Inflammatory state of lymphatic vessels and miRNA profiles associated with relapse in ovarian cancer patients

Lymphogenic spread is associated with poor prognosis in epithelial ovarian cancer (EOC), yet little is known regarding roles of non-peri-tumoural lymphatic vessels (LVs) outside the tumour microenvironment that may impact relapse. The aim of this feasibility study was to assess whether inflammatory status of the LVs and/or changes in the miRNA profile of the LVs have potential prognostic and predictive value for overall outcome and risk of relapse. Samples of macroscopically normal human lymph LVs (n = 10) were isolated from the external iliac vessels draining the pelvic region of patients undergoing debulking surgery. This was followed by quantification of the inflammatory state (low, medium and high) and presence of cancer-infiltration of each LV using immunohistochemistry. LV miRNA expression profiling was also performed, and analysed in the context of high versus low inflammation, and cancer-infiltrated versus non-cancer-infiltrated. Results were correlated with clinical outcome data including relapse with an average follow-up time of 13.3 months. The presence of a high degree of inflammation correlated significantly with patient relapse (p = 0.033). Cancer-infiltrated LVs showed a moderate but non-significant association with relapse (p = 0.07). Differential miRNA profiles were identified in cancer-infiltrated LVs and those with high versus low inflammation. In particular, several members of the let-7 family were consistently down-regulated in highly inflamed LVs (>1.8-fold, p<0.05) compared to the less inflamed ones. Down-regulation of the let-7 family appears to be associated with inflammation, but whether inflammation contributes to or is an effect of cancer-infiltration requires further investigation.


Introduction
tumours. In inflamed LECs, upregulated miR-1236 reduced expression of vascular-97 endothelial growth factor-3 receptor, inhibiting lymphangiogenesis [18]. A separate study 98 showed that inflamed dermal LECs increased miR-155 expression, a miRNA with known 99 involvement in PD-L1 expression which may contribute to peripheral tolerance changes [19]. 100 Furthermore, a panel of differentially expressed miRNAs was identified in inflamed rat LECs 101 and human LEC culture, indicating effects on inflammation, angiogenesis, endothelial 102 mesenchymal transition, cell proliferation and senescence pathways [13]. Tumour expression 103 profiling has identified changes in lymphangiogenesis-inducing miRNAs, such as miR-126 in 104 oral cancer and miR-128 in non-small cell lung cancer, that could influence metastasis 105 [20,21]. A study using human LEC co-culture models of gastric cancer, representing 106 tumoural or draining LVs, identified changes in differential miRNA expression associated 107 with lymphangiogenesis and with positive lymphatic metastasis [22]. Further miRNA 108 expression studies of draining LVs are lacking but many gene dysregulations were identified 109 in LECs isolated from afferent LVs that drained a metastatic gastric tumour in rats, including 110 up-regulation of chemokine CXCL1 [23]. 111

112
The aim of this feasibility study was to provide insight into, and assess the prognostic value 113 of the changes in the inflammatory state and miRNA expression of the transporting LVs in 114 EOC. It is expected that these changes will be indicative of the immune response in the LVs 115 and the downstream LN, and therefore could be predictive of patient prognosis. 118 LVs were collected from patients undergoing debulking surgery for advanced stage (FIGO 119 IIC, III or IV) epithelial ovarian cancer at Hammersmith Hospital, Imperial College 120

