Morphoregulatory ADD3 underlies glioblastoma growth and formation of tumor-tumor connections

Glioblastoma is a major unmet clinical need characterized by striking inter- and intra-tumoral heterogeneity and a population of glioblastoma stem cells (GSCs), conferring aggressiveness and therapy resistance. GSCs communicate through a network of tumor-tumor connections (TTCs), including nanotubes and microtubes, promoting tumor progression. However, very little is known about the mechanisms underlying TTC formation and overall GSC morphology. As GSCs closely resemble neural progenitor cells during neurodevelopment, we hypothesised that GSCs’ morphological features affect tumour progression. We identified GSC morphology as a new layer of tumoral heterogeneity with important consequences on GSC proliferation. Strikingly, we showed that the neurodevelopmental morphoregulator ADD3, is sufficient and necessary for maintaining proper GSC morphology, TTC abundance and cell cycle progression as well as required for cell survival. Remarkably, both the effects on cell morphology and proliferation depend on the stability of actin cytoskeleton. Hence, cell morphology and its regulators play a key role in tumor progression by mediating cell-cell communication. We thus propose that GSC morphological heterogeneity holds the potential to identify new therapeutic targets and diagnostic markers.


Introduc9on
Glioblastoma (GBM) is the most aggressive and common form of primary brain malignancy in adults and an unmet clinical need (Tran and Rosenthal, 2010).Its high chance of relapse is largely due to its striking inter-and intra-tumoral heterogeneity along with its infiltraEon into the healthy brain parenchyma (Garofano et al., 2021;Petrecca et al., 2013;Spiteri et al., 2019).
Given the striking similariEes between neurodevelopment and GBM progression, neural progenitor cells could offer key insights into the molecular and cellular underpinnings of GBM cell morphology and its role in cancer progression.Moreover, a specific type of GBM cells, known as glioblastoma stem cells (GSCs), which confer aggressiveness and therapy resistance to the tumor (Azzarelli et al., 2018;Neiel et al., 2019), shows remarkable similariEes with a populaEon of neural progenitor cells called basal or outer radial glia (bRG or oRG), a key cell type underlying fetal development of the human cortex (Fietz et al., 2010;Hansen et al., 2010;Reillo et al., 2011).Not only do GSCs show transcriptomic signatures of bRG (Bhaduri et al., 2020;Couturier et al., 2020), but they also undergo a characterisEc type of cell movement, called mitoEc somal translocaEon (MST), previously reported only in fetal bRG (Bhaduri et al., 2020;Hansen et al., 2010;LaMonica et al., 2013).In bRG, cell morphology was shown to have an important role in underlying cell proliferaEon, migraEon and MST (Del-Valle-Anton and Borrell, 2022;Kalebic and Hu,ner, 2020;Molnar et al., 2019;Ostrem et al., 2017;Taverna et al., 2014).In fact, different bRG morphotypes were idenEfied (BeEzeau et al., 2013;Kalebic et al., 2019;Reillo et al., 2017) and increased morphological complexity has been linked to a greater proliferaEve potenEal (Kalebic et al., 2019).Considering such role of cell morphology in neurodevelopment and the presence of tumor microtubes in GBM, we hypothesized that morphological complexity affects GBM progression.
Here we idenEfied adducin-γ (ADD3), an acEn-associated protein known to control bRG morphology and proliferaEon, as a putaEve master morphoregulator of GSCs.We next invesEgated the morphological heterogeneity of GSCs in different GBM cell lines and found that they exist in four morphotypes, similar to neural progenitors in the developing brain.We demonstrated that ADD3 regulates the morphology of GSCs by inducing their elongaEon, branching and the formaEon of tumor microtubes.We further showed that the effect of ADD3 on cell morphology are necessary for cell survival and proliferaEon.Hence, we described cell morphology as a new layer of heterogeneity in GBM and idenEfied morphoregulatory proteins as potenEal targets to tackle GBM progression.

Iden9fica9on of morphoregulatory adducin-g (ADD3) in GBM
Considering the resemblance between GSCs and bRG, we first sought to idenEfy genes that might govern the GSC morphology by mining datasets of morphoregulators in fetal bRG.We combined previously published transcripEonal (Fietz et al., 2012) and proteomic (Kalebic et al., 2019) analyses and idenEfied 45 morpho-regulatory genes whose expression is enriched in bRG vs. other cell types of the developing brain.We next intersected this list with a published list of genes expressed in glioblastoma (Bhaduri et al., 2020) (Figure 1A).Among the 30 idenEfied genes, the adducin family was prominently present (Fisher exact test p-value = 3.34 x 10 -9 ) with all its three members (Figure 1B, B').
Adducins are morphoregulatory proteins involved in the assembly of the acEn-spectrin network and are implicated in growth of cell protrusions, membrane trafficking and providing mechanical stability to plasma membrane (Baines, 2010;Kiang and Leung, 2018;Kiang et al., 2020;Lou et al., 2013).Taking advantage of data from the Cell Model Passports (van der Meer et al., 2019) and the Cancer Dependency Maps (Behan et al., 2019;Pacini et al., 2021;Tsherniak et al., 2017), we excluded genes that were not expressed at the basal level in a panel of commercially available and mulE-omically characterised GBM cell lines (Figure 1B) and that are core-fitness essenEal genes (VinceE et al., 2021) (Figure S1A) shortlisEng a set of 15 candidate genes (Figure 1C).Of the 3 adducins, 2 (ADD1 and ADD3) were in this list, with adducin-γ (ADD3) showing a strong and seemingly context-specific, essenEality (Figure 1C).
The role of ADD3 in glioblastoma is not clear, as it has been reported to both promote and reduce tumor growth and invasiveness (Kiang et al., 2020;Rani et al., 2013).Furthermore, ADD3 has been associated with temozolomide resistance (Poon et al., 2015), glioma progression (Rani et al., 2013;van den Boom et al., 2003) and reduced glioma cell moElity (Mariani et al., 2001).Strikingly, we have previously shown that ADD3 is required for the correct morphology of human basal progenitors and that its depleEon results in a reducEon of their proliferaEon (Kalebic et al., 2019).

