Leukemic stem cells (LSCs) adhere to bone niches through adhesion molecules. These interactions, which are deeply reorganized in tumors, contribute to LSC resistance to chemotherapy and leukemia relapse. However, LSC adhesion mechanisms and potential therapeutic disruption using blocking antibodies remain largely unknown. Junctional adhesion molecule C (JAM-C, also known as JAM3) overexpression by LSCs correlates with increased leukemia severity, and thus constitutes a putative therapeutic target. Here, we took advantage of the ability of nanoscopy to detect single molecules with nanometric accuracy to characterize junctional adhesion molecule (JAM) dynamics at leuko-stromal contacts. Videonanoscopy trajectories were reconstructed using our dedicated multi-target tracing algorithm, pipelined with dual-color analyses (MTT2col). JAM-C expressed by LSCs engaged in transient interactions with JAM-B (also known as JAM2) expressed by stromal cells. JAM recruitment and colocalization at cell contacts were proportional to JAM-C level and reduced by a blocking anti-JAM-C antibody. MTT2col revealed, at single-molecule resolution, the ability of blocking antibodies to destabilize LSC binding to their niches, opening opportunities for disrupting LSC resistance mechanisms.

Advances in fluorescence microscopy over the past decades have allowed for detecting a single molecule (SM) with nearly nanometer accuracy. These innovations, initially conceptualized by Werner Heisenberg (Heisenberg, 1949), and which led to the award of the Nobel Prize to Eric Betzig (Betzig, 1995; Betzig et al., 2006), Stephan Hell (Hell, 2007; Hell and Wichmann, 1994) and William Moerner (Dickson et al., 1997; Moerner and Kador, 1989) in 2014, are termed super-resolution microscopy or nanoscopy. These approaches, first optimized on fixed cells, then extended to living cells, recently revealed an intense and unexpected dynamics of macromolecular structures, such as synapses and focal adhesions (Sergé, 2016). This completely revised the classic view of these as static structures, with a molecular composition essentially constant over time. Thus, it is now thought that major players, like integrins and their adapters, are finely regulated by mechanisms such as association/dissociation, local diffusion within the membrane and/or direct targeting by local exocytosis (Bakker et al., 2012; Ishibashi et al., 2015; Rossier et al., 2012). This dynamic conception of cell adhesion has started to reveal an unsuspected plasticity that is modulated by physiopathological conditions and signals received by the cell.

Hematopoiesis is the process by which hematopoietic stem cells (HSCs) replenish platelets, red blood cells and immune cells over lifetime. It occurs in the bone marrow (BM) of adult mammals and requires the retention of HSCs in stromal niches, constituting specialized microenvironments controlling HSC quiescence, proliferation and differentiation into transient amplifying progenitors. Acute myeloid leukemia (AML) cells are hierarchically organized in a similar manner to normal hematopoietic cells. The bulk of leukemic cells originates from leukemic stem cells (LSCs), which retain their capacity for self-renewal, pluripotency and quiescence. LSCs, which are responsible for cancer relapse, are functionally defined as leukemic-initiating cells, and are a subset of leukemic cells able to initiate AML in xenografted mice. Crosstalk interactions between HSCs and their surrounding stroma leads to reciprocal exchange, to the benefit of the metabolism of both cells (De Grandis et al., 2015). This process is particularly important for LSCs, promoting a higher resistance to chemotherapy and favoring relapse after treatment (Schepers et al., 2015), a process called cell-adhesion-mediated drug resistance (Meads et al., 2008). Of note, cell adhesion to the extracellular matrix has already been studied by single-molecule tracking (SMT) and super-resolution approaches, notably addressing the dynamics of integrins and associated molecules (Sergé, 2016). However, cell–cell adhesion is experimentally more challenging to investigate since it inherently occurs between the two membranes of contacting cells, or by replacing one cell by a lipid bilayer, while integrin-mediated cell adhesion can be studied on extracellular matrix fibers deposited directly on the coverslip.

The junctional adhesion molecule (JAM) family contains three main members, JAM-A (Martin-Padura et al., 1998), JAM-B (Aurrand-Lions et al., 2001) and JAM-C (Arrate et al., 2001), also known as JAM1, JAM2 and JAM3, respectively. JAM-C expressed by normal HSCs interacts with JAM-B expressed by BM stromal cells, contributing to the maintenance of HSC quiescence (Arcangeli et al., 2011, 2014). In a follow-up study, we found that LSCs express JAM-C, a new biomarker for disease outcome in AML (De Grandis et al., 2017). LSC expression of JAM-C correlates with increased adhesion to BM stromal cells and is correlated with poor disease outcome. This suggests that JAM-C is involved in LSC resistance to chemotherapy by supporting adhesion of LSCs to the stroma. Thus, better understanding of the dynamics of LSC adhesion to BM stromal cells is critical for developing new therapeutic strategies targeting LSCs.

Here, we characterized JAM dynamics during LSC interactions with stromal cells by advanced nanoscopy and analytic approaches. SM trajectories were reconstructed using our in-house algorithm called multiple-target tracing (MTT) (Rouger et al., 2012; Sergé et al., 2008). SMT was completed with a set of analytic tools, named MTT2col for MTT dedicated to two colors, delivering maps of SM trajectories and colocalizations, SM axial-position profiles in z-stacks, trajectory descriptors and classification according to their anomalous diffusion exponent (Saxton, 1994). We focused on the role and perturbations of JAM interactions at the SM level during the establishment of JAM-C-expressing LSCs contacting JAM-B-expressing stromal cells. AML cell lines expressing high levels of JAM-C or not were used, mimicking LSCs or not, respectively (De Grandis et al., 2017). Our trajectory analysis revealed: (1) how JAM-B and JAM-C diffusion is modulated at contacts between stromal and leukemic cells, (2) JAM-B–JAM-C interaction events during the establishment and stabilization of contacts, (3) the strengthening of contacts at high levels of JAM-C expression, in relation with the characterization of JAM-C as an LSC marker, and (4) the weakening of contacts upon treatment with a blocking anti-JAM-C antibody. Taken together, the results of these innovative analyses reveal how SM dynamic measurements provide fundamental insights into LSC adhesion. The therapeutic potential of anti-JAM-C antibodies has never been tested. This study paves the road for evaluating in depth the mechanism and impact of antagonists, such as blocking antibodies, on the cell adhesion mechanisms responsible for resistance to chemotherapy (Meads et al., 2008).

Dynamic measurements of JAM interactions and enrichment at leuko-stromal contacts

We screened BM-derived cell lines for JAM expression. We selected the murine stromal cell line MS5 and the human acute myeloblastic cell line KG1, according to their respective expression of JAM-B or JAM-C. Noteworthy, the high-affinity JAM-B–JAM-C interaction occurs across species (i.e. human JAM-C interacts with mouse JAM-B). A subpopulation of the AML cell line KG1 expresses JAM-C and the stromal cell line MS5 homogeneously expresses JAM-B. Importantly, high JAM-C level in AML is associated with poor prognosis, characterized by an increased risk of relapse (De Grandis et al., 2017). Since the parental KG1 population expresses JAM-C at low level, within a stem cell subpopulation, KG1 cells were regularly sorted from the parental population, based on the stem cell criterion (Al-Mawali et al., 2016; Vergez et al., 2011) CD45+CD34+CD38low/−CD123+JAM-C+ as described previously (De Grandis et al., 2017). As assessed by flow cytometry, and in agreement with our previous findings (Arcangeli et al., 2011; De Grandis et al., 2015, 2017), MS5 cells express JAM-B and not JAM-C, while sorted KG1 cells have a high level of JAM-C expression and are negative for JAM-B expression (Fig. S1).

We next performed SMT experiments to study the dynamic aspects of the leuko-stromal interaction mediated by JAM-C on parental KG1 cells and JAM-B on MS5 cells (Fig. 1A). To this aim, we developed an experimental study system based on a short-term coculture of both cell lines. Live cells were separately stained with immunofluorescent antibodies at the SM level. KG1 cells were next seeded on MS5 cells that were previously adhered to glass-bottom Petri dishes, prior to proceeding to video microscopy. MTT2col analysis allowed reconstruction of JAM trajectories, then generation of (1) maps of trajectories and colocalizations, (2) z-profiles corresponding to the axial distribution of SM, (3) descriptors of trajectory confinement or directionality, and (4) sorting according to the anomalous diffusion exponent.

Fig. 1.

Monitoring JAM-B and JAM-C dynamics at MS5–KG1 contacts by MTT2col. (A) Schematic overview of SMT. MS5 cells spread in glass-bottom Petri dishes were labeled with primary and secondary antibodies at SM density. KG1 cells were immunolabeled separately and inoculated in co-culture on MS5 cells (left) to initiate leuko-stromal contacts and then videonanoscopy acquisitions were undertaken (middle). Cellular contacts (in) were determined by MTT2col from JAM-C positions in the focal plane. The outside of cell contacts (out) was defined for JAM-B as the rest of the field of view on MS5 cells and for JAM-C by focusing at the top of KG1 cells (dashed blue rectangles). MTT2col generated trajectories and sets of dedicated analyses (right). (B) Maximum projection of a 500-frame videonanoscopy acquisition, showing JAM-B and JAM-C positions over time (left). Maps of JAM-B and JAM-C trajectories represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells (right). Inserts show magnifications of the framed areas. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. Several concentric circles correspond to successive colocalizations at a nearby locations but with different durations. (C) Images from the same videonanoscopy acquisition corresponding to the area framed in B, with colocalization events or not (white circle or green/magenta arrowheads, respectively). (D) Trajectory density of JAM-C and JAM-B inside and outside cellular contacts. Data are presented as scatter clouds, with mean±s.e.m. for each video. (E) Distribution of diffusion coefficient D for JAM-C and JAM-B inside and outside cellular contacts. Data from all trajectories are presented as histograms. Note that the recruitment of JAM-B and JAM-C at MS5–KG1 cell contacts is associated with reduced JAM-C diffusion and increased JAM-B diffusion. *P<0.05; ***P<0.001 (Wilcoxon test). n=201 JAM-C trajectories and 281 JAM-B trajectories at contacts, 80 JAM-C trajectories and 7453 JAM-B trajectories out of contacts, from 56 videos (seven independent experiments).

Fig. 1.

Monitoring JAM-B and JAM-C dynamics at MS5–KG1 contacts by MTT2col. (A) Schematic overview of SMT. MS5 cells spread in glass-bottom Petri dishes were labeled with primary and secondary antibodies at SM density. KG1 cells were immunolabeled separately and inoculated in co-culture on MS5 cells (left) to initiate leuko-stromal contacts and then videonanoscopy acquisitions were undertaken (middle). Cellular contacts (in) were determined by MTT2col from JAM-C positions in the focal plane. The outside of cell contacts (out) was defined for JAM-B as the rest of the field of view on MS5 cells and for JAM-C by focusing at the top of KG1 cells (dashed blue rectangles). MTT2col generated trajectories and sets of dedicated analyses (right). (B) Maximum projection of a 500-frame videonanoscopy acquisition, showing JAM-B and JAM-C positions over time (left). Maps of JAM-B and JAM-C trajectories represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells (right). Inserts show magnifications of the framed areas. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. Several concentric circles correspond to successive colocalizations at a nearby locations but with different durations. (C) Images from the same videonanoscopy acquisition corresponding to the area framed in B, with colocalization events or not (white circle or green/magenta arrowheads, respectively). (D) Trajectory density of JAM-C and JAM-B inside and outside cellular contacts. Data are presented as scatter clouds, with mean±s.e.m. for each video. (E) Distribution of diffusion coefficient D for JAM-C and JAM-B inside and outside cellular contacts. Data from all trajectories are presented as histograms. Note that the recruitment of JAM-B and JAM-C at MS5–KG1 cell contacts is associated with reduced JAM-C diffusion and increased JAM-B diffusion. *P<0.05; ***P<0.001 (Wilcoxon test). n=201 JAM-C trajectories and 281 JAM-B trajectories at contacts, 80 JAM-C trajectories and 7453 JAM-B trajectories out of contacts, from 56 videos (seven independent experiments).

