Emerging experimental and clinical evidence has suggested that metastasis may be an earlier event in cancer progression than previously realized. Experimental studies in transgenic mice have observed the presence of breast cancer cells in the bone marrow even at the stage of atypical ductal hyperplasia or DCIS within the breast [1
]. In observational studies in humans, both circulating tumor cells (CTCs) and disseminated tumor cells (DTCs) in the axillary lymph nodes and bone marrow have been noted in early stage disease, even at the stage of in situ
Observations made in experimental model systems should be validated in actual human tumors, if possible, in order to strengthen the relevance of the findings to humans. Observations such as epithelial-mesenchymal transition (EMT), autophagy, emergence of the multidrug resistance phenotype and vasculogenic mimicry are examples of important phenomena of tumor progression observed experimentally but only with modest confirmatory support in human cancers [15
]. The findings of early metastasis in the experimental studies have begun to see some validations in the human situation. However more validations are certainly required. In the present study our experimental findings suggest cancer clusters can stimulate human myoepithelial cells or murine embryonal fibroblasts to engage in encircling lymphovasculogenesis which allows the clusters to become lymphovascular emboli. Our observational findings suggest that, in fact, in situ
carcinomas can directly become lymphovascular tumor emboli. Both findings support the conclusion that early metastasis can occur in both an experimental animal model system as well as in humans.
Our hypothesis is that mammary myoepithelial cells or myoepithelial stem cells committed to the myoepithelial or fibroblast lineage transform to either endothelial cells or a commitment to the endothelial lineage. Our hypothesis to explain the genesis of tumoral emboli within lymphovascular channels is to consider the possibility that myoepithelial cells which surround clusters of DCIS, or stromal stem cells which lie adjacent to clusters of invasive carcinoma might, under the right circumstances, transform into encircling endothelial cells, causing the clumps of in situ or invasive carcinoma cells to become tumoral emboli.
In the experimental studies, the gross pattern of fluorescence observed in the collective experiments supported the conclusions that the unlabeled tumoral spheroids initially stimulated the growth of the labeled HMS-1 and MEFs which subsequently stimulated the influx of host cells within the tumoral extracellular matrix. On IHC examination the presence of GFP/RFP immunoreactivity within the cells which lined these lymphovascular channels containing the tumoral emboli supported our conclusions that the tumor cell clusters were inducing encircling lymphovasculogenesis from the injected HMS-1 or MEFs. This encircling lymphovasculogenesis eventually formed communicating anastomoses with murine vessels. This is why hybrid fluorescence emerged in the GFP-transgenics injected with unlabeled tumor cells and RFP-labeled HMS-1 cells or MEFs. We did not observe fluorescence or encircling GFP/RFP immunoreactivity when the HDFs were coinjected with the unlabelled spheroids. We interpret these findings to indicate that myoepithelial to endothelial or embryonal fibroblast to endothelial transformation may be mediated by stem/progenitor cells present in either the HMS-1 or the MEFs populations capable of pluripotency and the ability to switch to endothelial differentiation. It would not be unanticipated that an embryonal line like MEFs or a benign myoepithelial tumor cell line like HMS-1 would contain such pluripotent stem/progenitor cells. On the other hand, mature adult fibroblasts such as HDFs might lack such pluripotent cells and be incapable of encircling lymphovasculogenesis or a switch in differentiation.
In the mouse studies, the coinjected RFP-labeled HMS-1 or MEFs will always emit a red color and hence in the setting of the GFP-transgenics, the presence of anatomizing vessels is suggested by a yellow hybrid color. These experiments alone do not necessarily mean that the myoepithelial cells and embryonal fibroblasts developed into cells which lined the lymphovascular channels but taken together with the in vitro studies (Figure ) and the other set of mouse studies showing GFP immunoreactivity in the cells lining the spaces (Figure ), the aggregate studies suggested this possibility. The yellow hybrid color observed in the murine studies would not further transition into a green color because there would always be a combination of RFP-labeled HMS-1 or RFP-labeled MEFs and murine GFP-labeled endothelial cells or endothelial precursors in the tumor microenvironment giving the hybrid yellow fluorescence.
Even though our experimental model utilized spheroids or clumps of invasive breast cancer cells and not true DCIS, when coinjected or cocultured with human myoepithelial cells, the myoepithelial cells encircled the spheroids both in vitro as well as in mice in the manner in which myoepithelial cells lie juxtaposed to DCIS in vivo. When the spheroids were coinjected or cocultured with murine embryonal fibroblasts, these latter cells similarly encircled the spheroids both in vitro as well as in mice in the manner in which stromal fibroblasts lie juxtaposed to invasive ductal carcinoma. Both the human myoepithelial cells and the murine embryonal fibroblasts developed into cells which lined the lymphovascular channels containing the tumor emboli. In both situations then, the carcinoma cells induced an encircling lymphovasculogenesis.