Morphological heterogeneity of GSCs and subcellular localiza9on of ADD3
Analysis of Cancer Dependency Map datasets revealed that the glioblastoma cell line Onda-11 exhibits the strongest dependency on ADD3 (scaled depleEon fold-change upon CRISPR-Cas9 targeEng = -0.59,with -1 indicaEng the median depleEon fold-change of strongly essenEal core-fitness genes, such as ribosomal protein genes Figure 1C and S1B).To promote stemness of Onda-11 cells, we maintained them in serum-free culture condiEons and confirmed their stem-like features by immunofluorescence for nesEn, SOX2, L1CAM, OCT4, GFAP and CD44 (Figure S1C-I).
Upon transfecEon with GFP we examined the morphology of Onda-11 GSCs, and found a remarkable heterogeneity idenEfying eight morphotypes, which we grouped into four principal morphoclasses: non-polar, flat polar, circular mulEpolar and elongated (Figure 1D-F).
Morphologically these cells were reminiscent of neural progenitor cells during corEcal development (Kalebic and Hu,ner, 2020).Specifically, elongated morphoclass (radial and bifurcated morphotypes) and bipolar cells of the flat polar morphoclass, morphologically resemble morphotypes of bRG, whereas circular mulEpolar cells resemble mulEpolar basal progenitors (Kalebic et al., 2019).Instead, flat mulEpolar and flat non-polar GSCs do not seem to have a corresponding developmental morphotypes and likely arise during tumorigenesis.This suggest that, in addiEon to the molecular and cell behavioral features (Bhaduri et al., 2020), GSCs also recapitulate the morphological features of embryonic neural progenitors, in parEcular of bRG.We further confirmed the existence of the four morphoclasses in U87-MG glioblastoma cell line (see Figure S3D).
We examined the subcellular localizaEon of ADD3 in Onda-11 GSCs by confocal microscopy.
We observed that ADD3 readily localizes to the proximity of plasma membrane, to cellular protrusions and, specifically, TTCs (Figure 1G).Whereas ADD3 was enriched in protrusions that contained both microtubules and acEn, in TTCs it was present irrespecEvely of whether they contained acEn only or acEn and microtubules (Figure 1H).Considering the morphological heterogeneity of GSCs and the localizaEon of ADD3 to cellular protrusions, we next sought to examine the potenEal ability of ADD3 to affect GSC morphology and its role in glioblastoma growth.(H) Quan-fica-on of the expression of ADD3 in Onda11 GSC protrusions and microtubes.Error bar, SD; n = 3 independent cell cultures.

ADD3 is sufficient and required to control the number of protrusions and elonga9on of GSCs
We transfected Onda-11 GSCs with ADD3-over-expressing (ADD3 OE) and control plasmids along with GFP, to visualize cell shape, and performed a morphological analysis three days following transfecEon (Figure 2A and S2A).ADD3 OE led to an altered distribuEon of morphoclasses with a marked increase in the proporEon of elongated cells at the expense of the other three morphoclasses (Figure 2B).To examine various features of the cell morphology in a quanEtaEve manner, we established a machine learning-assisted pipeline for the automaEc segmentaEon and analysis of microscopy images (Figure 2C).Employing this pipeline to examine the effects of ADD3 OE, we observed a striking increase in the number of cellular protrusions (Figure 2D), both primary protrusions that grow directly from the cell body (Figure 2E), and all protrusions, which include also secondary and other higher-order protrusions, compared to the control.This was accompanied by an increase in both the average and the maximum length of cell protrusions (Figure 2F, G), which was confirmed also by the Scholl analysis (Figure S2B), and by an increase in protrusion branching (Figure 2H).
Together, this suggestes that ADD3 promotes both the formaEon and the growth of new protrusions.Besides, such an increase in cellular protrusions also enlarged cell perimeter and area (Figure 2I, J).Finally, the overall shape of ADD3-over expressing cells became more elongated, as their major axis was significantly longer than in the control cells, whereas the length of the minor axis was not affected (Figure 2K, L).Accordingly, cell eccentricity was increased, indicaEng a more ellipEc and elongated shape as opposed to circular (Figure 2M).
We next examined if ADD3 was required to maintain the correct Onda-11 morphology.We performed a CRISPR/Cas9-mediated knock-out (KO) of ADD3 and confirmed its efficiency by both immunoblot and immunofluorescence three days aier transfecEon (Figure S2C, D).
InspecEon of the Onda-11 morphology upon ADD3 KO showed altered distribuEon of morphoclasses with an apparent reducEon in the proporEon of the elongated cells and a relaEve increase in the non-polar cells (Figure 2N, O).Consistent with this and opposite to the effects of the over expression, ADD3 KO resulted in the reducEon of the number of protrusions, their length, branching index, cell perimeter and area (Figure 2P-R, S2E-J).This was accompanied by a reducEon in both major and minor axis length (Figure 2S, T), but did not result in a staEsEcally significant reducEon in cell eccentricity (Figure 2U).(D-M) Number of total (D) and primary (E) cell protrusions, average (F) and maximum (G) protrusion length, branching index (H), perimeter (I), area (J) , major (K) and minor (L) axis length and eccentricity (M) upon ADD3 OE vs. Control, calculated as described in (C).
(N-U) ADD3 KO reduces protrusion abundance and induces cell shrinkage.Onda 11 cells were transfected either with ADD3 KO plasmid or with gLacZ KO plasmid as control and their morphology was analyzed.
(O) Distribu-on of the 4 morphoclasses in control and ADD3 KO Onda 11 GSCs.
(P-U) Sholl analysis (P), perimeter (Q), area (R) , major (S) and minor (T) axis length and eccentricity (U) upon ADD3 KO vs. Control, calculated as described in (C).We examined if the above effects of ADD3 on cell morphology are perEnent to other glioblastoma cell lines.We performed ADD3 KO in U87-MG glioblastoma line and H4 neuroglioma line (Figure S3A, B, E, F).Whereas U87-MG showed strong morphological heterogeneity, which was comparable to Onda-11, H4 cells exhibited rather uniform morphologies (Figure S3C, D, G).Accordingly, KO of ADD3 resulted in a change in morphotype distribuEon in U87-MG, but not H4 cells (Figure S3D, G).Similarly to Onda-11, ADD3 KO in U87-MG cells resulted in an increase in non-polar cells at the expense of elongated ones.This suggests that the effects of ADD3 are perEnent to other glioblastoma cell lines, parEcularly those that exhibit a heterogeneous cell morphology.
Taken together, these analyses show that ADD3 is both sufficient and required to maintain correct cell morphology, including the correct number and length of cellular protrusions, their branching, cell size and elongaEon.Instead, ADD3 is sufficient to increase cell eccentricity, while its KO resulted in cell shrinkage without modifying the eccentricity.