LV sample collection
Healthcare NHS Trust, London, United Kingdom (Tables 1 and S1). Informed consent was 121 provided from each patient. Ethics committee approval was obtained from Hammersmith and 122 Queen Charlotte's and Chelsea Research Ethics Committee (REC reference: 05/Q0406/178), 123 and tissue samples were provided by the Imperial College Healthcare NHS Trust Tissue Bank 124 (ICHTB). The following LV processing steps are summarised in S1 Fig.  125 126 Macroscopically normal appearing LVs along the external iliac vessels with surrounding 127 fibro-fatty tissue were removed with the nearby LN as part of the debulking procedure. 128  flow cabinet (Fig 1 A-C). Each vessel was divided into two sections. One half of each vessel 153 was further cleaned of surrounding fatty tissue, placed in RNA stabilising solution fixed in 4% paraformaldehyde (VWR, UK) for 45 minutes, followed by submersion in 15% 156 sucrose for 3 hours, then 30% sucrose for 6-12 hours. This was followed by embedding in 157 OCT (Scigen, UK) using dry ice and isopropyl alcohol (ThermoFisher, UK) and stored at -158   Table 1). 60% of the lymph 260 vessels were harvested at primary debulking surgery and 40% at interval debulking surgery 261 (S1 Table). Only one positive para-aortic LN was detected at final histopathology. In the 262 13.2±3.6 months follow up period, 4 of the 10 patients had relapsed. Of the four patients that 263 relapsed, three vessels demonstrated high inflammation and one vessel showed medium 264 inflammation. Vessels from patients that did not relapse showed low inflammation in 4 cases 265 and medium inflammation in the remaining 2 cases. Of the four relapse cases, cancer-cell 266 infiltration was detected in three vessels (two with high inflammation and one showing 267 medium inflammation). The overall relative range of inflammation in the lymphatic vessels 268 was small with high inflammation defined as an average of >10% MPO:DAPI ratio and a 269 narrow range of an average of 3-10% MPO:DAPI ratio for medium inflammation. All vessels 270 showed less inflammation than similarly stained ovarian tumour tissue (Fig 2I). 271 Consequently, to compare LVs with a notable difference in inflammation we focused initially 272 on comparison of the LVs with high inflammation versus low inflammation.  (Table 1). Tumour cells were detected in 50% of 289 LVs with medium inflammation and two-thirds of LVs displaying high inflammation. The 290 LVs presented with high inflammation in 3 of 10 samples, two of which were also positive 291 for cancer cells. Three LV samples presented with medium inflammation, two of which 292 contained cancer cells, and the remaining 4 LV samples presented with low inflammation and 293 no detectable cancer cell infiltration ( Table 1). The samples were macroscopically normal in 294 appearance, however positive staining for the ovarian tumour markers WT1 and Pax8 was 295 observed in four of the 10 samples collected (Fig 2 G,K). The LVs with cancer cells showed 296 (p=0.114). A moderate but statistically non-significant agreement was found between cancer-298 infiltrated LVs and Stage IV and inflammation (κ =0.583, 0.61 respectively). 299 300   Expression analysis of a panel of miRNA involved in auto-immunity and inflammation was 334 carried out. This showed that let-7 was differentially expressed in the highly-inflamed LVs 335 (Table 4, S2 Table). Significant differential expression of 8 miRNA was identified when 336 comparing miRNA expression of LVs presenting with high inflammation versus low inflammation (p<0.05, n=7). Of the 8 miRNA, 6 belonged to the let-7 family with miR-23a-3p and miR-23b-3p completing the set. Differential expression can also be clearly visualised 339 when the estimated probability density functions, calculated using kernel density estimation 340 for inner groups, for the differentially expressed individual miRNA are compared (S4 Fig). 341 Correction for multiple testing by applying a post-hoc Bonferroni correction showed that 342 differential expression of three miRNA, let-7b, let-7c and let-7d remained significant 343 (p<0.0017) . Comparison of miRNA expression in LVs displaying low inflammation versus 344 medium or high inflammation showed a similar trend towards down-regulation of the let-7 345 family (Table 4, S3 Table). Six miRNA were down-regulated >1.8 fold with significant 346 down-regulation of let-7g-5p and let-7i-5p (p<0.05). 347 348 identified in relapse LVs (let-7c-5p, let-7b-5p, miR-23b-3p and miR-186-5p), with three of 359 these (let-7c-5p, let-7b-5p, miR-23b-3p), also significantly differentially expressed in highly 360 inflamed LVs (Table 5). Differential expression of miRNA with correction criteria of 361 p<0.002 was not identified. Of the 17 miRNA that were down-regulated >1.8 fold in relapse 362 vessels, 8 were also down-regulated in inflamed LVs, and 11 in the highly-inflamed LVs, 363 including 5 and 6 members of the let-7 family respectively and miR-23b-3p (S2, S3 and S5 364 Table). 365 regulation such as miR-16-5p, miR-93-5p and miR-497-5p, miR-381-3p, miR-144-3p and 374 miR-181a-5p (p<0.05) ( Table 5). Of these, up-regulation of miR-93-5p and miR-144-3p 375 remained significant with application of a Bonferroni correction (p<0.002). In contrast, 376 inflamed LVs presented with primarily down-regulated miRNA, in particular, significant 377 down-regulation of members of the let-7 family along with miR-23a-3p, miR-23b-3p and 378 tumour suppressor miR-15-5p [28]. In the cancer-containing LVs, only 7 miRNA showed a 379 >1.8 fold down-regulation, 6 of which were members of the let-7 family (S4 Table). Of these 380 six, five were also down-regulated >1.8-fold in the LVs of patients who relapsed during the 381 study period. 382 383 Similar trends in the differential miRNA expression in cancer-infiltrated LVs and highly 384 inflamed LVs were also identified, as 15 miRNA were found to undergo an expression 385 change of >1.8 fold (up-or down-regulated) in both of the comparisons (S2 and S4 Table). 386 Six of the 7 miRNA that were down-regulated in both groups belonged to the let-7 family. 387 This common trend in the fold-regulation difference in the highly inflamed and cancer-388 infiltrated LVs may be reflective of the fact that 50% of the cancer-infiltrated LVs also 389 presented with high inflammation. We also noted that the number of differentially expressed 390 miRNA was higher when comparing LVs presenting with high or low inflammation than 391 when comparing vessels with or without cancer-infiltration. Expression of several miRNA 392 were consistently below qPCR detection levels and thus omitted from further analysis.  Table). 408 409