To analyze dual-color data by MTT, we first aligned the green and magenta channel images (Fig. S2A). Correctly aligning several channels in multicolor microscopy is always challenging. Indeed, chromatic aberrations and mechanical positioning of the cameras and optical elements can hardly warrant nanometer accuracy. However, rigorous alignment can be efficiently completed by post-acquisition registration. This is critical for assessing colocalization at sub-pixel super-resolution (Fig. S2B,C). Transient interactions were defined as spatiotemporal proximity between JAM-B and JAM-C, as assessed by colocalization (Schnitzbauer et al., 2018; Triller and Choquet, 2008) for distances below 320 nm, close to the Rayleigh criterion, which gives a resolution limit of 257 nm for maximum emission at 616 nm (Fig. 1B,C). The mean value of colocalization occurrences and durations, tested over a range of colocalization threshold values, from 80 to 640 nm (0.5 to 4 pixels), reaches a plateau after the retained value, 320 nm (2 pixels, depicted by gray dashed lines in Fig. S2D). This plateau indicates that this value provides a reliable threshold, since below that value some colocalization events are missed, while above it, most are detected. The fact that this value is greater than the size of most molecules (typically ∼10 nm) and our SMT lateral resolution (∼60 nm) could indicate that some colocalization events correspond to indirect interactions, maybe through intermediate molecules, such as sub-membrane scaffolds, or by residing within aggregates or mesoscale structures, such as lipid rafts or the actin sub-membrane meshwork (Kusumi et al., 2012).

Proper evaluation of colocalization by SMT results from a compromise; the labeling level should be low enough to allow SM detection, but altogether high enough to allow observation of colocalization events. For a typical cell radius of ∼10 µm (for KG1 cells, the radius is 7.4±1.3 µm, mean±s.d.), the area of the membrane is ∼1000 µm2. A homogeneous repartition would lead to a surface density of ∼3–70 molecules/µm2. However, a partial but consequent concentration of the molecules will occur when they are at adhesion sites, such as those of the adhesion belt, focal adhesions or leuko-stromal contacts. These adhesion sites, typically representing ∼10% of the plasma membrane surface, would lead to a membrane surface area of ∼100 µm2 over which the molecules could distribute, and therefore to a surface density of ∼30–700 molecules/µm2. For SMT, the density should be low enough to ensure that SM signals (corresponding to the point-spread function, with 160-nm diameter and 0.2-µm2 area) rarely overlap. This leads to a maximum density of five labeled molecules per µm2 (see Mathematical Appendix 3 in Sergé et al., 2008). We achieve ∼0.13 labeled JAM-B and ∼0.3 labeled JAM-C molecules per µm2, and about ten times more at contacts. Hence, our level of labeling remains below the ability of SMT to statistically discriminate SMs.

By focusing on MS5–KG1 contacts, we observed JAM-B SM staining at the upper membrane of MS5 cells and JAM-C SM staining at the bottom of KG1 cells contacting MS5 cells. Hence, JAM-C was only visible at cell contacts while JAM-B was visible within and outside contacts, given that stromal cells are relatively flat (Fig. 1A,B). Analysis of JAM-C trajectories out of contacts needed focusing to the upper plane at the top of KG1 cells (Fig. 1A). Trajectory densities and diffusion coefficients D were compared at (in) and outside (out) contacts. It is noteworthy that KG1 motion was negligible, at least one order of magnitude slower, as compared to JAM motion. We found increased trajectory densities at cell contacts, revealing local recruitment of both proteins (Fig. 1C,D). This enrichment was associated with a decrease in JAM-C diffusion (median 0.020 µm2/s at contacts and 0.029 µm2/s out of contacts; Fig. 1E). This resulted from the appearance of a slower fraction on the histogram of D, at ∼0.002 µm2/s. By contrast, JAM-B diffusion was lower overall than for JAM-C but slightly increased at contacts (0.0078 µm2/s versus 0.0062 µm2/s out of contacts). Taken together, this suggests that JAM interactions constitute a driving element in establishing myeloblastic contacts with stromal cells.

JAM-Chigh LSCs preferentially interact with stromal cells

We further investigated the impact of the JAM-C expression level by leukemic cells. To this aim, we compared the membrane dynamics and interactions of JAMs at contacts between MS5 stromal cells and KG1 leukemic cells with a low (Movie 1) and high (Movie 2) JAM-C expression level (Fig. 2A). This distinction was made by sorting our data according to the number of JAM-C trajectories obtained by MTT. Fitting the distribution of trajectory numbers by a bimodal Gaussian distribution allowed the discrimination of JAM-Clow and JAM-Chigh subpopulations within KG1 cells (Fig. 2B). Videonanoscopy analysis of JAM recruitment to the leuko-stromal interface revealed qualitative and quantitative differences in JAM trajectories at cell contacts with respect to the JAM-C level in KG1 cells (Fig. 2C; Fig. S3A). Using low numbers of KG1 cells allowed the measurement of one cell per field of view and to subsequently separate data concerning JAM-Clow and JAM-Chigh cells. Trajectory numbers directly reflect JAM-C expression level at the membrane. Leuko-stromal contacts involving JAM-Chigh KG1 cells showed recruitment of JAM-C and, to a lower extent, of JAM-B (Fig. 2D). Quantification of the interaction kinetics revealed similar colocalization frequencies and durations for low and high levels of JAM-C (Fig. 2E), both of which were higher than fortuitous colocalizations that occurred in simulated random walks (Fig. S3B). It is noteworthy that colocalization events will be underestimated, since only a fraction of JAMs are labeled (Fig. S3C). JAM-B–JAM-C interactions are thus favored, occurring more often than just by random while having little or even no sensitivity to JAM expression level.

Fig. 2.

Monitoring JAM-B–JAM-C interactions for different JAM-C expression levels. (A) Schemes of cellular contacts for an MS5 cell and a KG1 cell expressing JAM-C at low (left) or high (right) level. (B) The histogram of the number of JAM-C trajectories detected on parental KG1 cells was adjusted with a weighted sum of two Gaussian distributions using the least-squares method, allowing the identification of cells expressing JAM-C at high and low level, respectively. n=56 KG1 cells (seven independent experiments). (C) Maps of trajectories of JAM-B on MS5 cells and of JAM-C, expressed at low (left) or high (right) level, on KG1 cells, with magnifications of the framed areas. Trajectories are represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. (D–G) JAM-C (magenta) and JAM-B (in, green, or out of cell contact, gray) dynamics for KG1 cells with low or high JAM-C expression level. Density of trajectories (D), occurrence and duration of colocalization episodes (E), MSD (F) and distribution of D (G) are shown. Data are presented as scatter clouds for each video, with mean±s.e.m. (D,E) or as histograms for all trajectories (G). *P<0.05; ***P<0.001 (Wilcoxon test). n=2689 JAM-C trajectories and 149 JAM-B trajectories for JAM-Clow KG1 cells, 3014 JAM-C trajectories and 117 JAM-B trajectories for JAM-Chigh KG1 cells, from 56 videos (seven independent experiments). (H) Scheme and maximal intensity y-projection of xz-side views from confocal z-stacks illustrating the location of JAM-B and JAM-C at the level of contacts between leukemic KG1 cells and stromal MS5 cells. Dashed lines correspond to cell edges. (I) Quantification of the axial positions (mean number by z-plane, normalized to the maximum value, with s.e.m. shown as shaded areas) of JAM-B (green) and JAM-C (magenta). Measurements with KG1 cells expressing JAM-C at low or high level (light or dark colors, respectively. n=22 and eight videos respectively, from seven independent experiments). The blue dash line represents the theoretical distribution expected for JAM-C uniformly distributed at the plasma membrane and in the cytosol of KG1 cells, modelized by spheres.

Fig. 2.

Monitoring JAM-B–JAM-C interactions for different JAM-C expression levels. (A) Schemes of cellular contacts for an MS5 cell and a KG1 cell expressing JAM-C at low (left) or high (right) level. (B) The histogram of the number of JAM-C trajectories detected on parental KG1 cells was adjusted with a weighted sum of two Gaussian distributions using the least-squares method, allowing the identification of cells expressing JAM-C at high and low level, respectively. n=56 KG1 cells (seven independent experiments). (C) Maps of trajectories of JAM-B on MS5 cells and of JAM-C, expressed at low (left) or high (right) level, on KG1 cells, with magnifications of the framed areas. Trajectories are represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. (D–G) JAM-C (magenta) and JAM-B (in, green, or out of cell contact, gray) dynamics for KG1 cells with low or high JAM-C expression level. Density of trajectories (D), occurrence and duration of colocalization episodes (E), MSD (F) and distribution of D (G) are shown. Data are presented as scatter clouds for each video, with mean±s.e.m. (D,E) or as histograms for all trajectories (G). *P<0.05; ***P<0.001 (Wilcoxon test). n=2689 JAM-C trajectories and 149 JAM-B trajectories for JAM-Clow KG1 cells, 3014 JAM-C trajectories and 117 JAM-B trajectories for JAM-Chigh KG1 cells, from 56 videos (seven independent experiments). (H) Scheme and maximal intensity y-projection of xz-side views from confocal z-stacks illustrating the location of JAM-B and JAM-C at the level of contacts between leukemic KG1 cells and stromal MS5 cells. Dashed lines correspond to cell edges. (I) Quantification of the axial positions (mean number by z-plane, normalized to the maximum value, with s.e.m. shown as shaded areas) of JAM-B (green) and JAM-C (magenta). Measurements with KG1 cells expressing JAM-C at low or high level (light or dark colors, respectively. n=22 and eight videos respectively, from seven independent experiments). The blue dash line represents the theoretical distribution expected for JAM-C uniformly distributed at the plasma membrane and in the cytosol of KG1 cells, modelized by spheres.

Molecules mostly undergo Brownian motion, which can be characterized by computing their mean square displacement (MSD) from experimental trajectories. This allows the diffusion coefficient D of each molecule to be determined. MSD and D were lower for JAM-C at cell contacts that had a high JAM-C level (Fig. 2F,G). D decreased from 0.012 µm2/s to 0.0067 µm2/s, as assessed by fitting the average MSD curve (Fig. 2G, arrow). At contacts involving a high JAM-C level, JAM-B motion was overall similar with respective to JAM-C level, 0.0021 µm2/s versus 0.0028 µm2/s, but with a wider distribution, indicating modified dynamics. JAM-B diffusion out of contacts provided a reference of 0.0029 µm2/s versus 0.0028 µm2/s. This emphasizes that JAM-C is key in recruiting JAM-B for establishing leuko-stromal contacts.