ADD3 promotes morphological transi9ons during interphase
Considering the above change in the distribuEon of morphoclasses we sought to examine potenEal transiEons between GSC morphoclasses using Eme-lapse microscopy.Two days following transfecEon with GFP and ADD3 or control plasmids, Onda-11 GSCs were imaged for 60 hours.We first focused on the morphological dynamics during interphase and observed that Onda-11 GSCs only rarely undergo a transiEon between morphoclasses (Figure 3A As mitosis involves characterisEc morphological changes, we specifically examined the inheritance of the mother cell morphology upon the cell division.During the duraEon of the live imaging, around 40% of cells underwent mitosis (Figure S4B).In control the mother cell morphology was generally inherited by both daughter cells (Figure 3D, E, S4C).Among the four morphoclasses, non-polar cells again displayed the greatest number of transiEons (Figure 3E and Movie S4).Differently to what observed in interphase (Figure 3B), ADD3 OE was not able to alter the frequency of morphoclass transiEons in mitosis (Figure S4C and Movie S3).
However, upon ADD3 OE we observed (1) morphological transiEons of the progeny of flat polar dividing cells (Figure 3D, E and S4D) as well as (2) a subtle increase in the elongated progeny of the diving cells (Figure 3E, S4D and Movie S5).Both effects were similar to what described above for the interphase.InteresEngly, in a subset of elongated cells, we observed an MST-like behavior, which however did not seem to be regulated by the over expression of ADD3 (Figure S4E).
Taken together, these data show that the morphoclass idenEty is largely conserved both in interphase and in relaEon to mitosis.The morphological heterogeneity instead seems to be principally generated by the morphological dynamics of non-polar cells in both interphase and mitosis.ADD3 over expression led to an increase in transiEons from all morphoclasses into elongated cells, which is consistent with the increase in the proporEon of elongated cells described above (Figure 2).Finally, while this effect was mild in mitosis, it led to a marked increase in elongated cells during interphase.See also Supplemental Movies S1-5.

ADD3 controls Onda-11 GSCs prolifera9on and survival
In light of the (1) effects of ADD3 on GSC morphology (Figure 2) and the previous data showing that (2) ADD3 underlies progenitor morphology and proliferaEon during corEcal development (Kalebic et al., 2019), we sought to examine the putaEve effects of ADD3 on the proliferaEon of Onda-11 GSCs.We first examined the expression pa,ern of Ki67, a marker of cell proliferaEon, and categorized cells in three phases of the cell cycle (Figure S5A).Upon ADD3 OE we detected a relaEve increase in the proporEon of cells in G0 and early G1 phases (Figure 4A, B).This led to a marked reducEon in the proporEon of cells in late G2 and M-phase, which was confirmed also by immunostaining for a mitoEc marker, phospho-vimenEn (pVim, Figure S5B-C).InvesEgaEng the Ki67 expression pa,ern across the four morphoclasses, revealed the strongest effect in circular mulEpolar cells and less prominently in elongated and nonpolar cells (Figure 4C).Upon EdU treatment of Onda-11 GSCs, we detected no difference in the proporEon of cells in S-phase (Figure S5D-E), suggesEng that the principal effects of ADD3 OE on GSC proliferaEon are related to G0/early G1 phases and mitosis.
We next examined the effects of the ADD3 KO on Onda-11 proliferaEon.Consistently with the above, the KO resulted in the opposite phenotypes compared to the OE.The proporEons of cells in both G0/early G1 and late G1/S/early G2 phases were reduced, as revealed by both Ki67 expression pa,ern and EdU treatment (Figure 4D-F and S6A).We further detected an increase in the proporEon of cells in G2/M (Figure 4E), but no specific increase in mitoEc pVim+ cells (Figure S6B-D).
Finally, we examined if the above effects of ADD3 on cell proliferaEon are perEnent to U87-MG glioblastoma and H4 neuroglioma cell lines.Similarly to the effects on cell morphology (Figure S3), ADD3 KO only affected the proliferaEon of U87-MG cells (Figure S7A-C), but not H4 cells (Figure S7D-F).Taken together, ADD3 enables correct Onda-11 proliferaEon and this effect is relevant also to other glioblastoma cell lines that show morphological heterogeneity.
Considering the dependence of Onda-11 on ADD3 (Figure S1B), we examined the apoptosis of KO cells by immunofluorescence for cleaved caspase-3 (Figure S6B) and detected a marked increase in cell death compared to the control (Figure 4G).Strikingly, this effect was not specific to transfected cells, but we detected a two-fold increase in apoptosis also in the surrounding cells (Figure 4H).Hence, ADD3 is required for the survival of Onda-11 GSCs in both cell-autonomous and non-autonomous manners.Such effects on both the targeted and the neighboring cells prompted us to (1) dissect the changes in the Onda-11 molecular signature following ADD3 manipulaEon and (2) examine the effects of ADD3 on intercellular connecEons mediaEng communicaEon between GSCs.

