DNMT3A mutation leads to leukemic extramedullary infiltration mediated by TWIST1

We collected five AML cell lines (OCI-AML3, THP-1, Kasumi-1, U937, and NB4) to examine their migration capacity by using transwell assay. After 18 h of incubation, the number of migrated cells of OCI-AML3 was higher than those of four other cell lines (p < 0.001) (Fig. 1a). OCI-AML3 cells harbor D3Amut and NPM1 abnormality. NPM1 mutation is considered a marker for relatively good prognosis, whereas D3Amut is associated with poor outcome. Therefore, we focused on the possibility of enhanced invasion of malignant cells rendered by DNMT3A R882 mutation, which is an established poor prognosis indicator. The protein level of DNMT3A in OCI-AML3 was not the highest among the five cell lines investigated (Additional file 1: Figure S1a). Subsequently, all exons of DNMT3A gene were sequenced in the five lines, and the results showed the existence of D3Amut only in OCI-AML3 (Additional file 1: Figure S1b). We further constructed two DNMT3A knockdown sublines of OCI-AML3 via specific siRNAs. The silence of DNMT3A could dramatically reduce the capacity of cell migration compared with control siRNA (p < 0.001) (Fig. 1b). To rule out the effect caused by WT DNMT3A, DNMT3A mRNA was downregulated in U937 cell line, which was derived from a monocytic leukemia case without D3Amut (Additional file 1: Figure S1b and S1c). Transwell assay showed that, with WT DNMT3A decreasing, migration capacity of U937 was not significantly altered (Additional file 1: Figure S1c).

To further assess the effect of D3Amut on cell migration, we overexpressed the WT and mutant DNMT3A in U937, which exhibited low migration capacity (Fig. 1a, c). Repeated independent transwell assays showed that the numbers of migrated U937 cells stably expressing DNMT3A R882C were higher than those without D3Amut (p < 0.001) (Fig. 1c). In addition to leukemia cells in suspension culture, the effect of D3Amut on migration behavior in adherent cells was also investigated. To eliminate the possible effect of endogenous DNMT3A protein, we tested a series of solid tumor cell lines and found little DNMT3A expression in MDA-MB-231, a breast cancer cell line (Additional file 1: Figure S1d). Scratch-wound experiment was conducted, and cell motility was efficiently increased in MDA-MB-231 cells expressing D3Amut but not in those expressing WT DNMT3A (Fig. 1d). Taken together, these data suggest that DNMT3A R882 mutation contributes to enhance the migration of malignant cells.

To evaluate the migration capacity of OCI-AML3 cell line in vivo, we transplanted OCI-AML3 cells into semi-lethally irradiated NOD/SCID mice through tail veins. Different quantities of OCI-AML3 cells, namely, 1 × 10⁶, 5 × 10⁶, and 1 × 10⁷ cells, were inoculated into mice. Around 20 days after xenografting, 36 of 37 animals investigated developed weakness in hind limbs and walking in unequal steps, with the symptoms appearing the earliest in mice inoculated with 1 × 10⁷ cells, followed by those with 5 × 10⁶ and 1 × 10⁶ cells (Fig. 2a). X-ray examination eliminated the possibility of pathological bone fracture in their hind limbs (Additional file 1: Figure S2a). Mice were executed when they became moribund. The average survivals in the three groups challenged with OCI-AML3 cells were 20.6, 23.9, and 30.0 days (p < 0.001 compared with the controls) (Additional file 1: Figure S2b). Mild splenomegaly was observed in sick mice. Immunophenotypic staining of human leukocyte-specific antigens hCD45 and hCD44 revealed the presence of OCI-AML3 cells in spleens (Additional file 1: Figure S2c), and the percentage of exogenous leukemia cells in the bone marrow (BM) was relatively low (Additional file 1: Figure S2c).