Axial profile at SM resolution reveals JAM enrichment at cell contacts

Recruitment of JAM-B and JAM-C at leuko-stromal contacts is difficult to detect by classical immunofluorescence labeling on fixed co-cultures. The phenomenon is not as massive as that of patching and capping of specific receptors at neuronal or immune synapses, or at focal adhesions (Sergé, 2016). Protein accumulation is more gradual in the current case. However, it can be detected and quantified by taking advantage of the sensitivity provided by SM detection. Indeed, the MTT algorithm, based on hypothesis testing, efficiently differentiates between SM signals and background noise, partly generated by autofluorescence. We quantified the number of JAM molecules on z-stacks as a function of the axial position z (Fig. 2H). This analysis shows a preferred position of JAM-B at and below contacts, as expected given the relatively flat geometry of MS5 stromal cells. The axial distribution of JAM-B is mostly restricted to around the basal plane, where MS5 cells are in contact with the glass slide. The width of the distribution is ∼3 µm. This is consistent with the average size of cellular extensions, convoluted by the axial resolution. Indeed, the axial distribution of quantum-dots deposited on a coverslip was fitted with a Gaussian profile (R2=0.998) leading to an axial resolution equal to the full width at half maximum (FWHM) of 1.3 µm (Fig. S4A,B).

For KG1 leukemic cells, which are relatively round, we computed the theoretically expected profile for a homogeneous distribution of a molecule at the surface (plasma membrane) and/or in the volume (i.e. undergoing vesicular trafficking) (Fig. S4C–F). We indeed visually observed a gradual internalization of JAM molecules over the timecourse of measurements. This was mostly visible for JAM-C, due to the round shape of the leukemic KG1 cells, as opposed to the flat shape of the stromal MS5 cells, which limited visualizing internalization. The axial distribution of JAM-C molecules, obtained by acquisition and quantification of z-stacks by MTT2col, exhibited a profile that was substantially biased as compared to the theoretical distribution on a sphere, mostly by being larger at low focal positions, corresponding to the bottom of KG1 cells. This reveals an enrichment of JAM-C at the z-position of the contact, which was more pronounced for high JAM-C expression, further supporting increased JAM recruitment at contacts (Fig. 2I; Fig. S4G, Movie 3).

Using a blocking anti-JAM-C antibody perturbs LSC contact with stromal cells

Since our results indicate that JAM interactions represent a driving force in establishing LSC contacts with the stroma, we evaluated the potential of targeting JAM-C to destabilize LSCs from their niches. We performed SMT with cells treated with a blocking anti-JAM-C antibody, expecting to hamper JAM-B–JAM-C interactions. We also performed measurements with cells treated with an unlabeled blocking anti-JAM-C antibody in competition with fluorescently labeled non-blocking anti-JAM-C antibody. We therefore compared JAM dynamics using three strategies: with non-blocking anti-JAM-C antibodies (Nbl; Movie 4, as used for Figs 1 and 2), with blocking anti-JAM-C antibodies (Bl; Movie 5), or with both antibodies, called in competition (Cp) (Fig. 3A). We used high KG1 cell numbers in order to increase data amount per video. Densities of JAM-C trajectories measured at cell contacts were comparable between all experimental conditions. However, densities of JAM-B trajectories at cell contacts and colocalization durations were drastically reduced when using the blocking antibody, either alone or in competition, with colocalization frequency reducing to levels comparable to the values obtained when simulating two non-interacting molecular populations (Fig. S3B). It is noteworthy that the JAM-B density out of cell contacts was unaffected (Fig. 3B,C; Fig. S3D). Remarkably, when using a non-blocking antibody, the distribution of the nearest distances from JAM-B to JAM-C exhibited a bimodal distribution, with substantial values below the cutoff set for assessing colocalization. This distribution was almost abolished with the blocking antibody. Furthermore, distances among JAMs were overall increased in the presence of the blocking antibody, reflecting sparser localizations within cell membranes (Fig. 3E, arrows). D decreased from 0.0097 µm2/s to 0.0065 µm2/s for JAM-C and from 0.0053 µm2/s to 0.0028 µm2/s for JAM-B at contacts, remaining at 0.0029 µm2/s or 0.0028 µm2/s out of contacts (Fig. 3F,G, arrows). The competition assay overall led to similar results. JAM-C enrichment at contacts was abrogated when using a blocking anti-JAM-C antibody, with an axial profile closer to the theoretical one (Fig. 3H,I). Altogether, these data show that a blocking anti-JAM-C antibody hampers JAM recruitment at cell contacts.

Fig. 3.

Blocking anti JAM-C antibody reduces LSC/stroma interactions. (A) Schematic overview of SM labeling with non-blocking (in-house antibody 19H36, left), blocking (R&D clone 208206, middle) or both anti-JAM-C antibodies, in competition (right). (B) Maps of JAM-B and JAM-C trajectories, using non-blocking (left), blocking (middle) or both, in competition (right), anti-JAM-C antibodies. Trajectories are represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. (C–G) JAM-C (magenta) and JAM-B (in, green, or out of cell contact, gray) dynamics with non-blocking anti-JAM-C antibody (Nbl), blocking anti-JAM-C antibody (Bl), or both, in competition (Cp). The density of JAM-C and JAM-B trajectories (C), occurrence and duration of colocalization episodes (D), distance from JAM-B to nearest JAM-C, allowing the identification of colocalizations (E), MSD (F) and distribution of D (G). *P<0.05, ***P<0.001 (Wilcoxon test). n=5683 JAM-C trajectories and 15,513 JAM-B trajectories from 74 videos (five independent experiments) for non-blocking anti-JAM-C antibody, 4513 JAM-C trajectories and 6892 JAM-B trajectories from 51 videos (four independent experiments) for blocking anti-JAM-C antibody, 7365 JAM-C trajectories and 1405 JAM-B trajectories from 34 videos for non-blocking anti-JAM-C antibody, as control for competition, 5120 JAM-C trajectories and 744 JAM-B trajectories from 23 videos for competition. (H) Scheme and maximal intensity y-projection of xz-side views from confocal z-stacks illustrating the location of JAM-B and JAM-C at the level of contacts between leukemic KG1 cells and stromal MS5 cells. Dashed lines correspond to cell edges. (I) Quantification of the axial positions (mean number by z-plane, normalized to the maximum value, with s.e.m. shown as shaded areas) of JAM-B (green) and JAM-C (magenta). Measurements using either a blocking (n=39 videos from five independent experiments) or non-blocking (n=75 videos, from six independent experiments) anti-JAM-C antibody (light or dark colors, respectively). The blue dashed line represents the theoretical distribution expected for a uniform distribution at the membrane and in the cytosol of KG1 cells.

Fig. 3.

Blocking anti JAM-C antibody reduces LSC/stroma interactions. (A) Schematic overview of SM labeling with non-blocking (in-house antibody 19H36, left), blocking (R&D clone 208206, middle) or both anti-JAM-C antibodies, in competition (right). (B) Maps of JAM-B and JAM-C trajectories, using non-blocking (left), blocking (middle) or both, in competition (right), anti-JAM-C antibodies. Trajectories are represented by gradients of green and magenta, respectively, according to time, and superimposed on the transmission image of the cells. Spatiotemporal colocalizations are denoted by white circles with a size proportional to duration. (C–G) JAM-C (magenta) and JAM-B (in, green, or out of cell contact, gray) dynamics with non-blocking anti-JAM-C antibody (Nbl), blocking anti-JAM-C antibody (Bl), or both, in competition (Cp). The density of JAM-C and JAM-B trajectories (C), occurrence and duration of colocalization episodes (D), distance from JAM-B to nearest JAM-C, allowing the identification of colocalizations (E), MSD (F) and distribution of D (G). *P<0.05, ***P<0.001 (Wilcoxon test). n=5683 JAM-C trajectories and 15,513 JAM-B trajectories from 74 videos (five independent experiments) for non-blocking anti-JAM-C antibody, 4513 JAM-C trajectories and 6892 JAM-B trajectories from 51 videos (four independent experiments) for blocking anti-JAM-C antibody, 7365 JAM-C trajectories and 1405 JAM-B trajectories from 34 videos for non-blocking anti-JAM-C antibody, as control for competition, 5120 JAM-C trajectories and 744 JAM-B trajectories from 23 videos for competition. (H) Scheme and maximal intensity y-projection of xz-side views from confocal z-stacks illustrating the location of JAM-B and JAM-C at the level of contacts between leukemic KG1 cells and stromal MS5 cells. Dashed lines correspond to cell edges. (I) Quantification of the axial positions (mean number by z-plane, normalized to the maximum value, with s.e.m. shown as shaded areas) of JAM-B (green) and JAM-C (magenta). Measurements using either a blocking (n=39 videos from five independent experiments) or non-blocking (n=75 videos, from six independent experiments) anti-JAM-C antibody (light or dark colors, respectively). The blue dashed line represents the theoretical distribution expected for a uniform distribution at the membrane and in the cytosol of KG1 cells.

JAM trajectories are affected by JAM-C level and blocking anti-JAM-C antibody

Distinct types of movement can be characterized and discriminated by using a geometrical descriptor of trajectory linearity, L (Fig. 4A), and the potential non-linearity of the MSD (proportional to time for a purely Brownian motion). The experimental MSD versus time curve was fitted with the theoretical function for anomalous diffusion, 4Dγtγ+2σ2, that may imply free diffusion, directionality or confinement (Saxton, 1994). This allowed computing three parameters for each trajectory: the generalized diffusion Dγ, the anomalous exponent γ and the location accuracy σ (Fig. 4B). γ is lower and higher than 1 for confined and directed motion, respectively. Descriptors were computed for Monte Carlo simulations and experiments (Fig. 4C). The distribution of descriptors assessed from experimental measurements was compared with those provided by simulations established for a known diffusion D, velocity v or confinement radius R. This allowed us to infer whether the experimental values underlying the measured trajectories arose from confined, Brownian or directed motion (Fig. 4D). We overall found lower values for each descriptor, L and γ, for KG1 cells expressing high levels of JAM-C (Fig. 4E) with a minor effect out of contacts (lower Cohen's coefficient, although this was significant for some conditions, due to a larger number of trajectories). More confined motion can be partly related to more frequent and longer interactions among JAM-C and JAM-B. Note that we similarly obtained lower values for L and γ when using a blocking anti-JAM-C antibody (Fig. 4F).

Fig. 4.