Cell-autonomous effects of ADD3 over expression
To elucidate the cell-autonomous effects of ADD3 OE, we performed a bulk RNA sequencing of GFP+ FACS-sorted cell co-transfected with ADD3 or control plasmids.The differenEal expression analysis revealed 10 upregulated and 7 downregulated genes upon ADD3 OE (Figure 4I).We demonstrated that the genes differenEally expressed upon ADD3 OE are indeed exhibiEng an expression pa,ern correlated with ADD3 also at the basal level in other GBM cell lines, i.e., the up-regulated genes are correlated, whereas down-regulated genes are anE-correlated with ADD3 (Figure S8), thus showing robustness of the ADD3 OE signature.
We next examined whether the effects of ADD3 on cell morphology and proliferaEon had consequences on cell fate and idenEty.Since ADD3 induced elongated and branched morphologies of GSCs (Figure 2) and led to a reducEon in cell cycle progression and division (Figure 4A-C, S5C) we examined the stemness of ADD3 OE cells and observed that ADD3 sustained as high level of stemness markers as control GSCs (Figure S9A-I).The ADD3 KO in turn led to a minor, albeit not staEsEcally significant, reducEon in some of the stemness markers (Figure S9J-R).
Considering that the same morphological and proliferaEon-related features are also linked to GBM invasiveness (Bhaduri et al., 2020;Venkataramani et al., 2022b), we generated neurospheres from FACS-sorted GFP+ cells over expressing ADD3 or control plasmid and examined their infiltraEon into the surrounding Matrigel.However, within one week, we did not observe any difference in the invasion index between ADD3 OE and control (Figure S10).
Finally, slowly dividing cells are oien associated with therapy resistance (Bao et al., 2006;Chen et al., 2012;Lathia et al., 2015).InteresEngly, expression of ADD3 has been previously linked with a populaEon of cells resistant to temozolomide, the main chemotherapeuEc used in the GBM treatment (Poon et al., 2015).Furthermore, its expression was also linked to mulEdrug resistance upon profiling 30 cancer cell lines (Gyorffy et al., 2006).We hence examined the potenEal signature of chemoresistance among the genes upregulated upon ADD3 OE (Figure 4I) and found CHI3L1 as a key molecule involved in temozolomide and radioresistance in GBM cell lines (Akiyama et al., 2014;Shao et al., 2014;Zhao et al., 2020).
Such chemoresistance is strongly associated to a network of tumor-tumor connecEons (TTCs), including TNTs and TMs (Kolba et al., 2019;Osswald et al., 2015;Wang et al., 2022;Weil et al., 2017).TMs are long protrusions enabling intercellular connecEons between largely slowly cycling GBM cells (Osswald et al., 2015;Ratliff et al., 2023).Considering that ADD3 promotes protrusion growth and branching along with a reducEon in cell cycle progression and that the ADD3 KO has both cell-autonomous and non-autonomous effects on cell survival, we next sought to examine if ADD3-related phenotypes are specifically mediated by TTCs.

ADD3-induced tumor cell-tumor cell connec9ons (TTCs) are required for the effects of ADD3 on GSC prolifera9on
To invesEgate if ADD3 could affect the TTC abundance, we stained Onda-11 GSCs over expressing ADD3 with phalloidin and α-tubulin to detect acEn and microtubules, respecEvely (Figure 5A).We detected doubling of TTCs connecEng adjacent cells and containing acEn cytoskeleton upon over expression of ADD3 (Figure 5B).Using correlaEve light-electron microscopy we idenEfied GFP+ co-transfected cells and then examined the ultrastructure of ADD3-induced TTCs using cryo-electron tomography (Figure 5C).This showed that such TTCs are strikingly enriched in acEn and that no microtubules were observed.Since the majority of TTCs were short and thin, they were likely TNTs.Nevertheless, we also observed TMs in control Onda-11 (see Figure 1G, H) and upon ADD3 OE (see Figure 5A), which was confirmed by IF for Connexin-43 (Figure S11).
We thus examined if intact acEn cytoskeleton is required for the maintenance of ADD3induced protrusions by treaEng the transfected Onda-11 GSC with Cytochalasin D, which causes disrupEon of acEn filaments and inhibits acEn polymerizaEon (Figure 5D).Consistently with the above (Figure 5B), DMSO-treated cells exhibited 2-fold increase in TTCs upon ADD3 OE (Figure 5E).In contrast, Cytochalasin D-treated cells lost all the ADD3-induced TTCs and showed similar levels between the control and OE (Figure 5E).
In light of the associaEon of TTCs with cell proliferaEon (Lu et al., 2019;Osswald et al., 2015;Ratliff et al., 2023;Valdebenito et al., 2018;Venkataramani et al., 2022a) and the ADD3induced phenotypes on both TTCs and cell cycle progression (Figures 4 and 5A-E), we sought to examine if the effects of the morphoregulatory ADD3 on cell morphology and TTCs are required for its effects on cell proliferaEon.We treated control and ADD3 OE Onda-11 GSCs with DMSO and Cytochalasin D and examined the expression pa,ern of Ki67 (Figure 5F), as a key indicator of the effects of ADD3 on cell cycle progression (see Figure 4).Our analysis shows that control cells treated with Cytochalasin D do not have different cell cycle progression compared to DMSO-treated control cells, suggesEng that the stability of acEn cytoskeleton is not required for their normal proliferaEon of onda-11 cells.In agreement of what we observed in untreated cells, ADD3 OE GSCs treated with DMSO showed a significant effect on cell proliferaEon, (Figure 5G and compare to 4B), whereas this effect was completely lost upon treatment with Cytochalasin D (Figure 5G).Taken together, these data suggest that ADD3 acts as a key regulator of GSC morphology to induce new acEn-rich TTCs, which in turn enable cell-cell contacts and mediate the downstream effects on cell proliferaEon.

