine). Furthermore, in our discovery cohort several miRs were
   identified as possible biomarkers for UC but most were not
   differentially abundant in the second independent validation cohort,
   and so it is possible that miR-16 and miR-103b might not be
   differentially expressed in other cohorts of dogs with UC.

   Currently there is no gold standard method to normalise biomarker or
   miRNA data from urine or urine EVs^[113]16, although there is ongoing
   effort to map out the effects of background variables and normalisation
   for urine EV studies^[114]43. Here, we chose to normalise the miRNA
   abundance determined by ddPCR to three normalisation miRNAs to account
   for differences in RNA isolation efficiency between samples and we
   further normalised for urine total creatinine content to adjust for
   differences in urine concentration and volume between samples. Since
   the volume and concentration of urine was not standardised among
   patients, we believe our normalisation strategy should allow for
   meaningful comparison between the groups whilst also accounting for
   differences in RNA isolation efficiency.

   Since the number of dogs included was small it should be considered a
   pilot study, and further investigations of the clinical sensitivity and
   specificity of these miRNA biomarkers in larger cohorts of dogs with UC
   and UTIs are warranted. In addition, all of the UC dogs we studied
   presented with clinical signs and had gross evidence of disease on
   ultrasound, so it would be interesting to investigate if these miRNAs
   are also useful biomarkers for detecting early stage/microscopic UC,
   and hence might be useful as a screening test for UC in predisposed
   breeds (for example West Highland White Terriers).

   In conclusion, we identified novel urinary EV-based miRNA biomarkers
   for canine UC which demonstrated excellent discrimination between dogs
   with UC vs. either dogs with UTI or a combined group of healthy dogs
   and dogs with UTI. Further investigation of the clinical sensitivity
   and specificity of these miRNA biomarkers and the role of these DE
   miRNAs in the biology of canine UC is warranted.

Methods

   All methods were performed in accordance with the relevant guidelines
   and regulations. The Ethics and Welfare Committee of the Department of
   Veterinary Medicine approved the study protocol (CR264).

Patient recruitment

   Dogs with confirmed UC (UC group; based on compatible clinical history,
   abdominal ultrasound and bladder suction cytology findings) were
   recruited to the study. In addition, two control groups were included;
   one was dogs with confirmed urinary tract infections (UTI group) and no
   clinical suspicion of UC and the second were healthy dogs (owned by
   staff at our institution) without any clinical history of urinary tract
   disease and which were not on any medications other than routine
   parasiticides (HC group). Samples were collected by catheterisation or
   free catch for the UC group, by cystocentesis for the UTI group and by
   free catch for the HC group.

   Dogs were recruited into two cohorts; the first ‘discovery’ cohort was
   recruited between July 2017 and November 2019, whilst the second
   ‘validation’ cohort was recruited between February 2020 and May 2021.

   Urine samples were collected and processed within 4 h. Urine was
   centrifuged at 2000×g for 10 min (to pellet cellular components) and
   the cell-free supernatant harvested prior to freezing at − 80 °C until
   batch analysis.

EV isolation and analysis

   Frozen urine aliquots were thawed, and a protease inhibitor (Pierce
   Protease Inhibitor, Thermo Fisher) was added. Extracellular vesicles
   were isolated from the urine samples by combined ultrafiltration and
   size-exclusion chromatography (UF-SEC) method, as per our previously
   published protocol^[115]17. In brief, urine was centrifuged at 4000×g,
   for 20 min at 4 °C, and the supernatant was filtered using 0.22 µm
   syringe PES filters (Millex GP, #SLGP033RS). The filtered urine
   (1–3.6 mL) was concentrated to < 130 µL using Amicon Ultra centrifugal
   ultrafiltration columns (100 kDa cut off, #UFC810024, Millipore,
   Burlington, MA, USA). The final volume of concentrated urine was
   adjusted to 130 µL with PBS and inserted into the size-exclusion
   chromatography column (qEV single 70 nm, Izon Science Ltd, Burnside,
   New Zealand) that had been prewashed with 5 mLs of PBS. The SEC
   fraction volumes were further optimised for dog urine samples which
   resulted in collection of five fractions: void fraction (Fraction A,
   1 mL), EV enriched fraction (Fraction B, 600 µL), second void fraction
   (Fraction C, 1.1 mL), protein fraction (Fraction D, 1.4 mL) and third
   void fraction (Fraction E, 500 µL) (Supplementary Fig. 1,
   Fig. [116]1A). RNA was isolated from 500 µL of Fraction B and Fraction
   D; each fraction was combined with 2.5 mL Qiazol lysis reagent (#79306,
   Qiagen, Venlo, The Netherlands), incubated at room temperature for
   5 min and stored at − 70 °C. Remaining SEC fractions were stored
   at − 70 °C for further EV characterisation.