Characterization of simulated and experimental trajectories. (A) Trajectories have been characterized by their linearity L, a geometric parameter measuring the deviation from a line. The descriptor L varies from zero for confined motion to one for directed motion. (B) SM motion was also characterized by the curvature of the MSD, which is linear for pure diffusion, bent upward for directed motion and downward for confinement. This was quantified by an anomalous exponent γ associated with the expression MSD=4Dtγ+2σ2. The descriptor γ is equal to, greater than and less than one for Brownian, directed and confined motion, respectively. (C) Sample trajectories (grouped to a common origin and separated examples) obtained by Monte Carlo simulations (left) generated with MATLAB for Brownian motion (purple), confined motion (purple to black) or with an added directed component (purple to blue). Sample trajectories of JAM-C (magenta) and JAM-B (green) for experimental measurements with high (dark colors) and low (light colors) levels of JAM-C expression on KG1 (middle). Sample trajectories of JAM-C (magenta) and JAM-B (green) for experimental measurements with blocking (dark colors) or non-blocking (light colors) anti-JAM-C antibody (right). (D–F) Violin plots, with mean±s.d. highlighted, and Cohen's d coefficient depicting the variation of each descriptor with the type of motion, from confined to directed simulations (D), for JAM-B and JAM-C in the context of KG1 cells expressing JAM-C at high or low level (E) and using a blocking (Bl) or non-blocking (Nbl) anti-JAM-C antibody (F). *P<0.05; **P<0.01; ***P<0.001 (Wilcoxon test). n=2689 JAM-C trajectories and 149 JAM-B trajectories for JAM-Clow KG1 cells; 3014 JAM-C trajectories and 117 JAM-B trajectories for JAM-Chigh KG1 cells; 5683 JAM-C trajectories and 15,513 JAM-B trajectories for non-blocking anti-JAM-C antibody; 4513 JAM-C trajectories and 6892 JAM-B trajectories for blocking anti-JAM-C antibody.

Fig. 4.

Characterization of simulated and experimental trajectories. (A) Trajectories have been characterized by their linearity L, a geometric parameter measuring the deviation from a line. The descriptor L varies from zero for confined motion to one for directed motion. (B) SM motion was also characterized by the curvature of the MSD, which is linear for pure diffusion, bent upward for directed motion and downward for confinement. This was quantified by an anomalous exponent γ associated with the expression MSD=4Dtγ+2σ2. The descriptor γ is equal to, greater than and less than one for Brownian, directed and confined motion, respectively. (C) Sample trajectories (grouped to a common origin and separated examples) obtained by Monte Carlo simulations (left) generated with MATLAB for Brownian motion (purple), confined motion (purple to black) or with an added directed component (purple to blue). Sample trajectories of JAM-C (magenta) and JAM-B (green) for experimental measurements with high (dark colors) and low (light colors) levels of JAM-C expression on KG1 (middle). Sample trajectories of JAM-C (magenta) and JAM-B (green) for experimental measurements with blocking (dark colors) or non-blocking (light colors) anti-JAM-C antibody (right). (D–F) Violin plots, with mean±s.d. highlighted, and Cohen's d coefficient depicting the variation of each descriptor with the type of motion, from confined to directed simulations (D), for JAM-B and JAM-C in the context of KG1 cells expressing JAM-C at high or low level (E) and using a blocking (Bl) or non-blocking (Nbl) anti-JAM-C antibody (F). *P<0.05; **P<0.01; ***P<0.001 (Wilcoxon test). n=2689 JAM-C trajectories and 149 JAM-B trajectories for JAM-Clow KG1 cells; 3014 JAM-C trajectories and 117 JAM-B trajectories for JAM-Chigh KG1 cells; 5683 JAM-C trajectories and 15,513 JAM-B trajectories for non-blocking anti-JAM-C antibody; 4513 JAM-C trajectories and 6892 JAM-B trajectories for blocking anti-JAM-C antibody.

Anomalous diffusion allows classification of JAM dynamics

Plotting trajectories according to Dγ versus γ allows the visualization and classification of molecular dynamics. JAM trajectories were sorted into three categories: those with low, intermediate and high γ values, which are associated with confined, Brownian and linear motion, respectively (Fig. 5A). The three categories were separated by thresholds determined from the dispersion of γ assessed for Brownian simulations (Fig. 5B). A low or high expression level of JAM-C had little impact on this classification (Fig. S5). However, using a blocking anti-JAM-C antibody led to a significant decrease in the proportion of trajectories assessed as Brownian for JAM-C and JAM-B, with an increase in the proportion assessed as confined for JAM-B (Fig. 5C) in agreement with Fig. 4F. JAM interactions may be attributed to a subset of confined (such as upon receptor aggregating and/or binding to scaffold partners) or directed motion (such as through receptor-mediated internalization upon activation). The blocking antibody that prevents JAM-B–JAM-C binding, could trigger JAM-B binding to other partners, that are yet to be identified, resulting in increased confinement. Our results further advocate that JAM dynamics and interactions are partly reversed by using a blocking anti-JAM-C antibody. Overall, when using a blocking antibody, JAM dynamics become analogous to those seen in a leukemic cell with a low JAM-C level, as expected for a non-stem cell.

Fig. 5.

JAM dynamics classification according to anomalous diffusion and blocking anti JAM-C antibody. (A) Trajectories can be sorted according to their anomalous diffusion exponent. Fitting the MSD to the expression for anomalous diffusion, MSD=4Dγtγ+2σ2, allows the determination of the generalized diffusion coefficient Dγ (µm2/s), anomalous exponent γ and accuracy σ (µm). Trajectory diversity can be visualized as a dot plot of Dγ versus γ. A trajectory with γ≪1, ∼1 or ≫1 is classified as confined, Brownian and linear, respectively. (B) Trajectories obtained by Monte Carlo simulations of directed (blue), Brownian (purple) or confined (black) motion were visualized as a dot plot of Dγ versus γ for comparison with experimental trajectories and visualization of the boundaries among the three modes of motion, set from the dispersion of γ for Brownian simulations. (C) Percentage of trajectories of JAM-C and JAM-B classified as confined, Brownian or directed when using a non-blocking or blocking anti-JAM-C antibody. Error bars are mean±s.e.m. *P<0.05; ***P<0.001 (Wilcoxon test). n=5683 JAM-C trajectories and 15,513 JAM-B trajectories from 74 videos (five independent experiments) for non-blocking anti-JAM-C antibody; 4513 JAM-C trajectories and 6892 JAM-B trajectories from 51 videos (four independent experiments) for blocking anti-JAM-C antibody.

Fig. 5.

JAM dynamics classification according to anomalous diffusion and blocking anti JAM-C antibody. (A) Trajectories can be sorted according to their anomalous diffusion exponent. Fitting the MSD to the expression for anomalous diffusion, MSD=4Dγtγ+2σ2, allows the determination of the generalized diffusion coefficient Dγ (µm2/s), anomalous exponent γ and accuracy σ (µm). Trajectory diversity can be visualized as a dot plot of Dγ versus γ. A trajectory with γ≪1, ∼1 or ≫1 is classified as confined, Brownian and linear, respectively. (B) Trajectories obtained by Monte Carlo simulations of directed (blue), Brownian (purple) or confined (black) motion were visualized as a dot plot of Dγ versus γ for comparison with experimental trajectories and visualization of the boundaries among the three modes of motion, set from the dispersion of γ for Brownian simulations. (C) Percentage of trajectories of JAM-C and JAM-B classified as confined, Brownian or directed when using a non-blocking or blocking anti-JAM-C antibody. Error bars are mean±s.e.m. *P<0.05; ***P<0.001 (Wilcoxon test). n=5683 JAM-C trajectories and 15,513 JAM-B trajectories from 74 videos (five independent experiments) for non-blocking anti-JAM-C antibody; 4513 JAM-C trajectories and 6892 JAM-B trajectories from 51 videos (four independent experiments) for blocking anti-JAM-C antibody.

Our MTT2col method is based on dual-color SMT, coupled to innovative and extended trajectory analyses. We characterized the dynamics of two binding partners, JAM-B and JAM-C, which are expressed by stromal and leukemic cells, respectively. Both are preferentially recruited at leuko-stromal contacts, together with reinforced interactions, as assessed by increased SM colocalization frequency and duration (Fig. 6A). At cell contacts, diffusion was increased for JAM-B, but reduced for JAM-C. Out of contacts, JAM-B movement was slower than JAM-C, possibly reflecting interactions (in cis or trans) with intra- or extra-cellular molecules, such as the cytoskeleton and scaffolds, possibly through the JAM PDZ-binding motif (Kusumi et al., 2012), or signaling partners such as integrins, which have been reported to interact with JAMs (Kummer and Ebnet, 2018). This suggests an increased membrane organization for the stromal cell as compared to the LSC, where JAM-C would be mostly freely diffusing. LSCs and stromal cells, respectively, express several binding partners, notably VLA-4 (ITGA4)–VCAM-1 and CD44–E-selectin, as well as signaling molecules, like CXCR4–CCL12 or HIF-1α (Konopleva et al., 2009; Rashidi and DiPersio, 2016). The differential effect observed on diffusion at contacts could be explained by JAM-B being destabilized by disengaging from its partners upon competition with JAM-C, or, by contrast, by being stabilized by new interactions and/or friction with JAM-B and other putative partners. Changes in dynamics at contacts may furthermore reflect local modifications of membrane composition and fluidity. Interestingly, the invariance of JAM-B–JAM-C interactions regarding JAM-C level could guarantee a robust commitment to adhesion and signaling mechanisms, hence remaining efficient at low JAM level.

Fig. 6.

Overview of JAM dynamics. (A) Scheme of JAM dynamics at leuko-stromal contact, for an LSC expressing JAM-C at high level (left) or for a differentiated leukemic cell either expressing JAM-C at low level or treated with a blocking anti-JAM-C antibody, mimicking a reduction of JAM-C level (right). (B) Scheme of the evolution of JAM dynamics upon leuko-stromal contact, leading either to JAM-B and JAM-C stabilization (reduced diffusion; top right) or JAM-B destabilization (increased diffusion) and JAM-C stabilization (reduced diffusion) (bottom right) or a combination of both, through modulated interactions with structural elements such as scaffold proteins, cytoskeleton and lipid rafts (see legend).

Fig. 6.

Overview of JAM dynamics. (A) Scheme of JAM dynamics at leuko-stromal contact, for an LSC expressing JAM-C at high level (left) or for a differentiated leukemic cell either expressing JAM-C at low level or treated with a blocking anti-JAM-C antibody, mimicking a reduction of JAM-C level (right). (B) Scheme of the evolution of JAM dynamics upon leuko-stromal contact, leading either to JAM-B and JAM-C stabilization (reduced diffusion; top right) or JAM-B destabilization (increased diffusion) and JAM-C stabilization (reduced diffusion) (bottom right) or a combination of both, through modulated interactions with structural elements such as scaffold proteins, cytoskeleton and lipid rafts (see legend).

Using a blocking anti-JAM-C antibody, either alone or in competition with a non-blocking anti-JAM-C antibody, resulted in reduced interactions, proving that changes in the dynamic behaviors of JAM-B and JAM-C are related to direct/indirect interactions between these two proteins. This may be functionally comparable to reversing the situation from a high to a low JAM-C level, with blocking antibodies leading to a reduced number of JAM-C molecules able to bind to JAM-B, limiting a possible cooperative effect. Blocking anti-JAM-C antibody led to reduced diffusion for JAM-C as for JAM-B (Figs 3F,G and 4F). This result may appear counterintuitive, since a blocking antibody, by limiting JAM-B–JAM-C interactions, would be expected to reduce their interactions, and hence to increase their diffusion. This apparent contradiction could be explained by other direct or indirect effects, as already mentioned, for example, binding to the actin cytoskeleton (Kusumi et al., 2012), or signaling or other adhesion molecules, such as integrins (Kummer and Ebnet, 2018). These putative effects of the blocking antibody, which is apparently able to affect both JAM-C and JAM-B, remain to be investigated. It is noteworthy that transient interactions of integrins with focal adhesions have also been described by SMT (Rossier et al., 2012). However, the reported distribution of D for integrins in and around adhesions was bimodal. This suggests a more stringent stop-and-go mechanism, as compared to the partial reshaping of the distribution of D reported here for JAMs. Noteworthy, in neurons, unimodal and bimodal distributions of D have been reported for adhesion and synaptic receptors, respectively (Letellier et al., 2018). We can hence hypothesize that the underlying interactions mediating stabilization are more gradual for JAMs than for integrins, at least within the considered cellular contexts. During cell contact establishment, JAM distribution was enriched at contacts. This is reminiscent, but in a less stringent way, of what occurs for neuronal or immune synaptic components, which are mostly localized at synapses, and only sparsely outside of them. A reversible diffusion-trap process likely generates this distribution, leading to a gradual enrichment at cell contacts. Such reversibility is expected to sustain a dynamic structure that is susceptible to consequent evolution over time.