Discussion:
In this study we idenEfied the GSC morphology as a key player underlying cell proliferaEon.
We further showed that the main driver of this effect are TTCs.There are three aspects of our study that deserve parEcular discussion: (1) Cell morphology is a new layer of GBM heterogeneity; (2) GSC morphology affects GSC proliferaEon and survival through cell-cell connecEons; (3) ADD3 is a key protein controlling GSC shape.

Morphology as a new layer of GBM heterogeneity
One of the key reasons for GBM's malignancy is its extraordinary inter-and intra-tumoral heterogeneity.The molecular heterogeneity, described at genomic, transcriptomic and epigeneEc levels, was shown to underlie a mulEtude of GBM cell types and states (Bhaduri et al., 2020;Chaligne et al., 2021;Couturier et al., 2020;Darmanis et al., 2017;Jacob et al., 2020;Neiel et al., 2019;Patel et al., 2014;So,oriva et al., 2013).In fact, it has been suggested that each GBM contains on average 11 different cell types (Bhaduri et al., 2020) that could be grouped into four principal cellular states which recapitulate disEnct neural cell types (Neiel et al., 2019).Notably, GSCs themselves show a striking molecular heterogeneity within the same tumor (Bhaduri et al., 2020).However, to link these specific cell types with cellular funcEons and oncological phenotypes, it is also necessary to study potenEal GBM heterogeneity at the cell biological level.
We have examined GSC morphology and idenEfied four different morphotypes in Onda-11 and U-87 MG cells (Figure 1), suggesEng that basic morphological nature is a cell-intrinsic property.The idenEfied morphotypes bore striking similarity to neural stem cells during corEcal development, in parEcular bRG (Kalebic and Hu,ner, 2020).This is consistent with a large body of evidence showing that GBM iniEaEon, maintenance and progression are controlled by the same signaling pathways and transcripEon factors that regulate brain development (Azzarelli et al., 2018;Curry and Glasgow, 2021).Despite good molecular understanding, the links between neurodevelopment and GBM at the cell biological level remain largely unexplored.Given that GSCs exist as similar morphotypes to those observed in bRG, where they promote cell proliferaEon (Kalebic et al., 2019), we asked what the funcEonal consequences of such neurodevelopmental features on GBM are.
A key quesEon to answer was whether the morphoypes are stable or transient cellular states.
Our live imaging experiments (Figure 3) suggested that the former is true both within and across cell cycles.It would hence be interesEng to examine if such stable morphotypes correspond to transcripEonally defined cell types.So, how is the morphological heterogeneity then generated?Our data showed that the cells of the nonpolar morphoclass are responsible for such heterogeneity, as they are able to generate all the remaining morphoclasses both in mitosis and by morphological transiEons in interphase.It is worth noEng that nonpolar morphoclass is not present among developmental neural progenitors in interphase (Kalebic and Hu,ner, 2020).Hence it is tempEng to hypothesize that morphological plasEcity of nonpolar cells might be a prominent feature of brain cancers, and that it might be linked to the characterisEc plasEcity among different GBM cell states (Neiel et al., 2019).

Cell-cell connec9ons link GSC morphology with prolifera9on and survival
To examine if different morphotypes have disEnct cellular funcEons, we analyzed their proliferaEon and observed differences in their cell cycle progression (Figure 4).Notably, modifying cell morphology, by over expression of ADD3 and thus generaEng more elongated cells, led to a reduced cell cycle progression.We found that the key morphological feature responsible for the change in proliferaEon, is the acEn-based tumor-tumor connecEons (TTCs), including both TNTs and TMs (Figure 5, S10).Whereas formaEon of TTCs has been implicated in increased cell proliferaEon (Joseph et al., 2022;Lu et al., 2019;Osswald et al., 2015), a recent study has shown that TM-rich, interconnected GBM cells have a slower cell cycle compared to the fast-dividing, unconnected cells in the invasion zone (Ratliff et al., 2023), which is in agreement with our data (Figures 4 and 5).
Furthermore, such TTC-rich cells over expressing ADD3, did not show altered invasive capacity (Figure S9).This is in line with recent in vivo studies showing that a different populaEon of GBM cells, which lacks connecEons to other GBM cells is the main driver of brain tumor invasion (Ratliff et al., 2023;Venkataramani et al., 2022b).Taken together, our data suggest that TTC-rich GBM cells over expressing ADD3 represent a populaEon of slowly proliferaEng cells either prior to the infiltraEon into the brain parenchyma or following it.TTC-rich GBM cells have also been associated with increased resistance to chemotherapy (Kolba et al., 2019;Osswald et al., 2015;Wang et al., 2022;Weil et al., 2017).Such cells were shown to be able to change their metabolic profile through a TNT-mediated mitochondria, vesicle and protein transfer (Hekmatshoar et al., 2018;Pinto et al., 2021).InteresEngly, upon ADD3 OE, we also observed upregulaEon of CHI3L1, which is involved in chemo-and radioresistance in GBM (Akiyama et al., 2014;Shao et al., 2014;Zhao et al., 2020), implying that ADD3-induced TTCs might be sustaining therapy resistance.This is further suggested by an increased expression of ADD3 in a populaEon of GBM cells resistant to various chemotherapeuEcs, including temozolomide (Gyorffy et al., 2006;Poon et al., 2015).
Hence, TTCs appear to be a key feature contribuEng to the funcEonal consequence of the morphological heterogeneity of GBM.Since GSCs' transcripEonal heterogeneity is mediated by both intrinsic and extrinsic factors (PrasetyanE and Medema, 2017), we propose that the morphological heterogeneity could also be controlled both cell-autonomously and nonautonomously and that TTCs might play a pivotal role in the la,er.In fact, we showed that ADD3, as an intrinsic factor promoEng morphological heterogeneity, has a criEcal role in cancer cell survival, both cell-autonomously and non-autonomously (Figure 4G, H).Such effect on surrounding cells might be due to the exchange of specific pro-apoptoEc signals from the ADD3 KO cell, or simply because of the lack of any signal from KO cells, following the striking reducEon in TTCs.
Taken together, GBM cell morphology mediates intercellular communicaEon and thus has important consequences on tumor cell proliferaEon, survival, invasiveness and resistance to therapy.Hence, in GBM, like in other cancers (Alizadeh et al., 2020;Barker et al., 2022;Wu et al., 2020), cell morphology has a strong potenEal to be used as a diagnosEc and prognosEc marker, through microscopy-based analysis of the tumor.