   EV characterisation was performed on Fractions B and D from individual
   HC dogs. Firstly, transmission electron microscopy (TEM) was performed
   as previously described^[117]17 and nanoparticle tracking analysis
   (NTA) using NanoSight NS300 (Malvern, Worcestershire, UK) as previously
   described^[118]17 was also performed on Fraction B of all samples to
   establish EV concentrations. Particle concentrations (assumed to be
   EVs) were normalized to urine creatinine concentrations (determined
   using the pyrogallol red method) as an estimate of 24-h urine EV
   excretion rate, since measurement of urine protein:urine creatinine
   concentrations on spot urine samples are highly correlated with 24-h
   protein excretion in dogs^[119]44.

RNA isolation and sequencing library preparation

   RNA was isolated from 500 µL of Fractions B and D of the sequencing
   cohort, and fraction B of the validation cohort samples. Qiazol-sample
   mixes were defrosted on ice, combined with 0.2 volume of chloroform,
   and left for 5 min at room temperature. Solutions were centrifuged
   (12,000×g, 15 min, 5 °C), the aqueous phase was collected and combined
   with 2.5 volumes of 96% ethanol. The sample mix was added to miRNeasy
   Mini Kit columns (Qiagen, Hilden, Germany, #217004) and washed
   according to the manufacturer’s instructions. Finally, the total RNA
   was eluted in 30 μL of nuclease free water. The small RNA profile of
   isolated RNAs was measured from 1 μL of sample using the Agilent
   Bioanalyzer Small RNA kit (#5067-1548, Agilent Technologies, Santa
   Clara, CA, USA). The concentration of 10–40 nucleotide (nt) and 40–80
   nt small RNA were calculated using the 2100 Expert software with custom
   ranges.

   Sequencing libraries were prepared from Fractions B and D of the
   sequencing cohort using QiaSeq miRNA library kit (#331505, Qiagen)
   combined with QIAseq miRNA 48 Index IL kit (#331595, Qiagen) according
   to the suggested manufacturer protocol. 99–9600 pg of miRNA sized RNA
   was used as starting material. Eleven samples were processed and
   analysed as replicate samples. As an additional step, instead of
   discarding the magnetic beads after the indexing step as instructed,
   the beads were collected and washed. As a result, two sequencing
   libraries from each sample with the same index were obtained: “miRNA
   library” with the original protocol and “longRNA library”, which
   included the DNA from the discarded beads. The library preparation
   failed for one healthy control fraction d sample, and it was excluded
   from the sequencing run. The samples from each library were combined
   separately and run in a NovaSeq SP100 flowcell in separate lan.

Bioinformatics analyses

   Bioinformatic analyses were performed with R version 4.2.1. The raw
   reads including the middle linker sequence were filtered from fastq
   files using the command line, and the number of unique UMIs for
   sequences longer than 15 nt in each sample were calculated using a
   self-written R-script. The results from the two obtained libraries were
   compared with PCA and correlation analyses (Supplementary Fig. 2) and
   as a result, the count numbers of both libraries (“mirna” and “long”)
   were combined for individual samples.