Molecular trajectories are essentially dominated by Brownian agitation and can deviate from this basal behavior in two characteristic and well-documented ways, by directed or confined motion (Jacobson et al., 1995). These behaviors can be variably accentuated and/or transient, occurring alternately. Thus, a membrane molecule, which is likely to interact reversibly with partners such as the cytoskeleton, lipid domains or cell contacts, can follow a trajectory showing alternating free and confined motion (Kusumi et al., 2012; Sergé, 2016; Sergé et al., 2002, 2008). Conversely, directed motion is mostly expected to occur from cytosolic vesicles upon labeled SM internalization (Sun et al., 2015) or when molecules are coupled to the actin flow (Sergé et al., 2003). It is noteworthy that our analytical tools can assess each trajectory, taken as a whole, or even the average MSD of a set of trajectories, measured under given experimental conditions.

Advances in optical microscopy allow study of the complexity of cellular mechanisms with ever-increasing resolution, which is now approaching the nanometric dimensions of macromolecules. Alternative labeling with monovalent and smaller probes would be appealing and may be reached by following four alternative strategies: (1) synthetizing and labeling Fab fragments, which may lead to lower affinity, (2) cloning JAMs with a HaloTag, SNAP tag or CLIP tag (Liss et al., 2015), which requires transfected or knock-in cells, (3) cloning single-domain antibody fragments (or nanobodies), which present several key advantages, such as high stability, solubility, affinity, specificity and putative therapeutic interest (Chanier and Chames, 2019), and (4) tagging with a photoactivatable fluorescent protein (Shcherbakova et al., 2014).

Cell adhesion molecules may be expressed at a broad range of copy number, according to the molecule and cell considered. Copy number may vary from 3000 to 70,000, according to values of the literature reported in Burkhart et al., (2012). On average, 130 JAM-B molecules were labeled per cell (five at contacts) (Fig. 1). The total number of labeled JAM-C molecules was evaluated using the z-stack acquisitions, leading to 315 JAM-C molecules per cell (four at contacts) (Fig. 2I). Hence, our cells would have between 0.2 to 4% labeled JAM-B and 0.5 to 11% labeled JAM-C, with a large variability in the estimation both of the labeled and total numbers. SMT thus leads to rather rare colocalization events, due to a low percentage of labeled molecules. Observing 0.1 colocalization events per cell on average (Fig. 2E) leads to 0.08% and 0.03% colocalization (2% and 2.5% at contacts) among labeled JAM-B and JAM-C, respectively. Assuming that (1) the stained fraction is similar for JAM-B and JAM-C, and (2) colocalization concerns either two labeled or two unlabeled molecules, thus neglecting the occurrence of colocalization among a labeled and an unlabeled molecule (simplified vs realistic schemes in Fig. S3C), then the proportion of colocalization events in the total populations would be the same as the one within the labeled populations. The realistic scheme indicates that this percentage is overestimated, to an extend that is hardy evaluable with our measures. Accessing longer colocalization events would allow detecting co-movement, a strong indication of physical interaction.

The variety of transient interactions initially measured in vitro can be validated and specified in cells. Many supramolecular structures often present relatively unexpected dynamics, such as intercellular contacts involving epithelial, cancerous or other cells. SM approaches reveal transient interactions that lead adhesion molecules and their partners to enter and exit contacts, focal adhesions or others. Such structures can thus be created, strengthened, weakened, suppressed or simply maintained by a homeostatic equilibrium of inflows and outflows, depending on the context (Triller, 2011). Our results raise issues about adhesion strengthening mechanisms, like recruitment of intracellular scaffolds, such as the cytoskeleton, or extracellular matrix reorganization. In this regard, we observed a higher diffusion for JAM-C, reducing at contacts, with higher confinement for JAM-B. This could be a signature of a more structured stromal cell membrane and a more dynamic leukemic cell membrane, getting more organized upon contact engagement.

The actin sub-membrane meshwork and lipid rafts participate in the emergence and evolution of most cell adhesions (Kusumi et al., 2012). More particularly, the ability of JAMs to bind to scaffold proteins, such as ZO-1, through their PDZ domain, directly points for a connection with actin, that is susceptible to favor JAM aggregation and stabilization, as extensively studied at tight junctions (Bazzoni et al., 2000; Ebnet et al., 2000). Molecular diffusion often spans a large range of values. Such is the case for JAM-B and JAM-C, with diffusion ranging from 10−4 to 1 µm2/s (Fig. 1E). This large variability is expected to reflect a broad heterogeneity of molecular dynamics, as can be seen on individual trajectories (see Figs 1B and 4C). JAM dynamics may hence vary in opposite, but compatible, ways. Notably for JAM-B, we observe both decreased and increased diffusion values (Figs 1E, 2G and 3G), which could be due to stabilization and destabilization through binding and unbinding to and from structural features, such as the actin meshwork, scaffold proteins and lipid rafts (Fig. 6B). For the whole JAM-B population, a mix of these modifications in dynamics and stabilization may occur, leading to a broader distribution of diffusion values.

JAMs are expected to undergo continuous exocytosis and endocytosis during the time course of 1 h experiments. This turnover is probably regulating a rather constant, homeostatic level of JAMs at the plasma membrane. Our data cannot preclude that a local accumulation of JAM-B and JAM-C at contacts (Fig. 1D) leads to their depletion in the rest of the membrane, which could subsequently trigger a global elevation of JAM number at the entire plasma membrane to maintain such a homeostatic level. Accordingly, the effect of the blocking antibody may be exerted directly at the level of the JAM trans-interaction, but could alternatively result from indirect consequences, notably through the reduction of JAM-C levels due to its internalization. JAM-C endocytosis could be promoted by the blocking antibody, possibly through increased ubiquitylation (Kostelnik et al., 2019). As reported for several surface receptors, this could allow triggering of alternative signaling pathways upon binding to other effectors localized in the endocytic pathway.

Intercellular interactions are associated with numerous signaling mechanisms, frequently leading to reciprocal exchanges (called crosstalk) that benefit the metabolism of both cells. This concept applies to the contact that LSCs establish with the stroma and the matrix constituting their niches. Unfortunately, this contact likely favors LSC resistance to chemotherapeutic treatments, and consequently increases risks of relapse. Adhesion molecules, such as integrins and cadherins, are widely associated with signaling pathways, both inside-out and outside-in. This was first reported for JAM-A (Peddibhotla et al., 2013), and more recently for JAM-B and JAM-C, through interactions with integrins (Kummer and Ebnet, 2018). Signaling is expected to mediate the effects of adhesion on LSC metabolism. The JAM-Chigh subpopulation of LSCs presents clonogenicity and leukemic-initiating activities as well as increased expression of genes associated with the LSC signature, which are all functionally related to leukemic-initiating cells (De Grandis et al., 2017). Using a blocking antibody, we studied in-depth, at the SM level, the targeting of the molecules involved in the LSC–stroma interaction, in which JAMs are key players. Our results indicate that blocking adhesion leads to a molecular dynamic pattern shifting from a JAM-Chigh LSC-related context to a JAM-Clow differentiated-cell-related context. This suggests that using a blocking antibody may promote LSC differentiation, and hence reduced AML aggressiveness. This approach is of undeniable interest to decipher the therapeutic potential of blocking antibodies or the mode of action of known drugs at SM resolution. Such knowledge is ultimately expected to contribute to destabilizing LSCs and making them more vulnerable to improved treatments against leukemia relapse.

Cell culture

The mouse BM stromal cell line MS5 (DSMZ ACC 441) was cultured in Iscove's modified Dulbecco's medium (IMDM) supplemented with 10% fetal calf serum, 1% L-glutamine, penicillin, streptomycin, HEPES, essential amino acids, sodium pyruvate and 25 µM β-mercaptoethanol (all from Gibco). The human myeloblastic cell line KG1 (ATCC CCL-246) was cultured in IMDM supplemented with 10% fetal calf serum, 1% L-glutamine, penicillin and streptomycin. Both cell lines were cultured according to ATCC guidelines and regularly tested for the absence of mycoplasma.

Cell sorting

For all experiments, KG1 cells were sorted from the parental population, on a stem cell criterion: CD45+CD34+CD38low/−CD123+, as previously published (Al-Mawali et al., 2016; Vergez et al., 2011). The sorting was completed on the JAM-C+ surface expression level, as previously published (De Grandis et al., 2017).

Flow cytometry

KG1 and MS5 cell lines were characterized for JAM-B and JAM-C expression. KG1 cells were saturated with IMDM containing 10% normal rabbit serum for 1 h before incubation. Both cell types were separately incubated on ice with either in-house anti-JAM-B polyclonal antibody (829 at 10 nM) or with allophycocyanin (APC)-conjugated anti-JAM-C antibody (R&D, clone 208212 at 10 nM). After washing, cells to be analyzed for JAM-B expression were stained with secondary DyLight 488-conjugated anti-rabbit-IgG secondary antibody.

SMT staining and videonanoscopy

MS5 cells were spread overnight in glass-bottom dishes at 100,000 cells per chamber (Greiner Bio-One). Cells were labeled for 10 min at room temperature with a mix of 10 nM anti-JAM-B 829 and Alexa Fluor 488-conjugated anti-rabbit-IgG (Thermo Fisher Scientific) antibodies in IMDM supplemented with serum. In parallel, 50,000 to 500,000 KG1 cells were labeled for 10 min with a mix of 10 nM anti-JAM-C (blocking: mouse, R&D clone 208206, non-blocking: homemade rat 19H36) and Alexa Fluor 594-conjugated anti-mouse-IgG or anti-rat antibodies (Thermo Fisher Scientific) in supplemented IMDM. For the competition assay, KG1 cells were first incubated with 10 nM blocking anti-JAM-C (mouse, R&D clone 208206) antibody for 30 min. Cells were next labeled for 10 min with a mix of 10 nM non-blocking anti-JAM-C (in-house rat, clone 19H36) and Alexa Fluor 594-conjugated anti-rat-IgG (Thermo Fisher Scientific) antibodies as for control experiments. Secondary antibodies coupled to quantum-dot 655 (Thermo Fisher Scientific) were occasionally used, leading to similar results, as reported previously (Mascalchi et al., 2012). Staining with secondary antibodies alone led to significantly reduced numbers of non-specific trajectory detections, 11.6% for JAM-B and 36.9% for JAM-C (Fig. S1B,C). Cells were washed extensively with HBSS with 1% HEPES and KG1 cells were dropped on MS5 cells before proceeding to imaging.