ADD3 as a key morphoregulator in GBM
We have idenEfied ADD3 as a key morphoregulator able to control GBM proliferaEon.We found that ADD3 exerts mulEple morphoregulatory funcEons on GSCs.Notably, it promotes cell elongaEon and induces various cell protrusions, including TTCs (Figures 2 and 5A-C).Such diverse roles are likely due to its close interacEon with acEn, a key cytoskeleton component regulaEng changes in cell shape.Indeed, when the acEn cytoskeleton is disrupted, ADD3 is not able to induce TTCs anymore (Figure 5D, E).The quesEon remains if ADD3 directly induces new protrusions by remodeling acEn in the membrane cytoskeleton or whether it stabilizes exisEng protrusions by connecEng acEn filaments to the plasma membrane.Previous work on other members of the adducin family of proteins seems to favor the la,er hypothesis as it has been shown that adducins regulate membrane stability by capping the fast-growing end of acEn filaments and connecEng spectrin-acEn cytoskeleton to membrane proteins (Anong et al., 2009;Baines, 2010;Kuhlman et al., 1996;Li et al., 1998).Accordingly, adducins were shown to stabilize neuronal synapses by controlling spine dynamics (Babic and Zinsmaier, 2011;Bednarek and Caroni, 2011;Pielage et al., 2011).In the context of GBM cells, it is tempEng to hypothesize that ADD3 stabilizes cellular projecEons by providing mechanical support.This ulEmately can lead to an increase in the number of stable cell protrusions, parEcularly long TTCs that enable cell-cell communicaEon.
Considering that such a role of ADD3 on acEn cytoskeleton is likely true across different types of cellular projecEons and cell types, it is plausible that its effects are not specific to Onda-11 GSCs, but generally applicable to GBM cells that are elongated and contain protrusions.In support of this, we showed that ADD3 plays an important role in maintaining the cell morphology of U-87MG cells (Figure S3).Furthermore, ADD3 has already been implicated in GBM progression, therapeuEc resistance and cell moElity (Kiang et al., 2020;Mariani et al., 2001;Poon et al., 2015;Rani et al., 2013;van den Boom et al., 2003).Nevertheless, the effects of ADD3 on both cell morphology and proliferaEon are more pronounced in Onda-11 GSC compared to U-87MG (compare Figure 2 to S3 and Figure 4 to S7).We link this to the noEon that Onda-11 cells were shown to be strongly dependent on ADD3 in the Cancer DepMap project (Behan et al., 2019;Pacini et al., 2021;Tsherniak et al., 2017), whereas U-87 were not.
Beyond the experimental validaEon of the findings reported in the Cancer DepMap, our results show that DepMap is an important resource for exploring the funcEon of cancer genes in an appropriate model system.In the future it would hence be interesEng to study the morphoregulatory mechanisms in more complex model systems such as in vivo or in paEentderived organoid system.Finally, ADD3 was previously shown to regulate the morphology of bRG during brain development (Kalebic et al., 2019).Its KO in human fetal brain Essue led to a reducEon in the number of protrusions of neural progenitors, which in turn resulted in a reducEon in the proliferaEve capacity of these cells (Kalebic et al., 2019), which ulEmately controls the developmental expansion of the cerebral cortex.This link between cell morphology and proliferaEon serves as a further example of how neurodevelopment can offer precious insights into brain cancers.It also provides a novel conceptual framework which allows for the idenEficaEon and mechanisEc characterizaEon of other potenEal molecular targets to be used in future diagnosEc and therapeuEc approaches in brain cancers.

Cell sor9ng
The cells were sorted 48 h aier transfecEon to isolate GFP+ cells.MoFLO Astrios EQ cell sorter, equipped with Summit 6.3.1 soiware (Beckman Coulter), was used for cell sorEng prior to the neurosphere formaEon assay, whereas Cytoflex SRT cell sorter, equipped with CytExpert SRT soiware (Beckman Coulter) was used for for cell sorEng prior to RNA and protein extracEon.An average sorEng rate of 500-1000 events per second at a sorEng pressure of 25 psi (for MoFLO Astrios EQ) or 15 psi (for Cytoflex SRT) with a 100 μm nozzle were maintained.

Plasmids
For the over expression of ADD3, human ADD3-encoding cDNA was amplified by PCR, using the forward and reverse primers CAAX_Xhol_Fw and CAAX_BgIII_Rev as reported above, and cloned into the pCAG vector.DNA was purified using the QIAquick PCR PurificaEon Kit (Qiagen, 28104) and all DNA plasmids were extracted and purified using the EndoFree Plasmid Maxi kit (Qiagen, 12362) following the manufacturer's instrucEons.
Neurospheres were kept Ell day 15 when they were fixed in 4% PFA for 20 minutes at RT.