samples.
415 PCA showed that differential miRNA expression identified in inflamed and cancer-infiltrated 416 LVs could also be identified without grouping the samples. This can be seen when assessing 417 the top 16 miRNA whose expression contributed the most to the first and second Principle 418 Components (PC1 and PC2). These components capture factors responsible for the most and 419 second most variation in the data set, respectively. Samples with the most similar miRNA 420 expression profile cluster together and clusters may vary over PC1/PC2 or both (S5 Fig).  (Table 8) showed a similar prediction accuracy compared to classification of relapse LVs, but 442 the Random Forest classifier achieved the best accuracy. Classification of inflamed and 443 <=Stage IV samples did not show a high accuracy (S7 Table). As a further evaluation, we 444 examined the effect of keeping as predictors only those miRNAs that have significant 445 associations with the outcome variables. Analysis was performed on classifiers build using 446 only significantly differentially expressed miRNAs (p<0.05). In the cancer-infiltrated LVs, 447 the differentially expressed miRNAs enhanced the accuracy of the predictive algorithm to 80-448 addition of miR-let-7b-5p and let-7c-5p improved the accuracy of the Support Vector 452 Machine from 80% to 90% but decreased the accuracy of Logistic Regression from 80% to 453 70%. However, the use of miR-23b and miR-186-5p only showed a bias towards classifying 454 the samples as non-relapsed when using Logistic Regression, which decreased when using all 455 4 miRNA. These results suggest that the miRNAs with significant associations to the 456 outcome variables have a capacity to be used as predictors for predicting the outcomes of 457 interest in new patients. 458 459  association with high inflammation. The lack of a significant correlation in both cases may 499 also be simply due to the possible scenario that cancer cells are just transiting through these 500 vessels, so detecting their presence becomes a matter of timing. Overall, this study suggests 501 that nearby LVs appear to play an important role in the immune response to cancer and the 502 association between inflammatory lymphatic changes and relapse deserves further 503 exploration while potentially representing a therapeutic target. 504

505
The use of miRNA expression classification algorithms also showed potential to predict 506 relapse, with relatively high accuracy for these 10 patients when applying a k-Nearest 507 Neighbours algorithm or a Support Vector Machine model (80%), rising to 90% when using 508 significantly differentially expressed miRNA only (Table 7, (Tables 2 and 3, S7 and S8 Table). Furthermore, due to the limited number of highly inflamed LVs, it was not possible to build a reliable classifier based on miRNA 516 expression in LVs with high inflammation versus miRNA expression in LVs with low or 517 medium inflammation. Indeed, all the classifiers are limited by the number of samples 518 required to train the predictive algorithms. Therefore, whilst providing insight, these 519 observations require increased sample input to maintain confidence in miRNA classification. 520 521 Analysis of miRNA expression of highly inflamed LVs compared to LVs with no or low 522 inflammation, showed down-regulation of 6 members of the let-7 family, 3 of which 523 remained significant with a Bonferroni correction (p<0.0017). A similar trend was observed 524 when comparing miRNA expression of LVs presenting with high or medium inflammation to 525 those with low inflammation. In this case 6 members of the let-7 family showed down-526 regulation >1.8 fold but only two members showed significant down-regulation (p<0.05), 527 which may be reflective of less difference in mean LV inflammatory state between the two 528 groups compared. The change in let-7 miRNA expression was large enough that without 529 designating groups, the overall variability in sample miRNA expression due to let-7 530 expression could be detected using PCA (S5 Fig). In ovarian tumour tissue, let-7 expression 531 is down-regulated and associated with poor prognosis and chemo-resistance [17,29,30]. Let-7 532 family members target oncogenes HMAG2 and RAS, and loss of let-7 is also associated with  Table 9. A logical progression of this 562 study would be to confirm targets in LECs. 563  The pathway analysis of the dysregulated miRNA identified involvement in transforming-587 growth factor TGF-β2 signalling, as well as glycoproteins in cancer and molecules that 588 contribute to the glycocalyx of the LV lumen (Table 6) Table. Table of Table). Similar improvements were found in the accuracy of classifiers 730 predicting relapse and LV cancer-infiltration (Tables 7 and 8).