Video acquisition was performed using an Axio Observer Z1 inverted microscope (Zeiss) equipped with an incubator thermostated at 37°C, a Yokogawa spinning disk device, 491- and 561-nm lasers (Cobolt Calypso 100 mW, 30%) and a 100× α Plan Neofluar NA 1.45 oil-immersion objective. Emission light was split by an LP565 dichroic and collected on two Evolve 512 EMCCD cameras (Photometrics) equipped with 525/50 (green channel) and 641/75 (red channel, shown as magenta) emission filters. The setup was controlled by Metamorph software (Version 7.7.10.0, Molecular Devices). EM gains were set to 800 on the green channel and 120 on the red channel, to compensate for overall differences in signal levels. For each video, 500 frames were taken at 5 MHz with 100 ms exposure time (50 s total length), followed by acquisition of a transmitted light image. For the competition assay, video acquisition was performed using an Axio Observer Z1 inverted microscope (Zeiss) thermostated at 37°C with a 488 nm laser (100 mW, 5%), a 561 nm laser (50 mW, 10%), an AOTF (Gataca System), a multiband filter set (Zeiss 81 HE), a 100× α Plan Apo NA 1.46 oil-immersion objective and a 1.6× optovar. The setup was controlled by Metamorph software (Version 7.10.2.240, Molecular Devices). For each video, 300 frames were taken using a Prime 95B sCMOS camera (Photometrics) at 100 MHz with 100 ms exposure time and 2×2 binning, for each channel in alternation, with 250 ms interframe delay (75 s total length).

We proceeded to acquisitions for no more than 1 h. This maximum duration was estimated to be appropriate for preserving cell physiology as well as reasonable levels of endocytosis and exocytosis. The same protocol was followed for blocking and non-blocking antibodies. Hence, cells were in contact with antibodies for no more than 75 min, including staining and acquisitions. Of note, videomicroscopy acquisitions could be alternatively performed successfully over hours with the same illumination intensity, without noticeable signs of photodamage.

Axial profile of SM distribution over cells and theoretical distribution over a sphere

After each video acquisition, a z-stack was acquired, with 0.2 or 0.5 µm step, starting at a focal position below the MS5 lower membrane, and finishing above KG1 cells. This z-stack was processed by MTT and trajectories with at least three steps were selected. The axial profile was computed for each channel, corresponding to either JAM-B or JAM-C, as the number of SMs detected in each plane. The bottom membrane of MS5 cells, adhering on the coverslip, is almost flat, as compared to the top membrane. This focal plane thus contains a high number of SMs and corresponds to the maximum of the distribution for JAM-B. Hence, to correct for variations of the lower focal position among z-stacks due to manual focusing, the focal position corresponding to the maximum of each JAM-B profile was determined and taken as a reference for the origin of the z-axis for JAM-B as for JAM-C. Profiles were next averaged, smoothed by moving average over five points, and normalized to their maximum.

For a sphere of radius R, we computed the analytic expression of the theoretically expected profile for a homogeneous distribution at the surface and/or in the volume, as a function of the axial position z (corresponding to the latitude for the sphere). The elementary surface dS and volume dV of a section of radius r located at a slice with latitude between z and z+dz (corresponding to an arc of length dl) are:
formula
(1)
Note that a purely surface location leads to a homogeneous distribution that is constant at all z values. The gradual inclination of the surface from the equator to the poles compensates exactly for the decrease of the radius r of the circle at latitude z. Alternatively, a homogeneous distribution over the entire volume of the sphere leads to an axial profile corresponding to a half-ellipse. A combination of both surface and volume distributions leads to the sum of uniform and half-elliptic profiles, dS+dV (Fig. S4C–F).

Monte Carlo simulation

Brownian motions were simulated with MATLAB (The Mathworks) by generating and summing 1000 random steps along the x- and y-axes, with a normal distribution of standard deviation (2)1/2. Each trajectory had a length set according to an exponential decay probability. We used parameters matching experimental values, with a pixel of 0.16 µm and a time lag τ=0.1 s.

For evaluating random colocalization, two sets of trajectories were generated, within a square with sides of 14.24 µm (89 pixels) equivalent to the average cell area at contacts. The average trajectory number (53 for JAM-C and 49 for JAM-B), diffusion coefficient (0.031 µm2/s for JAM-C and 0.0065 µm2/s for JAM-B) and trajectory length (24 frames for JAM-C and 49 frames for JAM-B) were set equal to those of the experimental values in control conditions (JAM-Clow KG1 cells labeled with non-blocking anti-JAM-C antibody).

To simulate confined, Brownian and directed trajectories, we used a diffusion coefficient of D=0.01 µm2/s. For confined motion, trajectories were restrained by symmetrical reflections on borders, within a square with a side length ranging from 1 to 10 µm. For directed motion of velocity v ranging from 1 to 10 µm/s, a constant displacement of amplitude was added to each step, along a constant and randomly set direction. For each condition, 100 trajectories were generated, starting at random x, y positions.

SMT determination of the input diffusion coefficient for MTT

After each experiment, videos were automatically analyzed by our in-house dedicated algorithm called multiple-target tracing (MTT) (Rouger et al., 2012; Sergé et al., 2008) to reconstitute SM trajectories. Thresholds were kept at their initial values, as optimized for the conception of MTT. The only parameter that was modified is the input diffusion coefficient Din. Indeed, the main parameter that may be modified before running MTT is the input diffusion coefficient Din (initially termed Dmax), which determines the search area for frame to frame reconnexion, as detailed in Mathematical Appendix 2 in reference Sergé et al. (2008). However, diffusion may inherently and substantially vary from one molecule, in a given cellular context, to another. Din thus needs to be optimized for every biological system. Din (initially set at 0.2 µm2/s for EGFR) was set at 0.02 µm2/s for JAMs. Considering that the distribution of JAM-B and JAM-C diffusion coefficients largely overlap, spreading essentially from 10−4 to 1 µm2/s, the determination of Din was performed on all trajectories, pooling JAM-B and JAM-C, leading to a single common value for Din, Note that since this parameter does not correspond to a fixed limit, but to the standard deviation of the Gaussian law used to compute the probability of reconnection, the diffusion coefficient of the reconstructed trajectories may be higher than Din.

Carefully adjustment of Din was performed not only in the current work, but is recommended for any MTT use. A script runs MTT on a representative dataset, for a range of input values, such as from 10−5 to 10 pxl2/frame. The code loops over Din values and computes the average Dout. Since running MTT in a loop can be time consuming, a first trial may be done with only one file, a reduced number of Din values and frames.

The plot of Dout versus Din (in log scale) is expected to display three regimes (Fig. S6A): (1) a slope close to 1 at low values (MTT first reconnects only very slow targets, then increasing Din generates new connections, with higher Dout, almost proportional to Din); (2) then a plateau (or at least a lower slope), when Din reaches the value that matches the system (ideally, Dout no longer depends on Din, at least to some extent), and (3), then again, Dout should be almost proportional to Din, allowing larger steps leads to larger diffusion values (but most of time this is irrelevant, as may easily be seen by plotting traces; at very large and irrelevant Din, MTT gradually reconnects any position).

Hence, the conclusion is that the plateau, or the smallest slope, determines the optimal value to use for Din. To account for all input values, the optimal Din may finally be computed as the average of Din values, weighted by the invert of the slope as optimal Din=sum(Din/δDout)/sum(1/δDout) with δDout=diff(Dout). For JAMS, we obtained Din=0.08 pxl2/frame=0.02 µm2/s.

SMT MTT2col trajectory analysis

Of note, MTT parameters are listed in the Table S1 in reference Sergé et al., (2008). Here is the list of the principal input values for MTT:

  • Spatial sliding window width used for particle detection: 7 pixels

  • Probability of false alarm: 24 (corresponding to 10−6 pixel−2)

  • Probability of false alarm for orphan peaks: 28 (corresponding to 10−7 pixel−2)

  • Disappearance probability (exponential decay) for blinking: 15 frames

  • Temporal sliding window length for past statistics: 5 frames

  • No cutoff on intensity

To discriminate at best neighbor particles, proximity is handled in combination with other past statistical parameters, namely, brightness, diffusion and blinking. The coefficients leading to the combined probability are given in the original article (Sergé et al., 2008). Of note, all input parameters are automatically saved as a text file in the MTT output folder, allowing their values to be checked afterwards.

Briefly, for SM detection, the intensity of the image is analyzed at the center of a sliding window (7×7 pixels) in which the probabilities of either presence (H1) or absence (H0) of a SM are compared, taking into account the intensity of the background noise. Targets are then identified by maximum likelihood estimation at each local maximum by a likelihood ratio test: the ratio H1/H0 gets high when peak intensities are separated enough from the noise. A threshold is set to ensure a low probability of false positives (spurious detection). Each peak amplitude and position, at sub-pixel accuracy, are estimated with a bi-dimensional Gaussian fit approximating the point-spread function with a width set by the diffraction limit, using the Gauss–Newton algorithm. Hence, small aggregates with a physical size remaining below the diffraction limit will be seen as point-spread function signals and detected as single targets, with higher signal intensity (Schmidt, 1996). However, clusters exhibiting a signal larger than the diffraction limit will lead to a poor maximum likelihood estimation score and will be statistically not detected (see for instance Movie 2). However, a higher signal in part of an image is statistically associated with higher variance. Hence, such areas are prone to lead to more false positives.

New targets are then reconnected with the trajectories already detected, if possible, considering all available criteria (not only proximity, but also intensity, diffusion and blinking). New trajectories are initiated when reconnection is not statistically achievable. The numbers of trajectories were normalized as densities per µm2, to account for area differences in and out of contacts. Mean squared displacement were computed for all trajectories, MSD=<r2>, where brackets denote average of square displacements r2 (in µm2) over time-steps. The diffusion coefficient D (in µm2/s) and positional accuracy σ (in µm) are next determined by a linear fit on points 1 to 4 of the MSD, according to the equation MSD=4Dt+2σ2 where t is time (in s). The value of the minimal diffusion coefficient was evaluated by analyzing immobile quantum-dots deposited on a coverslip, using the same experimental settings as used for live cells. This value is determined both by the experimental (notably optical and camera settings, acquisition speed) and analytic workflow (target detection and reconstruction of trajectories by MTT). When using optimal conditions, such measures led to values in the range of 0.001 μm2/s (Fig. S6B). Alexa Fluor 488 and Alexa Fluor 594 dyes led to slightly higher values, ∼0.0015 μm2/s. Accordingly, for SMT analysis, a cutoff was set at. 0.001 μm2/s to discard trajectories which, given this lower limit, cannot be assigned as either fixed or very slow. Of note, long trajectories, as obtained with quantum-dots, are susceptible to successively exhibiting multiple modes of motion and should thus be preferentially analyzed using a temporal sliding window. By contrast, we hypothesized that reasonably short trajectories (median length of 25 steps=2.5 s for our experimental results) would provide an intrinsic sampling of the different behaviors with relatively infrequent transitions.