Light microscopy
Confocal microscopy on fixed cells was performed using a Zeiss LSM980 point-scanning confocal or Zeiss LSM980-NLO point-scanning confocal based on Zeiss Observer7 inverted microscopes.The images were acquired with a PlanApo 10X/0.45dry or a PlanApo 20X/0.8dry or a PlanApo 40X/1.4NAoil immersion objecEves using 405 nm, 488 nm, 561 nm, 639 nm laser lines.The soiware used for all acquisiEons was Zen Blue 3.7 (Zeiss).Once the parameters of acquisiEon for control condiEons had been defined, they were kept constant for all the samples within the same experiment.
Time-lapse imaging on live Onda11 GSCs was performed as follows.48 h aier transfecEon, the sample was placed under a Zeiss LSM980 point-scanning confocal with a PlanApo 20X/0.8dry objecEve and imaged for approximaEvely 60 h.Z stacks of 18-20 µm range were taken with a Z step of 1 µm and an interval Eme of 30-40 min.

Correla9ve light-electron microscopy and cryo-electron tomography
QuanEfoil Gold Grids (R 2/2, Au, 200 mesh, QuanEfoil) were plasma-cleaned with a hydrogen and oxygen mix (20:80) for 15 seconds with a Gatan Solarus II and then washed for 1 hour with 100% EtOH.The grids were then coated with 5 ug/ml laminin (for 1 h at 37 °C) and around 25,000 Onda11 GSCs were seeded per grid.16 h later, the grids were plunged with a Leica EM GP2 plunger.During plunging, a drop of 3 μL BSA-coated 10-nm fiducial gold markers (Aurion) was applied on the EM grids for 2.5 seconds.Grids were stored into liquid nitrogen unEl acquisiEon.
Subsequently, cryo-fluorescent imaging was performed on Leica ThunderCryoCLEM system using the Navigator module of Leica LAS X soiware.Grids were focus-mapped using built-in soiware funcEons and imaged in Z-stacks of 10-12 slices and ≈1 µm step size in both transmi,ed light and green channel fluorescence.The grid maps were saved as .liffiles for subsequent idenEficaEon of the transfected cells at the cryo-transmission electron microscope (cryo-TEM).
Data acquisiEon was performed using a Thermo ScienEfic Titan Krios G4 TEM equipped with a Thermo ScienEfic Selectris X energy filter and a Thermo ScienEfic Falcon 4i direct electron detector.The microscope was operated at 300 keV in zero-loss mode with an energy filter slitwidth set to 10 eV.To idenEfy the area of interest for data collecEon the map acquired on the Leica Thunder were overlayed with the TEM images acquired with the MAPS (TFS) soiware.

Manual image analysis
All manual cell quanEficaEons were performed in Fiji ImageJ using the CellCounter funcEon, processed with Microsoi Excel, and plo,ed in GraphPad Prism.For manual analysis of Onda11, U87 and H4 cell morphology, we assigned GFP+ cells to one of the defined morphoclasses and morphotypes.The same was done for Ki67, where GFP+ Ki67+ cells were assigned to one of the three different Ki67 pa,erns of expression.PVim, Casp3, EdU, L1CAM, A2B5, NesEn, GFAP, OCT4, SOX2 posiEvity was also calculated using the CellCounter funcEon in Fiji ImageJ.All images were analyzed blindly.
For the Eme-lapse movies GFP+ morphoclasses were manually tracked overEme and scored for morphological change in interphase and mitosis.MitoEc somal translocaEon (MST) was defined as the distance the nucleus travels during the Eme step preceding mitosis.Maximum projecEons and generaEons of movies were carried out in Fiji ImageJ.
For neurosphere assay, Image analysis was carried out in Fiji ImageJ where the area of the core and the total neurosphere (including the protrusions) were measured with the freehand line tool.The invasion index was calculated by dividing the area of the core with total area of the neurosphere.

Automated image analysis
For the machine learning assisted pipeline for image analysis (Figure 2), we collected a total of 39 microscopy images, out of which we segmented the morphology of 1362 Onda11 cells, using CellPose, an arEficial neural network for automated cell segmentaEon.The "cyto2" pretrained model was chosen and retrained for improved Onda 11 cell segmentaEon.Each cell was labelled through its own image array using Python in a Jupyter notebook.As a first step, each cell was posiEoned singularly at the center of a new image array with the dimension of the biggest bounding box and saved as 'Eff' file.Subsequently, the following morphological features were extracted: area, perimeter, major and minor axis lengths, eccentricity.These properEes were engineered using the "regionprops_table" funcEon from the scikit-image library to compute properEes (measurements) out of labelled regions in the image arrays.
Eccentricity is a measure of cellular elongaEon and circularity, where an eccentricity equal to 0 indicates a circle, whereas values between 0 and 1 indicate an ellipse.
To analyse Onda 11 cell protrusions, we modified our previous semi-manual workflow named Progenitors Process Analysis (PPA) (Kalebic et al., 2019) and used to quanEfy number of primary and all protrusions, average and maximum protrusion length, branching index (raEo between the total number of protrusion and primary protrusions) and Sholl analysis.