To explore the cause of paralysis presentation, we proposed the possibility of CNS abnormality caused by extramedullary leukemia. Positron emission tomography-computed tomography (PET-CT) was performed to investigate the hypermetabolic regions that are suggestive of tumor infiltration. Compared with those of normal controls, the spinal cords of sick mice contained more radioactive substances (Fig. 2b), supporting the infiltration of OCI-AML3 cells into CNS. To detect the disease status at the entire organism level, we transfected luciferase reporter lentivirus vector into OCI-AML3 cells and subsequently transplanted the cells into the mice. All transplanted animals developed a paralysis syndrome at 1 month post xenografting. Xenografts were observed to infiltrate the brains, spinal cords, and BMs by using bioluminescent imaging (Fig. 2c). Flow cytometry showed quite a lot of xenografted cells in the brain when hCD44 antigens were labeled (Fig. 2d). To examine whether other leukemia cell lines could cause CNS infiltration, we used 5 × 10⁶ cells of U937 and THP-1 lines, both from acute monocytic leukemia patients, for transplantation. Around 20 days post xenografting, U937 cell-transplanted mice developed severe leukemia, but few exogenous cells were detected in the brain (Fig. 2d). In THP-1 cell-transplanted mice, no leukemia infiltration was observed in cerebral tissues at 40 days post xenografting when the mice became moribund (Fig. 2d). The above phenomenon demonstrates that OCI-AML3 cells could infiltrate into murine CNS and induce characteristic clinical symptoms.

Histopathologic examinations were carried out to further verify the CNS-infiltrating characteristics. In sick mice tissues, myeloid leukemia cells with large and round shape, pale cytoplasm, and poly or band form nucleus were accumulated in the brain meninges (Fig. 3a) and spinal cord meninges (Fig. 3b). Immunohistochemistry results showed that these cells were also positive for hCD44 (Fig. 3c). When mice transplanted with GFP vector-containing OCI-AML3 cells were examined, GFP-positive cells were shown along the brain meninges (Fig. 3d). HE staining and immunofluorescence showed the presence of around 10 % of exogenous leukemia cells in murine BMs (Additional file 1: Figure S3a and S3b). In addition, the genetic lesions of R882 loci were detected to validate the exogenous cells. Targeted sequencing results showed that GFP-positive samples from BMs and brains all harbored R882C mutation, which was the same as the control cells (Additional file 1: Figure S3c). Thus, OCI-AML3 cells are verified to invade the brain and spinal cord meninges and induce CNS leukemia.

Leukemic CNS infiltration indicates poor clinical outcome. In this study, mice with a large number of leukemia cells that infiltrated the meninges tended to have worse general status. OCI-AML3 cells could infiltrate into CNS in vivo, and D3Amut could promote cell migration in vitro; hence, we speculated that inhibition of heterozygous D3Amut in OCI-AML3 cells could affect the ability of leukemic cell migration in mice. A non-targeting control shRNA (sh-CTL) and two shRNAs targeting DNMT3A (sh-DNMT3A-1 and 2) were stably transfected into OCI-AML3. In these two knockdown strains, cells with sh-DNMT3A-2 were collected for transplantation because the expression of DNMT3A was more reduced than that in other strain (Additional file 1: Figure S4). Thus, OCI-AML3 cells with sh-CTL and sh-DNMT3A-2 were purified in a number of 1 × 10⁶ cells for xenograft in NOD/SCID mice. A luciferase reporter lentivirus vector was subsequently brought into OCI-AML3-sh-CTL (shControl) and OCI-AML3-sh-DNMT3A-2 (shDNMT3A) cells. With the same numbers of transplanted cells, the motility of OCI-AML3 with reduced D3Amut was largely decreased in the brains and spinal cords of mice at 1 month post xenografting (Fig. 4a). Notably, mice inoculated with shDNMT3A cells had prolonged life spans than those with shControl cells (p < 0.001) (Fig. 4b). With regard to the mortality in the shControl group, the contamination of a small number of OCI-AML3 cells without DNMT3A knock down might compete against shDNMT3A to cause late leukemic explosion. These data show that abrogation of D3Amut could exert an anti-infiltration effect and reduce the invasiveness of leukemic cells.