SMT MTT2col trajectory classification

Taking an envelope surrounding all JAM-C positions provided a mask allowing defining contact versus out of contact areas for JAM-B. Alternatively, masks were generated manually when JAM-C density was too low. For JAM-C, out of contact measures could not be obtained from the same focal plane. Instead, such measures were performed by focusing on top of KG1 cells. Registration parameters (translation, rotation and zoom) among channels were determined from the transmission images of the cells in each channel using the MATLAB function imregistration. Standard deviation of registration parameters among images was on average 0.18 pixel for translation in x and y, 0.03° for rotation and 4×10−4 for zoom. These values were low enough to ensure a registration error below 0.3 pixel, even in the corners of the images. Considering a colocalization threshold set at 2 pixels, such errors are unlikely to substantially contribute to colocalization. SM positions in the magenta channel were next registered respectively to the green channel, before detection of spatiotemporal colocalization events (Fig. S2).

The distinction between KG1 leukemic cells expressing JAM-C at low or high level within the parental KG1 cell line was made by sorting data according to the number of JAM-C trajectories, which was fitted by a bimodal Gaussian distribution. This allowed discrimination between JAM-Clow and JAM-Chigh subpopulations, with a cutoff set by the intersection of the two Gaussians, at 150 JAM-C trajectories per cell (Fig. 2B).

SMT MTT2col trajectory descriptors

Each SM trajectory can be characterized by descriptors such as geometric parameters measuring the deviation from either a line or a circle (Fig. 4A). The linearity was defined as L=(DP/TP)0.3, ratio of the direct path (DP, distance from first to last point of the trajectory) to the total path (TP, sum of elementary displacements). Note that L is close to 0 or 1 for confined or directed motion, respectively, with intermediate values for Brownian motion. The exponent 0.3 ensured mean values around 0.5 for Brownian trajectories.

The shape of the MSD versus time (right) is linear for pure diffusion, <r2>=4Dt+2σ2, upwardly curved for directed motion and downwardly curved for confinement (or obstruction by obstacles, or even immobility) (Fig. 4B). Indeed, for pure directed motion, r=vt, thus the MSD is quadratic with time, <r2>=v2t2. For confined motion, the MSD remains linear with time at short time, <r2>=4Dt+2σ2, when confinement can be neglected, but reaches an upper value at long time, directly set by the size of the confinement zone, leading to a negative curvature and a plateau of the MSD curve (Kusumi et al., 1993). For each trajectory, the MSD curve was fitted by least-square regression with the expression for anomalous diffusion, <r2>=4Dγtγ+2σ2, where γ is the anomalous exponent and Dγ is the generalized diffusion coefficient (Saxton, 1994). γ is close to 1 for Brownian motion, and lower and higher than 1 for confined and directed motion, respectively. Visual inspection of trajectories and MSD fits confirmed that the classification was statistically robust, in particular by considering longer trajectories (more than 30 steps) with consistent behavior over time. Alternatively, the offset 2σ2 could be first estimated by linear fit over the three first points of the MSD, allowing us to perform a linear fit on the logarithm of the resulting, corrected MSDcorr=MSD−2σ2, according to log(MSDcorr)=log(4Dγ)+γ log(t). This double linear fit procedure led to overall similar results. Anomalous fitting of the MSD curve allowed the sorting of trajectories into three distinct categories, therefore determining whether a movement is either Brownian (γ1), directed (γ1) or confined (γ1) (Fig. 5A). Boundaries among these three regimes were set at the mean plus or minus the standard deviation of γ for Brownian diffusion, as determined from the distribution of γ obtained from 400 Brownian simulations. A Gaussian fit led to a mean of 0.96 and a standard deviation of 0.42, leading to a lower threshold of 0.54 and upper threshold of 1.38. This ensured that 95% of Brownian trajectories would be adequately classified. Of note, simulations suggested that trajectories could be sorted according to a linear combination of Dγ and γ (Fig. 5B) resulting in boundaries defined by inclined lines, instead of horizontal lines, in the (Dγ, γ) space. However, this overall led to similar conclusions.

Statistics

Each data set was first tested for normality using a one-sample Kolmogorov–Smirnov test. When both data sets had at least seven values and if they were both evaluated as normal, comparison was evaluated by a two-tailed unpaired Student's t-test, assuming equal variance, and by a two-tailed Wilcoxon–Mann–Whitney rank sum test otherwise. All statistics were performed with MATLAB statistical toolbox (The Mathworks). Cohen's d coefficient, assessing the size effect between two data sets, was computed as d=(m2m1)/sdall where m1 and m2 are the mean of each data set, and sdall is the standard deviation of all data.

Data and code availability

The data that support the findings of this study, and the source code for MTT_2_colors, are available as open-source data and software for academic and non-profit research from the corresponding author upon reasonable request. The code is available at https://github.com/arnauldserge1/MTT2col.

We thank Dr Magali Irla (Center of Immunology of Marseille Luminy, Aix-Marseille University), Dr Olivier Théodoly (Laboratory of Adhesion and Inflammation, Aix-Marseille University) and Dr Laurence Salomé (Institute of Pharmacology and Structural Biology, Toulouse University) for critical reading of the manuscript. We thank Dr Michel Aurrand-Lions (Cancer Research Center of Marseille, Aix-Marseille University) for discussions and for providing in-house anti-JAM antibodies, Dr Maria de Grandis (Cancer Research Center of Marseille, Aix-Marseille University) for discussions and help in cell sorting, Dr Dalia El Arawi and Dr Laurent Limozin (Laboratory of Adhesion and Inflammation, Aix-Marseille University) for providing technical help for the microscopy setup. We also thank the microscopy, cytometry and statistics research facilities at CRCM as well as the cell culture facility at LAI for excellent technical support.

Author contributions

Conceptualization: A.S.; Methodology: A.S.; Software: A.S.; Validation: A.S.; Formal analysis: O.G., J.C., L.M., A.S.; Investigation: O.G., J.C., L.M., A.S.; Resources: A.S.; Data curation: O.G., J.C., L.M., A.S.; Writing - original draft: A.S.; Visualization: A.S.; Supervision: A.S.; Project administration: A.S.; Funding acquisition: A.S.

Funding

This work was supported by the Fondation ARC pour la Recherche sur le Cancer (PJA 20131200238 to A.S.) and institutional grants from Institut National de la Santé et de la Recherche Médicale (Inserm) and Aix-Marseille Université. O.G. has received funding from the European Horizon 2020 Framework Programme under the Marie Skłodowska-Curie grand agreement no. 713750, with the financial support of the Regional Council of Provence-Alpes-Cȏte d'Azur, A*MIDEX (no. ANR-11-IDEX-0001-02), funded by the Investissements d'Avenir project from the French Government, managed by the Agence Nationale de la Recherche (ANR).

The peer review history is available online at https://journals.biologists.com/jcs/article-lookup/doi/10.1242/jcs.258736