It would be interesting if we could directly show that the presumed new lymphovascular structures transported fluid or cells. However, direct evidence for the above would require extensive imaging studies beyond the scope of the present study. However we believe that we have indirect evidence to support the above conclusions. The new lymphovascular structures labeled with GFP contained luminal erythrocytes indicating that they were transporting circulating red blood cells (Figure ). Furthermore murine studies with coinjected MARY-X spheroids and labeled HMS-1 or MEFs exhibited spontaneous pulmonary metastasis which could only occur if the newly created lymphovascular structures were able to communicate with the resident lymphovasculature and transport both fluid as well as tumor cells.
We are not claiming in this study that the MARY-X model was a model of DCIS. Breast carcinomatous emboli are indeed invasive carcinoma and it is our hypothesis that they may gain access to lymphovascular channels by stimulating the growth of the latter around them. To show this, we used a model of invasive carcinoma, MARY-X, that possessed the phenotype of florid lymphovascular invasion. In terms of the experimental model, the induction of lymphovasculogenesis takes origin from either murine embryonal stem cells (MEFs) or human myoepithelial cells (HMS-1) within the tumor microenvironment. In the observational studies we used cases in which DCIS was seen juxtaposed to areas of lymphovascular invasion without intervening stromal invasion. In both the murine experimental studies as well as the human observational studies, the common phenomenon of the induction of encircling lymphovasculogenesis is illustrated. But in the experimental studies it is invasive carcinoma doing the induction. In the observational studies it is the DCIS.
In the observational studies, we noted that in situ carcinoma lay juxtaposed to LVI without intervening stromal invasion in 10 cases. Even after an exhaustive search for stromal invasion, we could not demonstrate it. Four possible explanations might still be considered to support stromal invasion as being the mechanism of lymphovascular invasion in these 10 cases despite our not being able to demonstrate it: a) stromal invasion might have occured a long time before the tissue was sectioned; b) evidence for stromal invasion might be present in an adjacent piece of tissue that was not subjected to sectioning; c) a very small number of tissues was analyzed in the present studies; d) one can not draw positive conclusions based on negative evidence. However in response to these considerations: a) the vast majority of human breast cancers show obvious and florid stromal invasion with lymphovascular invasion being rare. In these cases, the stromal invasion remains and does not disappear as the tumor grows. The proposed reason of why the stromal invasion is absent in our 10 cases was that it once was there and has now disappeared. This possibility is just not tenable; b) exhaustive sectioning was done and certainly much more than the routine degree of sectioning which is done in pathological analysis. If you applied this possible consideration to all of diagnostic pathology, you could argue that disease processes, like cancer, are routinely missed because their presence falls outside the area of sectioning. Although this can rarely occur, it certainly does not commonly occur or otherwise diagnostic pathology would be completely unreliable. In the 10 cases of the present study, a deliberate and intense search for areas of stromal invasion was carried out. We sectioned all areas of the cancer (outside, midzone and inside) and failed to find stromal invasion; c) we have analyzed 10 cases to date in which we have observed this phenomenon. Although this may be perceived as a small number, we believe it to be a significant number because it is these exceptions that can prove the rule. In other words in the typical case of infiltrating breast cancer, the phenomenon of encircling lymphovasculogenesis may also be occuring but it is not possible to prove because of the presence of stromal invasion juxtaposed to lymphovascular invasion. In these typical cases, one has to presume that stromal invasion progresses to lymphovascular invasion. However in the 10 cases where there is the absence of stromal invasion but the presence of lymphovascular invasion, this presumption can be excluded and one is left with support for our hypothesis: encircling lymphovasculogenesis; d) Although it is true that absence of proof is not absolute proof of absence, we are not drawing positive conclusions solely on the basis of negative evidence. We have positive evidence: the presence of clusters of DCIS juxtaposed to clusters of tumoral emboli within lymphovascular spaces. Both clusters show similar size, proliferation and immunocytochemical features. That is circumstantial but positive evidence. It is the totality of evidence, both positive as well as negative, that supports our conclusions.