Data driven selec9on of ADD3
To idenEfy genes that potenEally regulate glioblastoma stem cell morphology, we used a published tumor atlas of differenEally expressed genes in primary glioblastoma tumors (Bhaduri et al., 2020).We intersected this dataset with a list of morpho-regulatory genes involved in neurodevelopment idenEfied in (Kalebic et al., 2019).This yielded a list of 30 candidate genes.The enrichment of adducins among the 30 genes was calculated using the following parameters: total number of human protein coding genes = 19,396 (N), total number of adducins = 3 (n), number of selected genes = 30 (k), number of hits = 3 (x).We then invesEgated the expression level of the selected genes (29/30 genes as one of the genes, MGEA5, was not analyzed in the datasets menEoned below) in 48 annotated glioblastoma cell lines from Cancer Dependency Map Dataset (22Q2 version) (Behan et al., 2019;Pacini et al., 2021;Tsherniak et al., 2017) and the Sanger Cell Model Passport (van der Meer et al., 2019) observing a bimodal distribuEon from which we idenEfied 18 highly expressed genes (whose basal expression was seemingly generated by the distribuEon with the higher mean).
Subsequently, we derived the depleEon fold change of these 18 genes upon CRISPR-Cas9 targeEng in 48 GBM cell lines using the same resources.We excluded pan-cancer core fitness genes (as predicted in (VinceE et al., 2021)) and focused our a,enEon on ADD3 as an important morphoregulator during development (Kalebic et al., 2019), differenEally expressed in GBM (Bhaduri et al., 2020) and with a strong and context specific depleEon fold change in GBM cell lines.We then idenEfied Onda 11 as the GBM cell line with the highest dependency

Figure 1 .
Figure 1.Onda 11 GBM cells show morphological heterogeneity and are dependent on ADD3, a neurodevelopmental morphoregulator localised in GBM cell protrusions and TTCs.(A-C) Computa-onal iden-fica-on of ADD3 as a neurodevelopmental morphoregulator with a puta-ve role in GBM progression.Data come from (Bhaduri et al., 2020) and (Kalebic et al., 2019) (A) and Broad DepMap 22Q2 version and Sanger cell model passport (B and C) (A) Intersec-on between a list of 8509 differen-ally expressed genes in primary glioblastoma tumors and 45 neurodevelopmental morphoregulatory genes, resul-ng in 30 shared genes.(B) Log2(TPM+1) expression levels of the resulted gene list (29/30) averaged across 48 annotated glioblastoma cell lines from showing bimodal distribu-on (dashed line).(B') Density plot of the average expression levels of the adducin family of genes indica-ng the es-mated density with superimposed average expression levels of adducins.(C) Dependency of GBM cell lines (deple-on fold change (FC) distribu-on upon CRISPR/Cas9 targe-ng) on the 15 highly expressed non-core-fitness genes from panel (B).(D) Onda 11 GSCs were transfected with GFP and their cell morphology was analyzed 72 h later.MIP of 12 planes.Four different morphoclasses listed at the top of the images (elongated, circular mul-polar, flat polar, non-polar) are further divided into eight morphotypes annotated on the le^ of the images (radial, bifurcated, elongated branched, circular mul-polar, flat mul-polar, bipolar, flat non-polar, circular nonpolar).Scale bars; 10µm.(E-F) Analysis of Onda-11 morphology using GFP signal, 72 h a^er transfec-on, showing their morphological heterogeneity.Distribu-on of the 8 morphotypes (E) grouped into 4 morphoclasses (F).Mean of 8 independent transfec-ons.Error bars, SEM.(G-H) ADD3 is expressed in cellular protrusions and tumor cell-tumor cell connec-ons (TTCs) of Onda11 GSCs.(G) IF staining for ac-n (Phalloidin, white), microtubules (alpha-Tubulin, magenta), ADD3 (green) along with DAPI staining (blue), MIP of 12 planes.Image size 188.61x 188.61 µm (top); 35.69 x 53.14 µm (bodom).

(
A-M) ADD3 overexpression promotes cell elonga-on and protrusion abundance.Onda 11 cells were transfected either with GFP and ADD3 overexpressing plasmids (ADD3 OE) or with GFP and empty vector (control) and their morphology was analyzed.(A) Representa-ve examples of GFP+ (green) Onda 11 cell morphology in control (le^) and ADD3 OE (center).Scale bar, 200 µm.A close-up of elongated cells upon ADD3 OE (right, image width 250 µm).MIP of 12 planes.(B) Distribu-on of the 4 morphoclasses in control and ADD3 OE Onda 11 GSCs.(C) Schema-cs of the pipeline for automated cell segmenta-on and morphological analysis of cells.GFP+ cells from confocal microscopy images (MIPs of 25 planes) are segmented in CellPose and single cells are isolated to carry out morphological analysis in Python and Fiji using PPA 2.0 macro.
, B and Movie S1).In fact, only the nonpolar cells exhibited morphological dynamics in interphase (Figure 3C, S4A).ADD3 was however able to promote such morphoclass transiEons (Figure 3A, B) with an increase in transiEons into elongated cells (Figure 3A, C and Movie S2 and compare to Movie S1).

Figure 3 .
Figure 3. ADD3 promotes morphological transiEons in interphase.Onda 11 cells were transfected either with GFP and ADD3 overexpressing plasmids (ADD3 OE) or with GFP and empty vector (control) and their morphological dynamics were analyzed by live imaging in interphase (A-C) and in rela-on to mitosis (D, E). (A) Examples of the morphological dynamics in interphase of the 4 morphoclasses upon ADD3 OE vs Control.Note the increased elonga-on of the ADD3 OE cells.The -me lapse indicated in the upper le^ corner of the images.Scale bars, 50 µm.(B) Quan-fica-on of morphological changes in interphase.Note the increase in acquisi-on of new morphology upon ADD3 OE.Mean of 4 independent transfec-ons.Error bars, SD; *, P<0.05; two-way ANOVA with Bonferroni post hoc tests.(C) Quan-fica-on of morphological transi-ons in interphase for each morphoclass.Mean of 3 independent transfec-ons.Error bars, SEM.(D) Examples of the morphological dynamics in rela-on to mitosis of the 4 morphoclasses showing the -me point pre-(le^), during (middle) and post-mitosis (right) upon ADD3 OE vs Control.The -me lapse indicated in the upper le^ corner of the images.Scale bars, 50 µm.(E) Schema-c representa-on of the morphological inheritance shown as the percentage of morphoclass progeny for each mother morphotype.Number of mother cells (control, ADD3 OE): nonpolar (12, 13), flat polar (16, 16), circular mul-polar (30, 5), elongated (15, 29).Data come from 4 independent transfec-ons.