   For the analysis of mature miRNAs, the data was further filtered to
   contain only 16–24 nt read sequences matching mature miRNAs. The
   alignment of the reads and DE analysis were performed using scripts
   from DEUS 1.0 pipeline^[120]45. At first, remaining 16–24 nt long
   sequences were aligned to mature canine miRNAs from miRbase release
   22.1^[121]19 using blast-based alignment allowing one mismatch and the
   count numbers for sequences aligned to the same miRNA were combined.
   Not-aligned sequences were discarded. Differential expression analysis
   was performed separately for EV and protein fractions, combining HC and
   UTI groups as a control group, and comparing with the UC samples.
   During the analysis based on DEseq2 (Version 1.36.0)^[122]46, replicate
   samples (libraries prepared from the same RNA) were collapsed together.

ddPCR validation

   Based on the sequencing data, 9 differentially abundant (DA)-miRNAs and
   3 stable controls were validated using ddPCR performed using TaqMan
   miRNA assays described in Supplementary Table 2. Three microliters of
   1:3 diluted RNA were converted to cDNA using TaqMan™ MicroRNA Reverse
   Transcription Kit (#4366596, Thermo Fisher Scientific) according to the
   manufacturers’ instructions. 1.5 µL of the cDNA were combined with 11
   µL of 2X ddPCR Supermix for Probes (#1863024, BIO-RAD), 8.4 µL of
   nuclease free water and 1.1 µL of assay primer (Supplementary Table 4).
   Droplets were generated and analysed using the QX200 Droplet Digital
   PCR System (BIO-RAD).

Target prediction

   The biological targets of candidate miRNAs (miR-182 and miR-221/222)
   were predicted using two databases, miRDB v6.0^[123]20 and TargetScan
   v8.0^[124]21, since they have predictions specifically for dogs. For
   miRDB, genes with target scores lower than 60 were filtered out as the
   default setting, while all of the results from TargetScan were kept.
   Overlapped genes from the two databases were kept for further analysis,
   of which 377 genes were found for miR-182 and 121 were found for
   miR-221/222. A web-based tool Enrichr^[125]22 was used to analyze these
   two gene sets for the enrichment of common annotated biological
   features. Due to limited pathway information specific to dogs, we used
   general databases for mammals/humans instead, including the Elsevier
   Pathway Collection, the KEGG 2021 Human^[126]47 and the BioCarta 2016
   in Enrichr.

   To confirm if these miRNAs are involved in the regulation of UC
   progression, we collected DE gene lists from three recently published
   articles comparing canine UC and healthy
   controls^[127]23,[128]24,[129]48 and performed hypergeometric tests to
   assess the enrichment of these gene lists using the phyper function
   (Package stats version 4.3.1) in R with the parameter for total target
   genes (n) given as the mean of the number (18,092.5) of unique target
   genes in miRDB (16,710) and TargetScan v8.0 (19,475). The final p
   values were adjusted using the Benjamini-Hochberg (BH) method.

   Furthermore, we downloaded the available invasive UC bulk RNA-Seq
   datasets from NCBI SRA with accession numbers DRA005844^[130]24 and
   SRP217874^[131]49. Differential expression analysis was performed using
   the edgeR package in R (version 3.42.4). Batch effects between studies
   were considered in the linear model. Differentially expressed genes
   were defined as |logFC|> 1 and FDR < 0.05 (adjusted by the
   Benjamini–Hochberg method). PCA was calculated using the prcomp
   function in R. Plots were produced using the ggplot2 package (version
   3.4.2) in R.

Statistical analysis

   Numerical data were compared between groups using the Mann Whitney U
   test. For comparison of ddPCR data between groups, the abundance of
   each candidate miRNA was normalised to the abundance of a selected
   normalisation miRNA that had a similar expression level to the
   candidate miRNA. This normalised candidate miRNA abundance was then
   subsequently normalized to the sample creatinine content, to correct
   for variation in sample volume used and sample concentration between
   individuals.

   The accuracy of each normalised candidate miRNA (indexed to urinary
   creatinine content) for distinguishing dogs with UC from dogs with UTI
   (combining data from ‘sequencing’ and ‘validation’ cohorts) was
   evaluated using receiver operator curve analysis.

Supplementary Information

   [132]Supplementary Information 1.^ (1.3MB, docx)
   [133]Supplementary Information 2.^ (14.5KB, xlsx)
   [134]Supplementary Information 3.^ (10.4KB, xlsx)
   [135]Supplementary Information 4.^ (9.8KB, xlsx)

Acknowledgements