Furthermore lymphatic markers, eg., D2-40, VEGFR-3, Prox-1 were completely absent in normal ducts as well as ducts containing DCIS. Therefore we believe that we have excluded the possibility that lymphatic markers are being expressed by a subset of myoepithelial cells that represent ducts containing DCIS rather than lymphatics containing tumor emboli.
There was also no evidence for iatrogenic seeding. We reasoned that the only tenable hypothesis that would support these findings was one of induced encircling lymphovasculogenesis. In all 10 cases (Table ), the morphometric and immunohistochemical evidence supported our hypothesis. It is important to note that the vast majority of DCIS in patients does not manifest LVI in the absence of stromal invasion. Most DCIS, when associated with progression, exhibits frank stromal invasion. LVI occurs after there is significant stromal invasion. But the findings of our present study indicate that this usual type of progression need not be canonical---that in situ carcinomas may progress to lymphovascular tumor emboli through an alternate “noninvasive” mechanism.
Because it could also be argued that the apparent tumoral emboli within lymphovascular spaces might, alternatively, be DCIS with separation artifact and a reduced or lost myoepithelial layer, we had to also consider this possibility. We do not believe that we are observing a retraction artifact phenomenon here. First of all, retraction artifact is usually observed in islands of invasive ductal carcinoma and not DCIS. Secondly, the DCIS clusters which were obvious were surrounded by p63 positive myoepithelial cells with no obvious separation artifact. The lymphovascular tumoral emboli were surrounded by CD31 and D2-40 positive endothelial cells. Thirdly the 10 cases of DCIS with LVI had metastases to axillary lymph nodes (Table ). These findings all suggested that the LVI is true and not retraction artifact around DCIS, the latter of which would not give rise to axillary metastasis.
Our morphometric and tumoral IHC studies (Table ) indicated no differences in immunoreactivity or size (perimeter) between the in situ clusters and the tumor emboli (p=0.5) (p=0.1). This suggested that these structures were one and the same and remained intact during their in situ to LVI transition.
With respect to this transition, the near identity of proliferation (Ki-67) and immunocytochemical adhesion markers (E-cadherin) further supported the hypothesis that these two structures were, in fact, related. If the structures were not related, that is if the DCIS occurred independently and the lymphovascular emboli were derived from invasive cells, it would be more likely that these markers within the emboli would change. Invasion is thought to involve epithelial-mesenchymal transition where there is loss of E-cadherin. Furthermore there is generally thought in tumor progression that there is clonal selection for a more aggressive phenotype with increased proliferation. It would then be expected that if the lymphovascular emboli were derived from stromal invasion, that their E-cadherin would be decreased and their Ki-67 would be increased compared to the DCIS foci. On the other hand, if the DCIS clusters and lymphovascular emboli were one and the same, it would be expected that E-cadherin and Ki-67 would be nearly identical between the two structures and that is exactly what was observed. Slightly different morphometric differences in shape between these structures, however, would not negate their identity of origin because the lymphovascular tumor emboli would likely be subjected to hydrostatic pressures of lymph and blood flow that could easily alter their shape. Our collective histological and immunocytochemical findings, though not proving our hypothesis, support it. Although there are alternative explanations, the one that is most consistent with Occam's razor is in situ – LVI transition.
Our hypothesis to explain the genesis of tumoral emboli within lymphovascular channels then is to consider the possibility that myoepithelial cells which surround clusters of DCIS or stromal cells which lie adjacent to clusters of invasive carcinoma might, under the right circumstances, transform into encircling endothelial cells, causing the clumps of carcinoma cells to become tumoral emboli. If one invoked the canonical hypothesis of invasion into pre-existing lymphatics and blood vessels to explain LVI, one would expect to see similarities in immunoreactivity between vessels containing and devoid of emboli since all of these vessels would have antedated the LVI. But our study observed CD31, D2-40, VEGFR-3 and Prox-1 differences in the embolic channels. Newly derived vascular channels would be expected to be immature and express endothelial markers, podoplanin (D2-40), VEGFR-3 and PECAM-1 (CD31), to a lesser degree than mature preexisting vessels (Table ). If one invoked the invasion hypothesis to explain LVI, p63 immunoreactivity would be decreased in the DCIS-surrounding myoepithelial layer but not persistent within the CD31 and D2-40 lymphovascular spaces containing emboli. Our myoepithelial and endothelial IHC studies then also supported the concept of encircling lymphovasculogenesis.