To obtain insight into how D3Amut caused aberrant cell migration of OCI-AML3, we analyzed a group of EMT markers in OCI-AML3 strain. EMT-associated genes promote cancer cell plasticity and tumor metastatic spread, and the process can be induced by some factors, such as TWIST1 and SNAIL [21]. We observed that OCI-AML3 cells harbored high expression levels of TWIST1, SNAIL, and VIMENTIN. In other leukemia cell lines, these EMT markers were relatively in lower expressions except TWIST1 also highly expressing in THP-1 (Fig. 5a). These markers in CNS-infiltrating leukemia cells were further analyzed after cell sorting via GFP-tag (Additional file 1: Figure S5a). TWIST1, the most important transcriptional factor of EMT process, was highly expressed in the brain tissues compared with that in BMs (Fig. 5b). Subsequently, a series of AML patients’ primary BM samples that harbor DNMT3A R882H was analyzed. Interestingly, the mRNA and protein levels of TWIST1 were higher in mutant cases compared with those of DNMT3A WT ones (Figs. 5c, d). Immunofluorescence also showed that TWIST1 was highly expressed in primary leukemic cells with D3Amut (Additional file 1: Figure S5b). In the meanwhile, TWIST1 protein level was increased in U937 or MDA-MB-231 strain by overexpression of D3Amut but not WT DNMT3A (Additional file 1: Figure S5c). Also in Kasumi-1 cells, the expressional levels of TWIST1 were not changed after WT DNMT3A silencing so as to further exclude the possible effect made by WT DNMT3A (Additional file 1: Figure S5d). To evaluate the effect of D3Amut on the expression of TWIST1, OCI-AML3 cells stably transfected with sh-DNMT3A lentivirus displayed reduced TWIST1 with silenced D3Amut (Additional file 1: Figure S4). The OCI-AML3 cells in the brain had migrating capacity; thus, DNMT3A R882C was knocked down in those cells to determine whether phenotypes could be converted. The results showed that TWIST1 expression and cell migration were both reduced in brain-infiltrating cells (Fig. 5e). TWIST1 plays a key role during EMT. Therefore, we next transiently transfected two siRNAs targeting TWIST1 mRNA in brain-infiltrating cells. With decreasing TWIST1 level, DNMT3A remained in abundance, whereas the migration ability of brain-infiltrating cells was significantly reduced (Fig. 5f). This observation indicates that TWIST1 is probably a downstream of DNMT3A and could be involved in the regulation of leukemia cell mobility process.

The role of TWIST1 on the migration ability of leukemic cells in vivo was investigated. Two shRNAs (sh-TWIST1-1 and 2) targeting TWIST1 mRNA were stably transfected into OCI-AML3 cells by lentiviruses. Western blot analysis showed that both shRNAs successfully decreased TWIST1 expressions, and the knockdown effect was more remarkable in cells with sh-TWIST1-1 compared with shControl and sh-TWIST1-2 (Additional file 1: Figure S6a). Transwell assay also revealed that migrated cells were reduced after TWIST1 knockdown, particularly by sh-TWIST1-1 interference (Additional file 1: Figure S6b). Thus, we used sh-TWIST1-1 for further study. Exogenous leukemic cells with or without stable TWIST1 knockdown (shTWIST1 or shControl) were sorted through GFP tags and transplanted into NOD/SCID mice at a same number of 5 × 10⁵. One month later, shControl mice gradually showed paralysis symptom, but such phenotype was not observed in the group of shTWIST1 mice. Bioluminescent imaging showed that in addition to BMs, exogenous grafts in control mice were also distributed in the brains and spinal cords. However, the EMI of leukemic cells in shTWIST1 mice was few in CNS regions (Fig. 6a). Flow cytometry further revealed that hCD44-positive cells were detected in BMs of both control and shTWIST1 mice. Interestingly, the analysis of brain samples showed that the control one carried quite a few leukemic cells, whereas almost no xenograft was observed in shTWIST1 mice (Fig. 6b). Similar to the data from immunophenotyping assay, histopathologic examinations with HE staining also showed that transplanted cells were obviously noted in the marrow cavities and brain meninges in control mice. By contrast, leukemic cells in the shTWIST1 group were detected in BMs but not in the brain meninges (Fig. 6c). Immunohistochemistry confirmed few hCD44-positive cells infiltrated in the brains of shTWIST1 mice (Fig. 6d). Overall survival analysis demonstrated that shTWIST1 mice had longer life spans than control ones (p = 0.0022) (Fig. 6e). Additionally, all shTWIST1 mice were not paralyzed in their limbs even when they died because of leukemia. The results above indicate that TWIST1 is required for meninges migration of OCI-AML3 cells in mice, and TWIST1 abrogation could significantly reduce CNS infiltration.