Al-Mawali
,
A.
,
Gillis
,
D.
and
Lewis
,
I.
(
2016
).
Immunoprofiling of leukemic stem cells CD34+/CD38−/CD123+ delineate FLT3/ITD-positive clones
.
J. Hematol. Oncol.
9
,
61
.
Arcangeli
,
M.-L.
,
Frontera
,
V.
,
Bardin
,
F.
,
Obrados
,
E.
,
Adams
,
S.
,
Chabannon
,
C.
,
Schiff
,
C.
,
Mancini
,
S. J. C.
,
Adams
,
R. H.
and
Aurrand-Lions
,
M.
(
2011
).
JAM-B regulates maintenance of hematopoietic stem cells in the bone marrow
.
Blood
118
,
4609
-
4619
.
Arcangeli
,
M.-L.
,
Bardin
,
F.
,
Frontera
,
V.
,
Bidaut
,
G.
,
Obrados
,
E.
,
Adams
,
R. H.
,
Chabannon
,
C.
and
Aurrand-Lions
,
M.
(
2014
).
Function of Jam-B/Jam-C interaction in homing and mobilization of human and mouse hematopoietic stem and progenitor cells
.
Stem Cells
32
,
1043
-
1054
.
Arrate
,
M. P.
,
Rodriguez
,
J. M.
,
Tran
,
T. M.
,
Brock
,
T. A.
and
Cunningham
,
S. A.
(
2001
).
Cloning of human junctional adhesion molecule 3 (JAM3) and its identification as the JAM2 counter-receptor
.
J. Biol. Chem.
276
,
45826
-
45832
.
Aurrand-Lions
,
M.
,
Duncan
,
L.
,
Ballestrem
,
C.
and
Imhof
,
B. A.
(
2001
).
JAM-2, a novel immunoglobulin superfamily molecule, expressed by endothelial and lymphatic cells
.
J. Biol. Chem.
276
,
2733
-
2741
.
Bakker
,
G. J.
,
Eich
,
C.
,
Torreno-Pina
,
J. A.
,
Diez-Ahedo
,
R.
,
Perez-Samper
,
G.
,
van Zanten
,
T. S.
,
Figdor
,
C. G.
,
Cambi
,
A.
and
Garcia-Parajo
,
M. F.
(
2012
).
Lateral mobility of individual integrin nanoclusters orchestrates the onset for leukocyte adhesion
.
Proc. Natl. Acad. Sci. USA
109
,
4869
-
4874
.
Bazzoni
,
G.
,
Martinez-Estrada
,
O. M.
,
Orsenigo
,
F.
,
Cordenonsi
,
M.
,
Citi
,
S.
and
Dejana
,
E.
(
2000
).
Interaction of junctional adhesion molecule with the tight junction components ZO-1, cingulin, and occludin
.
J. Biol. Chem.
275
,
20520
-
20526
.
Betzig
,
E.
(
1995
).
Proposed method for molecular optical imaging
.
Opt. Lett.
20
,
237
-
239
.
Betzig
,
E.
,
Patterson
,
G. H.
,
Sougrat
,
R.
,
Lindwasser
,
O. W.
,
Olenych
,
S.
,
Bonifacino
,
J. S.
,
Davidson
,
M. W.
,
Lippincott-Schwartz
,
J.
and
Hess
,
H. F.
(
2006
).
Imaging intracellular fluorescent proteins at nanometer resolution
.
Science (80-.)
313
,
1642
-
1645
.
Burkhart
,
J. M.
,
Vaudel
,
M.
,
Gambaryan
,
S.
,
Radau
,
S.
,
Walter
,
U.
,
Martens
,
L.
,
Geiger
,
J.
,
Sickmann
,
A.
and
Zahedi
,
R. P.
(
2012
).
The first comprehensive and quantitative analysis of human platelet protein composition allows the comparative analysis of structural and functional pathways
.
Blood
120
,
e73
-
e82
.
Chanier
,
T.
and
Chames
,
P.
(
2019
).
Nanobody Engineering: Toward Next Generation Immunotherapies and Immunoimaging of Cancer
.
Antibodies
8
,
13
.
De Grandis
,
M.
,
Lhoumeau
,
A.-C.
,
Mancini
,
S. J. C.
and
Aurrand-Lions
,
M.
(
2015
).
Adhesion receptors involved in HSC and early-B cell interactions with bone marrow microenvironment
.
Cell. Mol. Life Sci.
73
,
687
-
703
.
De Grandis
,
M.
,
Bardin
,
F.
,
Fauriat
,
C.
,
Zemmour
,
C.
,
El Kaoutari
,
A.
,
Sergé
,
A.
,
Granjeaud
,
S.
,
Pouyet
,
L.
,
Montersino
,
C.
,
Chretien
,
A. S.
et al. 
(
2017
).
JAM-C identifies Src family kinase-activated leukemia-initiating cells and predicts poor prognosis in acute myeloid leukemia
.
Cancer Res.
77
,
6627
-
6640
.
Dickson
,
R. M.
,
Cubitt
,
A. B.
,
Tsien
,
R. Y.
and
Moerner
,
W. E.
(
1997
).
On/off blinking and switching behaviour of single molecules of green fluorescent protein
.
Nature
388
,
355
-
358
.
Ebnet
,
K.
,
Schulz
,
C. U.
,
Meyer Zu Brickwedde
,
M.-K.
,
Pendl
,
G. G.
and
Vestweber
,
D.
(
2000
).
Junctional adhesion molecule interacts with the PDZ domain-containing proteins AF-6 and ZO-1
.
J. Biol. Chem.
275
,
27979
-
27988
.
Heisenberg
,
W.
(
1949
).
The Physical Principles of the Quantum Theory
.
Dover Publications
.
Hell
,
S. W.
(
2007
).
Far-field optical nanoscopy
.
Science (80-.)
316
,
1153
-
1158
.
Hell
,
S. W.
and
Wichmann
,
J.
(
1994
).
Breaking the diffraction resolution limit by stimulated emission: stimulated-emission-depletion fluorescence microscopy
.
Opt. Lett.
19
,
780
-
782
.
Ishibashi
,
M.
,
Miyanaga
,
Y.
,
Matsuoka
,
S.
,
Kozuka
,
J.
,
Togashi
,
Y.
,
Kinashi
,
T.
and
Ueda
,
M.
(
2015
).
Integrin LFA-1 regulates cell adhesion via transient clutch formation
.
Biochem. Biophys. Res. Commun.
464
,
459
-
466
.
Jacobson
,
K.
,
Sheets
,
E. D.
and
Simson
,
R.
(
1995
).
Revisiting the fluid mosaic model of membranes
.
Science (80-.)
268
,
1441
-
1442
.
Konopleva
,
M.
,
Tabe
,
Y.
,
Zeng
,
Z.
and
Andreeff
,
M.
(
2009
).
Therapeutic targeting of microenvironmental interactions in leukemia: Mechanisms and approaches
.
Drug Resist. Update
12
,
103
-
113
.
Kostelnik
,
K. B.
,
Barker
,
A.
,
Schultz
,
C.
,
Mitchell
,
T. P.
,
Rajeeve
,
V.
,
White
,
I. J.
,
Aurrand-Lions
,
M.
,
Nourshargh
,
S.
,
Cutillas
,
P.
and
Nightingale
,
T. D.
(
2019
).
Dynamic trafficking and turnover of JAM-C is essential for endothelial cell migration
.
PLoS Biol.
17
,
e3000554
.
Kummer
,
D.
and
Ebnet
,
K.
(
2018
).
Junctional Adhesion Molecules (JAMs): The JAM-Integrin Connection
.
Cells
7
,
25
.
Kusumi
,
A.
,
Sako
,
Y.
and
Yamamoto
,
M.
(
1993
).
Confined lateral diffusion of membrane receptors as studied by single particle tracking (nanovid microscopy). Effects of calcium-induced differentiation in cultured epithelial cells
.
Biophys. J.
65
,
2021
-
2040
.
Kusumi
,
A.
,
Fujiwara
,
T. K.
,
Morone
,
N.
,
Yoshida
,
K. J.
,
Chadda
,
R.
,
Xie
,
M.
,
Kasai
,
R. S.
and
Suzuki
,
K. G. N.
(
2012
).
Membrane mechanisms for signal transduction: the coupling of the meso-scale raft domains to membrane-skeleton-induced compartments and dynamic protein complexes
.
Semin. Cell Dev. Biol.
23
,
126
-
144
.
Letellier
,
M.
,
Szíber
,
Z.
,
Chamma
,
I.
,
Saphy
,
C.
,
Papasideri
,
I.
,
Tessier
,
B.
,
Sainlos
,
M.
,
Czöndör
,
K.
and
Thoumine
,
O.
(
2018
).
A unique intracellular tyrosine in neuroligin-1 regulates AMPA receptor recruitment during synapse differentiation and potentiation
.
Nat. Commun.
9
,
3979
.
Liss
,
V.
,
Barlag
,
B.
,
Nietschke
,
M.
and
Hensel
,
M.
(
2015
).
Self-labelling enzymes as universal tags for fluorescence microscopy, super-resolution microscopy and electron microscopy
.
Sci. Rep.
5
,
17740
.
Martin-Padura
,
I.
,
Lostaglio
,
S.
,
Schneemann
,
M.
,
Williams
,
L.
,
Romano
,
M.
,
Fruscella
,
P.
,
Panzeri
,
C.
,
Stoppacciaro
,
A.
,
Ruco
,
L.
,
Villa
,
A.
et al. 
(
1998
).
Junctional adhesion molecule, a novel member of the immunoglobulin superfamily that distributes at intercellular junctions and modulates monocyte transmigration
.
J. Cell Biol.
142
,
117
-
127
.
Mascalchi
,
P.
,
Haanappel
,
E.
,
Carayon
,
K.
,
Mazères
,
S.
and
Salomé
,
L.
(
2012
).
Probing the influence of the particle in Single Particle Tracking measurements of lipid diffusion
.
Soft Mat.
8
,
4462
-
4470
.
Meads
,
M. B.
,
Hazlehurst
,
L. A.
and
Dalton
,
W. S.
(
2008
).
The bone marrow microenvironment as a tumor sanctuary and contributor to drug resistance
.
Clin. Cancer Res.
14
,
2519
-
2526
.
Moerner
,
W. E.
and
Kador
,
L.
(
1989
).
Optical detection and spectroscopy of single molecules in a solid
.
Phys. Rev. Lett.
62
,
2535
-
2538
.
Peddibhotla
,
S. S. D.
,
Brinkmann
,
B. F.
,
Kummer
,
D.
,
Tuncay
,
H.
,
Nakayama
,
M.
,
Adams
,
R. H.
,
Gerke
,
V.
and
Ebnet
,
K.
(
2013
).
Tetraspanin CD9 links junctional adhesion molecule-A to αvβ3 integrin to mediate basic fibroblast growth factor-specific angiogenic signaling
.
Mol. Biol. Cell
24
,
933
-
944
.
Rashidi
,
A.
and
DiPersio
,
J. F.
(
2016
).
Targeting the leukemia–stroma interaction in acute myeloid leukemia: rationale and latest evidence
.
Ther. Adv. Hematol.
7
,
40
-
51
.
Rossier
,
O.
,
Octeau
,
V.
,
Sibarita
,
J.-B.
,
Leduc
,
C.
,
Tessier
,
B.
,
Nair
,
D.
,
Gatterdam
,
V.
,
Destaing
,
O.
,
Albigès-Rizo
,
C.
,
Tampé
,
R.
et al. 
(
2012
).
Integrins beta(1) and beta(3) exhibit distinct dynamic nanoscale organizations inside focal adhesions
.
Nat. Cell Biol.
14
,
1057
-
1067
.
Rouger
,
V.
,
Bertaux
,
N.
,
Trombik
,
T.
,
Mailfert
,
S.
,
Billaudeau
,
C.
,
Marguet
,
D.
and
Sergé
,
A.
(
2012
).
Mapping molecular diffusion in the plasma membrane by Multiple-Target Tracing (MTT)
.
J Vis Exp
,
63
,
e3599
.
Saxton
,
M. J.
(
1994
).
Anomalous diffusion due to obstacles: a Monte Carlo study
.
Biophys. J.
66
,
394
-
401
.
Schepers
,
K.
,
Campbell
,
T. B.
and
Passegué
,
E.
(
2015
).
Normal and leukemic stem cell niches: Insights and therapeutic opportunities
.
Cell Stem Cell
16
,
254
-
267
.
Schmidt
,
T.
,
Schütz
,
G. J.
,
Gruber
,
H. J.
and
Schindler
,
H.
(
1996
).
Local stoichiometries determined by counting individual molecules
.
Anal. Chem.
68
,
4397
-
4401
.
Schnitzbauer
,
J.
,
Wang
,
Y.
,
Zhao
,
S.
,
Bakalar
,
M.
,
Nuwal
,
T.
,
Chen
,
B.
and
Huang
,
B.
(
2018
).
Correlation analysis framework for localization-based superresolution microscopy
.
Proc. Natl. Acad. Sci. USA
115
,
3219
-
3224
.
Sergé
,
A.
(
2016
).
The Molecular Architecture of Cell Adhesion: Dynamic Remodeling Revealed by Videonanoscopy
.
Front Cell Dev Biol
4
,
36
.
Sergé
,
A.
,
Fourgeaud
,
L.
,
Hémar
,
A.
,
Choquet
,
D.
,
Hemar
,
A.
and
Choquet
,
D.
(
2002
).
Receptor activation and homer differentially control the lateral mobility of metabotropic glutamate receptor 5 in the neuronal membrane
.
J. Neurosci.
22
,
3910
-
3920
.
Sergé
,
A.
,
Fourgeaud
,
L.
,
Hémar
,
A.
and
Choquet
,
D.
(
2003
).
Active surface transport of metabotropic glutamate receptors through binding to microtubules and actin flow
.
J. Cell Sci.
116
,
5015
-
5022
.
Sergé
,
A.
,
Bertaux
,
N.
,
Rigneault
,
H.
and
Marguet
,
D.
(
2008
).
Dynamic multiple-target tracing to probe spatiotemporal cartography of cell membranes
.
Nat. Methods
5
,
687
-
694
.
Shcherbakova
,
D. M.
,
Sengupta
,
P.
,
Lippincott-Schwartz
,
J.
and
Verkhusha
,
V. V.
(
2014
).
Photocontrollable fluorescent proteins for superresolution imaging
.
Annu. Rev. Biophys.
43
,
303
-
329
.
Sun
,
Y.
,
Li
,
N.
and
Fang
,
X.
(
2015
).
Probing the dynamics of growth factor receptor by single-molecule fluorescence imaging
.
Prog. Biophys. Mol. Biol.
118
,
95
-
102
.
Triller
,
A.
(
2011
).
[Molecular imaging: towards an in cellulo chemistry]
.
Med Sci
27
,
451
-
452
.
Triller
,
A.
and
Choquet
,
D.
(
2008
).
New concepts in synaptic biology derived from single-molecule imaging
.
Neuron
59
,
359
-
374
.
Vergez
,
F.
,
Green
,
A. S.
,
Tamburini
,
J.
,
Sarry
,
J.-E.
,
Gaillard
,
B.
,
Cornillet-Lefebvre
,
P.
,
Pannetier
,
M.
,
Neyret
,
A.
,
Chapuis
,
N.
,
Ifrah
,
N.
et al. 
(
2011
).
High levels of CD34+CD38low/-CD123+ blasts are predictive of an adverse outcome in acute myeloid leukemia: a Groupe Ouest-Est des Leucemies Aigues et Maladies du Sang (GOELAMS) study
.
Haematologica
96
,
1792
-
1798
.

Competing interests

The authors declare no competing or financial interests.

Supplementary information