If we limited our studies to the use of a single endothelial marker, we would have had to be more tentative in our conclusions. The use of single markers would limit the significance of our conclusions because in the tumor microenvironment slight changes might occur in the expression of a single gene, without implying any significance. While we did not exhaustively examine the expression of every known endothelial marker, in our experimental studies we studied the induction of VEGFR-1, VEGFR-2, VEGFR-3, CD31 and podoplanin, five different endothelial markers. In our human studies we studied the endothelial immunoreactivites for four markers, CD31, D2-40, VEGFR-3 and Prox-1. Random co-expression of several endothelial markers is unlikely and for this reason, we believe that our data supports our hypothesis. Nevertheless our findings could have been strengthened by the use of additional markers of lymphatic endothelium including neuropilin-2, FOXC2, CCR7, CCL19, CCL21 and the mannose receptor.
With these findings and this reasoning as background, we also observed lymphovasular channels with dual immunoreactivity. These channels showed CD31 and D2-40 immunoreactivities (red) as well focal p63 nuclear immunoreactivity (brown) (Figure ). The focal brown nuclear immunoreactivity can be contrasted with adjacent non-immunoreactive blue endothelial and tumoral nuclei present within the same section. The use of confocal microscopy with immunofluorescent staining would be theoretically helpful in confirming these findings, but because this was a retrospective study using paraffin-embedded archival pathological materials, the degree of autofluorescence would confound meaningful interpretation of confocal immunofluorescent studies. We therefore, were not able to use this additional approach.
It had been known that myoepithelial cells exert tumor suppressive effects on DCIS [30
]. More recently it had been demonstrated that myoepithelial cells can be paracrinely regulated by DCIS [33
] and undergo alterations in gene expression, and promoter methylation [35
]. In these studies the findings suggested that myoepithelial cells can become less differentiated and manifest aberrant gene expression including expression of endothelial-related genes [2
]. In our in vitro
experimental studies of the induction of endothelial differentiation within the myoepithelial cells, we observed an increase in podoplanin (the antigen recognized by D2-40), VEGFR-3, VEGFR-1, VEGFR-2 and CD31 transcripts over time when myoepithelial cells were cocultured with MARY-X spheroids. These studies suggest that endothelial (both lymphatic as well as vascular) proteins increase as the commitment to the endothelial lineage is made. Interestingly, we tested MARY-X for all the VEGF growth factors by real time PCR and it expressed relatively higher levels of VEGF-C and VEGF-D than VEGF-A and VEGF-B compared to a number of common ER positive and ER negative breast cancer cell lines. MARY-X would be expected therefore to stimulate lymphangiogenesis. Still the other breast carcinoma lines were not able to increase VEGFR-3, CD31 or podoplanin transcripts in HMS-1 cells, yet, in at least some of these other cell lines, VEGF-C transcripts were increased and presumbably VEGF-C was produced. The mechanism of induction of endothelial differentiation in myoepithelial cells may not therefore be mediated by the classic effectors of lymphangiogenesis, eg., VEGF-C.
We believe that we have provided supportive evidence for the acquisition of the lymphatic phenotype by myoepithelial cells in vitro. In the studies depicted in Figure , we examined the levels by RT-PCR of a number of endothelial markers in HMS-1 cells cocultured in media with MARY-X spheroids. At time 0, the expression of five different endothelial markers: VEGFR-1, VEGFR-2, VEGFR-3, CD31 and podoplanin was very low but each increased to various degrees when cocultured with MARY-X spheroids. The lymphatic endothelial markers, VEGFR-3 and podoplanin and the general endothelial marker, CD31 increased the most. We examined the presence of endothelial growth factors in MARY-X and compared the levels to other breast carcinoma lines and found that MARY-X makes high levels of VEGF-C. We did not measure the levels of VEGFR in MARY-X because this was not central to the present hypothesis and because in our analysis of the HMS-1 cells, there was no possibility of contamination by MARY-X. HMS-1 cells grew as a monolayer and the MARY-X spheroids grew in suspension without the ability to attach and the two cell lines are easily separated from each other. Furthermore at time 0 (Figure ), levels of VEGFR and other endothelial markers were low in HMS-1 cells and increased over time. This increase could only occur if there was an induction of endothelial differentiation of HMS-1 cells by the MARY-X spheroids. The rationale for studying the expression of VEGF in MARY-X is that we wanted to know what is different in MARY-X that might be responsible for its unique ability to induce encircling lymphovasculogenesis and possible candidates would include VEGF family members. VEGF-C, despite some conflicting data, represents an attractive candidate for future studies.
In the experimental models where induced vasculogenesis and/or vasculogenic mimicry has been observed [15
], the phenomenon facilitates metastasis from the standpoint that the vascular channels which are created anastomose with the resident vasculature. We can not tell in our observational studies whether anastomoses are occuring as we can not see three dimensions in two dimensional sections.
Though the lymphovascular tumor emboli were of similar size as the in situ carcinoma clusters (Table ), the lymphovascular channels containing the tumor emboli exhibited evidence of immaturity (Table ). Channels derived from encircling lymphovasculogenesis would be expected to be immature. We can not directly validate or investigate this assumption, however, in our observational studies. However in our experimental in vitro induction studies of endothelial differentiation of myoepithelial cells, we observed an increase in podoplanin, VEGFR-3, VEGFR-1, VEGFR-2 and CD31 transcripts over time. These studies suggest that endothelial (both lymphatic as well as vascular) proteins increase as the commitment to the endothelial lineage is made during encircling lymphovasculogenesis.
In a recent study which observed strong correlation of extensive retraction artifact in the primary carcinomas with nodal metastases, it was postulated that the retraction artifact seen around tumor nests might be an early stage of LVI, where the conversion of mesenchymal cells to endothelial cells had not yet occurred [42
]. Retraction artifact may also then be a reflection of early encircling lymphovasculogenesis.
Finally, it is known that “pure” DCIS can be associated with nodal metastases and explanations for this phenomenon have been proposed ranging from sampling bias, microinvasion misinterpretation, iatrogenic dissemination and “revertant DCIS”, a phenomenon where DCIS reverted back from invasive carcinoma [43
]. In light of the data presented in the present study one might propose yet another explanation --- encircling lymphovasculogenesis to explain how initially “pure” DCIS might metastasize without invading.
In both our experimental as well as observational studies, we have not shown that myoepithelial cells or embryonal fibroblasts are capable of direct transdifferentiation into endothelial cells. Even though coinjection experiments with HMS-1 and MEFs in mice suggested that they can form lymphovascular channels via encircling lymphovasculogenesis and that immunocytochemical observations in human cases indicated p63 nuclear/CD31 membrane and p63/D2-40 dual immunoreactivities within the same lining cell, stem cells within HMS-1 or MEFs or within the breast undergoing a switch in differentiation from myoepithelial or fibroblast to endothelial could account for our findings.
We certainly do not know the mechanism for the induction of this encircling lymphovasculogenesis observed experimentally in vitro or in mice or in the in situ–LVI cases. The resemblance of the in situ carcinoma clusters, the lymphovascular tumor emboli and the spheroids to the embryonal blastocyst raises the possibility that embryonic lymphovasculogenic signaling pathways may be involved.
Our findings acknowledge the general importance of endothelial progenitor cells to tumor lymphovasculogenesis. Recent evidence, for example, has argued that some endothelial progenitor cells can reside in the bone marrow [20
]. All endothelial cells do not need to take origin from pre-existing endothelial or endothelial precursor cells, however. And not every case of human cancer need involve active lymphangiogenesis [44
]. But what we are arguing in this study is that active lymphovasculogenesis can occur and when it occurs, can occur from mesenchymal or myoepithelial precursors.
In these experimental studies we used MARY-X. Since MARY-X represents an unusual type of breast cancer with a somewhat unique gene expression profile, cellular properties and behavior in vivo, one might question the general applicability of this model to other common forms of breast cancer. If one focuses on the lymphovascular embolus, however, which is also present in non-IBC breast cancer, one might reason that the exaggerated phenotype of LVI exhibited by MARY-X provides a model to help understand the phenomenon of LVI exhibited by common forms of breast cancer.
Our use of the term “encircling lymphovasculogenesis” must be distinguished from “circumferential lymphangiogenesis”, which has been used to describe a different process [46
]. This latter term refers to tumor-dependent induction of lymphangiogenesis in peritumoral normal tissue surrounding the tumor mass. Our definition of the former term is the development of lymphovascular channels which envelop or encircle tumoral clumps creating lymphovascular tumoral emboli. The mechanisms behind “encircling lymphovasculogenesis” and “circumferential lymphangiogenesis” may be similar, however.
The occurrence of encircling lymphovasculogenesis does not negate, however, the generally accepted phenomenon of classic lymphovascular invasion which may still represent the usual and dominant route of metastasis. However this pathway of classic lymphovascular invasion need not be obligate. Alternate pathways such as encircling lymphovasculogenesis may also be operating.
Collectively our experimental as well as our observational studies suggest that breast cancer tumor emboli may result from encircling lymphovasculogenesis rather than conventional lymphovascular invasion. This phenomenon may help short-circuit some of the steps of the metastatic process which may indeed give rise to “early metastasis”.